The Online Consciousness Conference
133 Comments and Responses on:
by
University of Southampton
&
Universite du Quebec a Montreal
[ paper in pdf format
movie in .m4v format ]
Note: Commentaries are in 12-point type-size and author’s responses are in 14-point type-size. The relevant passages from the commentaries are quoted in the responses. The full commentaries are also available at the Consciousness Online site. Commentaries not addressed to me are in grey.
Professor Harnad, I am highly sympathetic to your emphasis on robots, and to your recognition of the explanatory gap at the end of your paper. But I believe there are a couple of points along the way that are not quite right, and that interfere with a clear view of what Turing and Searle did and didn’t do. The first concerns the questions (on p. 2) of what is missing, that makes Searle not understand Chinese, and how Searle knows he doesn’t understand Chinese. Your answer is that what is missing is “what it feels like to be able to speak and understand Chinese”. And Searle knows (indeed, is the only one who can know) whether he has that feeling. But there is another answer that is implicit in Searle’s paper (though it is made along with some other points, and so does not stand out clearly). This is that Searle cannot respond to Chinese symbols in any way other than verbally. For example, the people outside the room may pass in a paper that says (in Chinese) “If you’d like a hamburger at any time, just press the green button to the right of the ‘out’ slot.” Assuming the right program, scripts, and so forth, Searle may well pass out a slip that says (in perfect Chinese) “Thank you very much, if I want one I’ll do that”. But meanwhile, even though he may be getting desperately hungry, he will not take himself to have any reason to press the green button. You do call attention to this fact later on in the paper. The point of my mentioning it here is to point out that this fact is by itself sufficient to show that Searle did not understand what was written on the input slip. His feelings about his understanding are neither here nor there. People can deny their blindness; it seems imaginable that a person who understood perfectly well might be under the illusion of lack of understanding (and conversely). On page 3, you give three reasons why Turing is to be “forgiven” for offering the imitation game. Clarity on the point I’ve just made should help us see that Turing needs no forgiveness. Turing’s target was named in the title of his paper – it was intelligence (not understanding, having indistinguishable cognitive capacities, having feelings, or having a mind). Intelligence is indeed the kind of thing that can be exhibited in conversation. The rationale for the imitation game could be put this way: If machine M can do as well in conversation as human H (shown by inability of judges to reliably tell which is which) then they have the property that’s exhibited in conversation (namely, intelligence). You are quite right that if we include, e.g., recognition, and ability to manipulate under “cognitive capacities”, we’ll need a robot, not just a computer. And if we want understanding, as Searle uses the term, we’re going to have to have grounding, i.e., something that counts as nonlinguistic behavior, i.e., at least some sensors and motors; in short, a robot. You suggest what seems to me to be the right thought experiment at the top of p. 8. In brief, we’re to imagine a feelingless robot that’s a good conversationalist and suits actions to words for its whole, lengthy existence. I’ll enrich the case by imagining the robot to have done my shopping for me for years. It knows all about bus schedules, product reliability, etc. It says “I’ve hooked your new speakers up to your receiver”. It has done so, and is reliable about such matters. Is it remotely plausible that it doesn’t know what “speaker” or “hook up” means? I am not sure what you meant us to think, but I think the answer is that of course it understands its words. What a robot has not got (by hypothesis), and does not need to have, is any feelings. The research program for intelligence and for the grounding that provides meaning for words is distinct from any research program for producing feelings. It’s the absence of feelings that accounts for its seeming justifiable to dismantle a robot. If it’s my shopping assistant that you’re recycling, I’ll be distressed, as I would if you took apart my car. But while a robot might defend itself against being dismantled (I’d design that into any robot I’d make) it can’t be genuinely afraid of dismantlement if it’s just a robot and has no feelings. It can’t be made miserable by threat of dismantlement. What feelings matter to is not using words meaningfully, but being a proper subject of genuine moral concern.
TURING’S CRITERION IS NOT FACE-VALID BY DEFINITION
(Reply to Robinson)
ROBINSON: “Searle cannot respond to Chinese symbols in any way other than verbally.”
Correct. And that’s why the right Turing Test is T3, the robotic TT. But that isn’t the reason Searle can be sure he does not understand Chinese. Nor is it true that passing T3 entails understanding, by definition. That’s no more true than that passing T2 does.
ROBINSON: “[Searle's] feelings about his understanding are neither here nor there.”
It’s not Searle’s feelings *about* his understanding that are at issue, it’s his feeling *of* understanding. And you’re right that it’s neither here nor there, because Searle is not understanding (Chinese) though he is passing T2 (in Chinese), and nor is anyone or anything else.
ROBINSON: “Turing’s target was… intelligence (not understanding, having indistinguishable cognitive capacities, having feelings, or having a mind).”
Correct, but having intelligence is synonymous with having indistinguishable cognitive capacities. (What else does it mean?) Whereas having feelings (which is synonymous with having a mind) is the bone of contention.
Turing was right that explaining the causal mechanism that generates our cognitive capacities answers all the answerable questions we can ask about cognition (intelligence, understanding, whatever you like). But that still leaves one question unanswered: Does the TT-passer feel (and if so, how, and why)? Otherwise it does not have a mind.
(And mindless “cognitive capacity” or “intelligence” can only be taken to be the same thing as our own if we accept that as a definition, which rather begs the question, since the question is surely an empirical one.)
ROBINSON: “[W]e’re to imagine a feelingless robot that’s a good conversationalist and suits actions to words for its whole, lengthy existence.”
Like you (I think), I too believe it (1) unlikely that anything could pass T3 unless it really understood, really had intelligence, really had our cognitive capacity. But I also believe — *for the exact same reason* — that it is (2) unlikely that anything could pass T3 unless it really felt, because, for me, understanding (etc.) *means* cognitive capacity + feeling. (In other words, again, I do not *define* understanding, etc. as passing T3.)
But, unfortunately, although I share your belief in the truth of (1), I also believe that we will never be able to explain why and how (2) is true (nor ever able to know for sure *whether* (2) is true — except if we happen to be the T3 robot).
ROBINSON: “What feelings matter to is not using words meaningfully, but being a proper subject of genuine moral concern.”
I am afraid I must
disagree. I do happen to be a vegan, because I don’t want to eat
feeling creatures; but the “hard” problem of explaining how and why we
feel is not just a moral matter.
Stuart W. Mirsky
I may be wrong but I took Professor Harnad’s use of “feeling” to be somewhat broader than your criticism seems to imply. Indeed, I read it as an allusion to awareness, as in having the sense of what is happening when an instance of information is understood. The use you allude to is the emotional aspect of the word “feeling” which is certainly a common way we understand the word but it is different, and in an important sense, more narrow than feeling as awareness. (We are aware of emotional content as well as intellectual content when we have either.) It’s feeling as awareness that I believe Professor Harnad seems to have in mind when discussing Searle’s insistence that, for a computational (or any) machine to be really intelligent, it would need to do more than just respond in the right way(s) verbally. It would also need to get the meaning of what it is “reading” and “saying.” Searle certainly makes an important point about what a machine entity would have to be in order for it to be called intelligent in the way we use the term for ourselves much of the time (i.e., conscious), whatever the merits of his Chinese Room argument (which, I think, is seriously flawed). But Professor Harnad, it seems to me at least, is right to note that there is something we could call “feeling” involved here, though perhaps it’s an unfortunate choice of term since it is so often tied up with references to our emotional lives. (Both emotions AND instances of intellection are part of our mental lives after all and, as such, are both felt by us, even if some instances of intellection have a perfectly flat modality — indeed, flatness, too, is felt on this usage, even if only as the absence of strong feelings of the emotional sort.) I do think your point about the machine’s capacity to act in certain ways in response to the information is useful because that’s integral to comprehending when the generation of action is part of the information in the input. But it doesn’t mean that a robotic body is essential. After all, many a paraplegic could likely understand the same proposal to order a hamburger but still not be able to act on it — just as the uncomprehending man in the Chinese Room is unable to, albeit for different reasons. (Searle’s man really doesn’t get it while the paraplegic just lacks the capacity to act.) More important, I would suggest, is to distinguish just what it is that constitutes understanding in us, what Searle calls getting the semantics of the “squiggles” and “squoggles”. And on that score, I think what’s finally needed is a satisfactory account of semantics that adequately captures what happens in us when we have understanding. We have a mental life within which understanding occurs and that is just what the Chinese Room, as sketched out by Searle, lacks by hypothesis. (It’s just a rote device for matching symbols.) The problem, finally, is what would it take to undo that lack — and could it be done using computational processes of the kind that run on today’s computers? I think it could but the only way to really see it, I think, is to unpack the notion of semantics.
COGNITION, COMPUTATION AND COMA
(Reply to Mirsky)
MIRSKY: “[T]he machine’s capacity to act in certain ways… [is] integral to comprehending when the generation of action is part of the information in the input. But it doesn’t mean that a robotic body is essential. After all, many a paraplegic could likely understand…”
We need a mechanism that can pass the full robotic T3 before we test what happens if we cut back on its sensorimotor capabilities. (The TT for a terminally comatose patient would be easy to pass, but no TT at all.) (Whether one could even contemplate blinding a T3 robot, however, is indeed a moral question of the kind raised by W.S. Robinson above.)
MIRSKY: “[W]hat’s finally needed is a satisfactory account of semantics that adequately captures what happens in us when we have understanding… — could it be done using computational processes of the kind that run on today’s computers? I think it could but the only way to really see it, I think, is to unpack the notion of semantics.”
Until further notice, what both Searle’s Chinese Room argument and the Symbol Grounding Problem show is that cognition cannot be just computation (symbol manipulation) — any computation, whether on today’s computers or tomorrows. What’s missing is sensorimotor capacity, which is not computational but dynamic. That will provide grounding. But for meaning you need to capture feeling too, and that’s a rather harder problem.
Joshua Stern
“He said, essentially, that cognition is as cognition does (or, more accurately, as cognition is capable of doing): Explain the causal basis of cognitive capacity and you’ve explained cognition.” Of course Turing never said any such thing, what he said was that if you cannot tell then you cannot tell. That was good positivist talk when he said it in 1950, and who could ever argue with it? Which would all be nitpicking, except we should let it remind us not to make the same mistake (so often as here mistakenly) alleged to Turing. “… and you really cannot ask for anything more”. Exactly, that is what Turing did say. The implication of (and I concede, by) Turing was that, if the thing on the other end of the test was not really intelligent, it might be the next best thing. But let us follow the invalid claim after all – let us reject the imitation, too, and demand the real thing. Now, even if Turing did not really claim the above, I – with less wisdom but sixty years more historic experience with the field – will happily do so in Harnad’s words: explain the causal basis of cognitive capacity and you’ve explained cognition. Well, who would argue that, either? Although one can argue it is a tautology – if it turns out you have NOT completely explained cognition, then the odds that you somehow managed to explain the causal basis of cognition anyway approaches zero. – Searle’s Chinese Room? My take is that Searle argues exactly correctly, but with exactly the opposite direction that he claims. If the CR can converse in Chinese without it anywhere “understanding” Chinese, then I take him at his (ex hypothesi) word – that is exactly how a computer will do it, thank you very much. I see nothing missing. The claim that something is missing is only asserted, never argued. I find nothing missing at all. If someone passes in a note that says, “The Chinese Room is on fire, run!”, and it returns out a note, “Thank you for telling me, I will run away, oh wait I can’t, please call the fire department, arrrgghh!”, and yet the operator sits still inside without even knowing his peril, well, we have learned several things, none of which Searle ever discusses, nor does most of the voluminous commentary upon it. If something is missing in the CR hypothesis, it is the rest of what is likely to be needed to fulfil the hypothesis, a narrative of self by the CR such that it would not be just a manipulation of symbols, but a manipulation of symbols in a context. I see no reason that cannot be just more symbol manipulation, but it does have to be a particular kind of symbol manipulation, and frankly, to make the argument interesting, we have to do a little more than hand-waving at how it would work – which is, sadly, the real conclusion one should reach about the CR in the first place. – Claims that some further “grounding” in physical terms is needed, I believe need to be unpacked. When I sit and talk to Fred about plans to explore Mars, neither of us has ever been there or is likely to be. So, we are grounded in what we know, and reason by analogy, right? Well, it turns out Fred only knows what he’s read in some books, he’s never roved over dry sand in a crater himself – but however far you’d like to stretch the analogy analogy, it’s beside the point. The point, I suggest, is that when Fred reasons, whether it is by analogy or by physical embedding, the reasoning in his head is (presumably) some sort of symbolic manipulation “and nothing but”. So, whatever the role of the real world in how Fred (and I) talk about Mars, the reasoning and the talking (perhaps the “talking” is by email) is all and only the original question which is, how are they doing that stuff, that linguistic stuff, that cognitive stuff, that logical stuff? – As a final note, I would just toss in that symbol manipulation never is just symbol manipulation, it is always a physical particular, and in that it is in no way removed from anyone’s claims that a true cognitive system must be physically grounded. Correspondingly, to wonder about the role of causal systems in cognition is an empty argument, as has often been pointed out, a lack of causality would defeat pretty much any kind of argument one would ever make about cognition – one can hardly imagine a workable definition of the term without a causal world for it to live in (and for us to discuss it in). So whatever the shortcomings of any particular causal system, it can hardly be said to be the causality that is the problem with it. So there never really was a “nothing but” in the symbol manipulation that Fred did in thinking about Mars, even his thinking involves various neurons and chemical reactions, or if Fred happens to be a computer, some circuits and bags of electrons shifting around. – Whether any of this bears on the problem of qualia is another matter. – The moral of the CR story (and it’s a rollicking good story no matter what its faults) may indeed be something about grounding, even without worrying about feelings and qualia, and that is, a lack of grounding, or embedding, in the full causal matrix of the world, including the social aspects thereof, is in some way crucial to telling a believable story about cognition, much less in realizing it. Does the room “know” that it has an operator sitting trapped in its midst? If not, can it really be said to be conversing with outsiders as if it were a full human agency? Well perhaps, as humans are never known for omniscience, nor about even modest levels of awareness of their internal states – and hey, wasn’t that a part of the original story?
IMPLEMENTATION ≠ GROUNDING
(Reply to Joshua Stern)
JSTERN: “Turing never said any such thing…”
My little essay was not about restating Turing words, it was about what those words meant (or, at some points, what they ought to have meant!).
JSTERN: “Searle argues exactly correctly, but… I find nothing missing at all”
Searle argues exactly correctly that if he were doing the Chinese T2-passing computations he would not be understanding Chinese. That’s what’s missing.
I then focussed on *how* Searle would know that he was not understanding Chinese. And the answer is that it *feels like something* to understand Chinese. And Searle would know that he did not have that feeling, if he were just implementing Chinese T2-passing computations.
JSTERN: “Claims that some further “grounding” in physical terms is needed… need to be unpacked… we are grounded in what we know, and reason by analogy, right?…”
Right. And “grounded in what we know” means grounded “in physical terms.” Otherwise it’s just the Chinese-Chinese dictionary definition chase all over again — which is just an infinite regress, from meaningless symbol to meaningless symbol. (And neither cognition, nor knowing, nor T2-passing are just “reasoning.”)
JSTERN: “The point, I suggest, is that when [one] reasons, whether it is by analogy or by physical embedding, the reasoning in his head is (presumably) some sort of symbolic manipulation ‘and nothing but’…”
And the point of Searle’s Chinese room argument, and of the symbol grounding problem, is that without sensorimotor grounding (“physical embedding”), computation is nothing but the ruleful manipulation of meaningless symbols. Therefore cognition (not just “reasoning”: *cognition*) is not just computation.
But that point has already been made many times before. The point of this particular little essay was to focus on what even a grounded cognitive system that could pass the Turing Test (whether T3 or T4) might still lack (namely, feeling), and to point out that whether or not the successful T3- or T4-passer really did feel, the full causal explanation (“reverse-engineering”) of the underlying mechanism that generates its T3/T4 success will not explain why or how (let alone *whether*) it feels.
JSTERN: “[S]ymbol manipulation never is just symbol manipulation, it is always a physical particular, and in that it is in no way removed from anyone’s claims that a true cognitive system must be physically grounded.”
Of course computation has to be physically implemented in order to be performed. That’s just as true of the physical implementation of a lunar landing simulation or the proof of a computational proof of the 4-color problem. But the physical implementation of any computation that is actually performed is not the same as the sensorimotor grounding of a computation that attempts to implement our cognitive capacity.
And that, again, is because cognition is not just computation.
JSTERN: “[T]o wonder about the role of causal systems in cognition is an empty argument… lack of causality would defeat pretty much any kind of argument one would ever make about cognition – one can hardly imagine a workable definition of the term without a causal world for it to live in (and for us to discuss it in).”
The only causality we need for cognition is the causality needed to pass T3. The causality to pass T2 is not enough (if it is just computational), because the connection between T2′s internal symbols and the external world of objects to which those symbols refer cannot be made by the T2 system itself: It has to be mediated by external interpreters: cognizers. Yet it is the cognizers’s cognition that the T2 mechanism is intended to explain.
Another infinite regress.
Hence, again, cognition is not just computation.
And it’s not just any old causality (or implementation) that’s needed: It’s the right one. T3 fixes that (insofar as grounding, hence doing, is concerned). But neither T3 nor T4 touch feeling, yielding not even a clue of a clue of how or why cognizers can feel.
JSTERN: “[T]here never really was a “nothing but” in the symbol manipulation… thinking involves various neurons and chemical reactions, or if… a computer, some circuits and bags of electrons shifting around.”
Nope. I’m afraid physical implementation ≠ grounding, whether the implementation is in wetware or in hardware. Grounding is the sensorimotor capacity to recognize and manipulate the external referents of internal symbols, whether the implementation is in wetware or in hardware.
Ceterum Censeo: And grounding still does not explain feeling (hence meaning).
JSTERN: “Whether any of this bears on the problem of qualia is another matter.”
None of your comments (which are only about the Chinese room and symbol grounding) bear on the problem of feeling (“qualia”); but the target essay does.
JSTERN: “The moral of the CR story… may indeed be [that] a lack of grounding, or embedding, in the full causal matrix of the world, including the social aspects thereof, is in some way crucial to telling a believable story about cognition, much less in realizing it.”
T2 does not require grounding, T3 does. The only moral is that if you want to tell “a believable story about cognition,” reverse-engineer whatever it takes to pass T3.
But the story the target essay is telling is not just about Searle and CR; it is about Turing and feeling.
Joshua Stern
“None of your comments (which are only about the Chinese room and symbol grounding) bear on the problem of feeling (“qualia”); but the target essay does.” Yes sir, that is exactly right. But what this shows is that you have given up on the symbol grounding problem, and are now completely engaged in a qualia grounding problem. My point is that this is another matter. I see no shortcoming to symbols that need grounding. I do not buy Searle’s assertion that the “feel” of understanding Chinese is missing, is illustrated by the CR. You apparently do buy his assertion, his non-argument. I see no reason to ground symbols in qualia. In fact, I argue – and you ignore – the idea that even if you could ground a symbol in a feeling, the symbol manipulation would still be exactly that which realizes cognition, it would simply do so with a better explained foundation. Grounding a symbol is not eliminating the symbol, and a grounded symbol is still nothing at all without a computational/ cognitive process. – Have I missed the detailed description of what your T3 is supposed to be? I gather it is a fully functional robot, not just a teletype T2 test, but I can’t quite see that in this published text. The problem is, your putative T3 in no way answers the questions you have about T2. Refer it you will to David Chalmer’s zombies – they might speak Chinese all day long, have a recollection of history, juggle bowling pins and complain about the heat, and still have no more understanding than Eliza. And you can extend that to T4 or Taleph2 and still not touch the issue. If you don’t believe me, ask Donald Davidson or Ruth Garrett Millikan about SwampMan. What that all misses is that Turing was right to focus on what differentiates the entire set of examples from non-cognitive systems, and that is whether an algorithmic process can explain cognition, for it is absolutely certain that nothing else can, save transcendental arguments. –
EXPLAINING FEELING
(Reply to Joshua Stern)
JSTERN: “[Y]ou have given up on the symbol grounding problem, and are now completely engaged in a qualia grounding problem.”
Actually, I have not given up on the symbol grounding problem. It just wasn’t the primary focus of my essay in this online symposium on consciousness.
And the essay was not about qualia “grounding” (I don’t even know what that would mean!) It was about the fact that neither the explanation of T3 nor the explanation of T4 capacity explains the fact that we feel.
(And the only time I mentioned the word “qualia” was in the video, as one of the countless synonyms of consciousness that we should stop proliferating, and just call a spade a spade: Feeling.)
JSTERN: “[E]ven if you could ground a symbol in a feeling, the symbol manipulation would still be exactly that which realizes cognition, it would simply do so with a better explained foundation.”
If you grounded symbols in sensorimotor capacity (T3) it is not at all clear that the resultant hybrid system could still be described as computational at all. (Symbols that are constrained by their dynamic links to their referents are no longer the arbitrary squiggles and squiggles that formal computation requires.)
And, to repeat, no one is talking about “grounding” symbols in feeling: The problem is explaining, causally, why and how some cognitive states (even hybrid symbolic/dynamic ones) are *felt* states.
JSTERN: “Have I missed the detailed description of what your T3 is supposed to be? I gather it is a fully functional robot, not just a teletype T2 test, but I can’t quite see that in this published text.”
As far as I can tell, you haven’t missed anything:
A system that can pass T2 has the capacity to do anything we can do, verbally, and do it in a way that is indistinguishable *from* the way any one of us does it, *to* any one of us, for a lifetime.
A system that can pass T3 has the capacity to do anything we can do, both verbally and robotically, and do it in a way that is indistinguishable *from* the way any one of us does it, *to* any one of us, for a lifetime.
A system that can pass T4 has the capacity to do anything we can do, both verbally and robotically, do it in a way that is indistinguishable *from* the way any one of us does it, *to* any one of us, for a lifetime, both behaviourally, and in all its measurable neurobiological function.
JSTERN: “David Chalmers’s zombies… might speak Chinese all day long, have a recollection of history, juggle bowling pins and complain about the heat, and still have no more understanding than Eliza.”
Never met a zombie, so far as I know. Nor do I know on what basis they are being supposed to exist. They sound like T3 or T4 robots without feelings, and as such they do not add or settle anything one way or the other.
The unsolvable (“hard”) problem remains that of explaining how and why T3 or T4 robots feel, if they do, and if they do not (i.e., if they are “zombies”), than explaining how and why *we* are *not* Zombies.
JSTERN: “And you can extend that to T4 or T-aleph2 and still not touch the issue.”
You can say that again…
JSTERN: “Turing was right to focus on what differentiates the entire set of examples from non-cognitive systems, and that is whether an algorithmic process can explain cognition, for it is absolutely certain that nothing else can, save transcendental arguments.”
No, what distinguishes a TT-passer from other systems is that it passes the TT totally indistinguishably from any of us. The only TT that can be passed by computation alone is T2, and that is not cognition because it is ungrounded.
Only a dynamical system can pass T3 or T4 (probably a hybrid dynamic/computational one). And a purely computational system can *simulate* just about any dynamical system, but it can’t *be* any dynamical system.
If T3 and T4 robots feel, then they feel, and then what we are looking for is not a “transcendental argument” but a down-to-earth explanation of how and why they feel,; but there does not seem to be the causal room for such an explanation.
If T3 and T4 robots don’t feel, then they don’t feel, and then what we are looking for is not a “transcendental argument” but a downto-earth explanation of how and why they don’t yet we do; but there does not seem to be the causal room for such an explanation.
Joshua Stern
First, my apologies for being mislead by the “qualia” objection, that I had failed to mention it in my first reply. Only, I’m not sure why you raised it. Let me try to amend my first reply with this insertion, then reply to one further point in this your second reply. Also, if my language here sounds curt that is not my intention, I’m only after brevity in discussion, and no doubt it suffers from my brevity in composition as well – or I would have hoped not to have made the excursion on qualia! So, to repair, please add this after where I _incorrectly_ said: “But what this shows is that you have given up on the symbol grounding problem, and are now completely engaged in a qualia grounding problem.” [Actually, your essay does mention qualia, and then at the end, waves them away. Yet, you accept Searle’s complaint that “something is missing” and at least Searle asserts that it is a quale, if “feeling” is even specific enough to be worth discussing as a quale (and of course the tradition is that it is), and that "something is missing" is what your essay, and your work on the grounding problem, attempts to solve. OK, first, of course there *is* a grounding problem for symbols, or else what is it that makes the lexical string c-a-t refer to a real-world feline. But, this is hardly unique to AI systems, the same question is asked why you refer to a cat that way, or even why I do. And certainly the “answer” is in the way that the agent, be it human or computer, relates to actual cats. So then, what is it that Searle is going on about? Not that at all. So, I take my lead from Searle’s confusions, rather than the much more universal linguistic problem. And, I should mention that Searle’s confusions are, at worst, a mereological problem. I’m sure this has been raised regarding the CR, but – within Searle himself, surely he does not expect his pineal gland, alone, to understand Chinese, nor any one of his neurons. The classic “systems approach” is that systems understand, not each component of the system, that false assertion is an ancient philosophical error, a fallacy of composition. So, I assert again, Searle’s description of the CR from the start is absolutely correct, he as a component of the CR of course does not understand Chinese, whether or not the proper understanding of Chinese should include feelings, whether or not the grounded understanding of Chinese requires physical actions. The actual direction your essay takes is to demand further behaviorist/positivist confirmation of understanding. You take Turing’s original test and chase it down a regress, then try to wave off Humean skepticism that any observational system could ever bring certainty. I agree with that to a large extent, but I’m afraid that in chasing the regress, you forget what it is we were discussing, what is critical to the discussion, and that is the symbol manipulation in the first place. If you succeed in chasing the regress and waving off the remnant (which I do not believe is quite the proper structure, but let’s allow it), then you have only established that the original system was right after all, you certainly have not *reduced* symbol manipulation into swinging a baseball bat. And so, I argue that starting the regress is an improper, unproductive move, and offer further reasons why this regress, to qualia or embedding, is both unnecessary and unpersuasive.] Finally, the additional point: “Only a dynamical system can pass T3 or T4 (probably a hybrid dynamic/computational one). And a purely computational system can *simulate* just about any dynamical system, but it can’t *be* any dynamical system.” But you have agreed that a computational system must have a physical realization, and that physical realization is certainly of the same dynamic systems quality as T3 or T4, so whatever difference you are asserting in T3 or T4, it is not the property of being a dynamic system. I’d suggest it is only in looking more humanoid, and I think we all know that is not a sufficient argument.
GROUNDING ≠ FEELING
(Reply to Joshua Stern)
JSTERN: “Actually, your essay does mention qualia, and then at the end, waves them away.”
I can’t grep a single token of “qualia” in my essay, but if you mean my preferred synonym, “feelings,” I would not say I wave them away! On the contrary, I say they are real, relevant — and inexplicable (with reasons)…
JSTERN: “‘something is missing’ is what your essay, and your work on the grounding problem, attempts to solve”
Actually, the essay points out that there are two things missing (grounding and feeling), and only one of them can be provided. What is missing in the verbal Turing Test, T2 (if passed via computation alone), is sensorimotor grounding. The solution is to move up to T3 (the robotic TT).
What is missing in any full, Turing-scale, reverse-engineering solution for generating and explaining cognitive capacity — whether T3 or T4 — is an explanation of why and how it feels (if it does); and the reason for this is that there is no causal room for feeling as an independent causal component in the explanation (except if we resort to psychokinesis, which is contradicted by all evidence).
JSTERN: “systems understand, not each component of the system… Searle… as a component of the CR of course does not understand Chinese”
Searle’s Chinese Room Argument has been generating fun as well as insights for over 30 years now, but it is rather late in the day to resurrect the hoary old “System Reply” without at least a new twist!
On the face of it, the “System Reply” (which is that it is not Searle that would understand if he executed the Chinese T2-passing computations, but “The System”) is laid to rest by Searle’s own original reply to this, which ran something like this (the words are not Searle’s but spicily improvised here by me):
“Ok, if you really believe that whereas I do not understand Chinese, the ‘system’ — consisting of me plus the symbols, plus the algorithms I consult to manipulate them — does understand Chinese, then please suppose that I memorize all the algorithms and do all the symbol manipulations in my head. Then “le système c’est moi”: There’s nothing and no one else to point to. So unless you are prepared to believe (as I certainly am not!) that memorizing and executing a bunch of symbol-manipulation rules is sufficient to generate multiple personality disorder — so that there are now *two* of me in my brain, unaware of one another, one of whom understands Chinese and the other does not — please do believe me that I would not be understanding Chinese under those conditions either. And there’s no one else there but me… Try the exercise out on something simpler, such as training a 6 year-old innocent of math to factor quadratic equations by rote symbol-manipulation formula, in his head. Given any quadratic equation as input, he can give the roots as output. See if that generates an alter ego who understands what he’s doing…”
(But, can I please repeat my hope that my little essay will not just become an occasion to rehearse the arguments for and against Searle’s Chinese Room Argument? With T3, we’ve already left the Chinese Room, and the problem at hand is explaining how and why we feel, not how and why the Chinese Room Argument is right or wrong…)
JSTERN: “The actual direction your essay takes is to demand further behaviorist/positivist confirmation of understanding.”
Reverse-engineering the mechanism underlying a device’s performance capacities is hardly behaviorism/positivism! Behaviorists did not provide internal causal mechanisms at all, and positivists were concerned with basic science, not engineering, whether forward or reverse.
JSTERN: “You take Turing’s original test and chase it down a regress, then try to wave off Humean skepticism that any observational system could ever bring certainty.”
The regress is the one involved in trying to chase down meaning by going from meaningless squiggle to meaningless squoggle (as in looking up a definition in a Chinese-Chinese dictionary when you don’t know a word of Chinese).
The Humean/Cartesian uncertainty is the usual one, the one that attends any empirical observation or generalization, but I do think the uncertainty is an order of magnitude worse in the case of the uncertainty about other (feeling) minds than it is about, say, the outside world, or physical laws.
But, with Turing, I agree that Turing-indistinguishability — which can be scaled all the way up to empirical-indistinguishability — is the best one can hope for, so we’re stuck with T3 or T4. The rest is the usual underdetermination of theory by data (there may be more than one successful causal theory that can explain all the data) plus the unique case of uncertainty about whether the system that conforms to the theory really has a (feeling) mind.
JSTERN: “in chasing the regress, you… have only established that the original system was right after all, you certainly have not *reduced* symbol manipulation into swinging a baseball bat.”
Passing T3 is not reducing symbol manipulation (computation) to bat swinging (dynamics). It is augmenting the task of reverse-engineering, scaling it up to having to explain not just verbal capacity, but the sensorimotor capacity in which verbal capacity (and much else in human cognition) is grounded. That means augmenting pure symbol manipulation to a hybrid symbolic/dynamic system (yet to be designed!) that can successfully pass T3.
But both T2 and pure symbol manipulation are by now dead and buried, along with the “system reply.”
JSTERN: “this regress, to qualia or embedding, is both unnecessary and unpersuasive.”
The T2 symbol-tosymbol regress, which is grounded (halted) by dynamic T3 capacity makes no appeal to feeling (“qualia”). (The regression is just as present if you simply point out that Searle could not point to a zebra even as he was explaining what a Ban-Ma was, in Chinese.)
And the problem of explaining feeling makes no particular use of the regress (groundedness); it is assumed that that has already been taken care of in T3 — but it doesn’t help!
JSTERN: “you have agreed that a computational system must have a physical realization, and that physical realization is certainly of the same dynamic systems quality as T3 or T4, so whatever difference you are asserting in T3 or T4, it is not the property of being a dynamic system. I’d suggest it is only in looking more humanoid, and I think we all know that is not a sufficient argument.”
Hardly. What distinguishes a T3 robot from a T2 computer is not a difference in what their respective computer programs can do, but a difference in what their dynamics can do. The ability to recognize a zebra is not just a difference in appearance!
Stuart W. Mirsky
(Apologies in advance as I’m not on my own computer and must necessarily type fast with less time to double check what I’ve written. Hopefully I will have enough.) Doesn’t the issue finally boil down to what it means to understand something at all? As Professor Harnad notes, Searle seems to be looking for something we can call the feel of understanding, what it feels like to understand Chinese in his mythical room. The author raises two key issues: First that understanding requires something more than the isolationism provided by the basic Chinese Room scenario. He suggests it requires a dynamical relation with the world (robotic capacity to sense and operate in the world). What he apparently means is not that the CR have the capacity to act on some things it understands but that it has the ability to connect symbols to inputs in a causal chain. Thus, we’re told that having meaning is to ground symbolic representations in elements of the physical environment. Meaning is a referring relation. Professor Harnad further differentiates between meaning as connection between symbol and referent on the one hand, and having feelings of getting it or understanding (in the sense already described above) on the other. A CR equipped with the necessary physical links might, he suggests, pass the requisite Turing test for understanding but we would still be at a loss to ascertain if the prescribed feeling is present or absent. A certain mystery is preserved in this account. But perhaps the mystery could be dispelled if we go back to the question of meaning as a matter of grounding. Perhaps grounding is not an adequate account of meaning after all, even if grounding plays a role in our mental lives (in which meaning is realized). It seems undeniable that we are grounded in an important sense and that we see meaning in things, at least to some degree, through the referential relation between word (or symbol) and elements in the world as is suggested. But does that imply that meaning is just an expression of this kind of causal linkage? Is it enough to suppose that without a causal chain to outside stimuli we are stuck with circularity, i.e., one symbol finding its meaning in another equally isolated symbol — and that this undermines any real possibility of meaning? Would a CR that depended for all its inputs on what it is fed about other symbols be unable to achieve understanding in any but an imitative sense? Suppose we were totally isolated from our sensory inputs about the world and had no inputs but a feed of digitized signals which had no analogical relation to the underlying causal phenomena which generated them because they were, say, generated in an abstract way (Morse code perhaps)? Assuming that they were not merely random but organized in patterns which were recognizable and so carried some information through their repetitions, would we be denied any possibility of achieving meaning at all? It would certainly be a much more limited world than we are used to — and probably a very different one, no matter how encyclopedic the information fed to us in symbols would be. But would it follow that we would not be able to have any understanding at all? (Didn’t Helen Keller learn about her world somewhat like this?) After all, what are we but organic machines which receive inputs about the world through physical systems? What we know of the world begins with the signals our sensory equipment pick up and pass up the line through our nervous to our neurological systems. What happens to those signals involves various physical transformations along the way, much as computers transfer their signals into more and more complex arrays of information. But how is that significantly different (operationally rather than in terms of the physical platform) from what computers do (though we are arguably much more complex)? Perhaps a better account of semantics is to be found in a more complex notion of the kind of system we are and which Searle’s CR manifestly isn’t?
After all, the system he describes is specked to do no more than rote responding, symbol matching, though no one thinks that symbol matching is all we’re doing when we read and match symbols through understanding their semantic content. (Indeed, that’s the basic intuition that seems to make the CR argument so compelling. If that’s all that a computer finally is, how can it understand in the way we do?) Yet, the fact that computational operations are largely mechanical in this way (and we need not deny it) doesn’t imply anything about more complicated features that may arise from increasingly complex operations formed by what are otherwise much simpler operations. Why should a series of ordered electrical signals yield a readable screen on my pc as I’m typing this? The letters I see aren’t the series of signals that convert the lighted areas on my screen into what are, for me, so many words. And yet they do it and produce something in which I can find meaning even if the digital signals themselves carry no such meaning. Why then should we abandon hope that feeling (qua awareness) could be causally explained. After all, if the CR is too thinly specked to do anything but rote symbol matching, it’s only natural to presume it’s too thinly specked to do whatever it is that amounts to the feeling of knowing or understanding something. If semantics involves the complexity of a system, why wouldn’t that same complexity be enough to account for the sense of understanding that accompanies instances of understanding in us in cases like this?
COMPUTATIONS AND DYNAMICS; COGNITION AND COMPLEXITY
(Reply to Stuart W. Mirsky)
SMIRSKY: “we’re told that having meaning is to ground symbolic representations in elements of the physical environment. Meaning is a referring relation.”
No. Sensorimotor grounding is a necessary, not a sufficient condition for meaning. Meaning = T3 grounding + feeling.
SMIRSKY: “A CR equipped with the necessary physical links might… pass the requisite Turing test for understanding but we would still be at a loss to ascertain if the prescribed feeling is present or absent”
Passing T3 is not “a CR plus physical links” (if by “CR” you mean a computer). Passing T3 requires a robot — a dynamical system with the sensorimotor power to do anything a person can do, verbally, and behaviorally, in the world. That’s highly unlikely to be just a computer plus peripherals.
(The brain and body certainly are not a computer plus peripherals; there’s not just [implemented] computations going on in the brain, but a lot of analog dynamics too.)
But even if it were possble to pass T3 with just computation + peripherals, cognitive states would not be just the computational states: They would have to include the dynamical states that include the sensorimotor transduction plus the computation. (Otherwise Searle could successfully re-run his CR argument against the computational component alone.)
SMIRSKY: “Perhaps grounding is not an adequate account of meaning after all, even if grounding plays a role in our mental lives (in which meaning is realized).”
Grounding is a necessary but not a sufficient condition for meaning.
SMIRSKY: “But does that imply that meaning is just an expression of this kind of causal linkage? Is it enough to suppose that without a causal chain to outside stimuli we are stuck with circularity, i.e., one symbol finding its meaning in another equally isolated symbol — and that this undermines any real possibility of meaning?”
Please see the earlier posting on “Cognition, Computation and Coma”: The T3 causal connection to the world is needed to test whether the T3 robot really has T3-scale capacity, indistinguishable from our own, for a lifetime. But apart from the need to *test* it
— in order to make sure that the robot has full T3 power — all that’s needed in order to *have* full T3 power is to have it. There’s no need for a prior real-time causal history of sensorimotor contact with objects. Just as toasters would be toasters, with all their causal powers, even if they had grown on trees rather than being designed and built by engineers, so my current T3 capacity would be what it is at this moment even if I had dropped freshly off a tree, fully formed, 10 minutes ago, with no real history (including having actually written the target essay several weeks ago, nor having seen and interacted with real objects throughout a lifetime).
This is not to say that it is likely to be possible to design a T3 robot with a viable yet prefabricated “virtual” history, rather than a real history — any more than it is likely that a T3 robot designed to pass the T3 for a comatose person would be able to pass the T3 for a normal, awake, ambulatory person.
But the most important point is that it is a big (and circular) mistake to assume that what goes on internally inside a T3 device (or inside any device, other than a computer) consists only of computations, plus whatever hardware is needed to implement them. Once we have left the domain of T2 — where it could, in principle, be just symbols in, symbols out, and nothing but symbol-manipulation (computation) in between — we have entered the world of dynamics, and not just external or peripheral dynamics, but also internal dynamics.
And that includes the internal dynamics of blind, deaf or paralyzed people who, even if they’ve lost the power to see, hear or move, retain some or all of the internal dynamics of their sensorimotor systems — which are not, we should keep reminding ourselves, simple the dynamics of implementing computations.
SMIRSKY: “Would a CR that depended for all its inputs on what it is fed about other symbols be unable to achieve understanding in any but an imitative sense?”
If by “CR” you mean just a computer, then indeed any computer is just as vulnerable to Searle’s argument and to the symbol grounding problem, no matter what it is fed by way of input.
If you mean a T3 robot that grew on a tree and could pass T3 as soon as it fell to the ground, yes, in principle it should be able to continue with just verbal (T2) input, if it really has full T3 power. But nothing hangs on that, because it is not just a computer, computing. It also has whatever internal dynamic wherewithal is needed to pass T3.
SMIRSKY: “Suppose we were totally isolated from our sensory inputs about the world and had no inputs but a feed of digitized signals which had no analogical relation to the underlying causal phenomena which generated them because they were, say, generated in an abstract way (Morse code perhaps)?”
Same answer as above!
SMIRSKY: “(Didn’t Helen Keller learn about her world somewhat like this?)”
Helen Keller was not a computer. She was a human with some sensory deficits — but enough intact sensorimotor capacity for normal human cognitive capacity.
SMIRSKY: “After all, what are we but organic machines which receive inputs about the world through physical systems? What we know of the world begins with the signals our sensory equipment pick up and pass up the line through our nervous to our neurological systems.”
All true. But whatever we are, we are not just digital computers receiving and sending digital digital I/O. We are dynamical systems
— probably hybrid analog/computational — with T3-scale capacity.
SMIRSKY: “how is that significantly different (operationally rather than in terms of the physical platform) from what computers do”
We see, hear, touch and manipulate the things in the world. Computers just manipulate symbols. And tempting as it is to think of all of our “input” as being symbolic input to a symbol-manipulating computer, it’s not. It’s sensorimotor input to a hybrid analog/digital dynamical system; no one knows how much of the structure and function of this T3 system is computational, but we can be sure that it’s not all computational.
SMIRSKY: “the system [Searle] describes is specked to do no more than rote responding, symbol matching, though no one thinks that symbol matching is all we’re doing when we read and match symbols through understanding their semantic content”
T2 is “specked” to do everything that any of us can do with words, in and out. Passing T2 with computation alone is specked to do whatever computation can do. But for T3, which is necessarily hybrid — dynamic/computational, all purely computational bets are off.
SMIRSKY: “more complicated features… may arise from increasingly complex operations formed by what are otherwise much simpler operations.”
T2 can be made as “complex” as verbal interaction can be made. So can the computations on the I/O. But as long they are just computations (symbol-manipulations), be they ever so complex, the result is the same: Symbols alone are ungrounded.
And the transition from T2 to T3 is not a phase-transition in “degree of complexity.” It is the transition from just implementation-independent computation (symbol manipulation) to the dynamics of hybrid sensorimotor transduction (and any other internal dynamics needed to pass T3.
SMIRSKY: “Why should a series of ordered electrical signals yield a readable screen on my pc as I’m typing this?”
A person is looking at the shape on the screen. And the screen is not part of the computer (it is a peripheral device). If you see a circle on the screen, that does not mean the computer sees a circle. The objective of cognitive science is to explain how you are able to detect, recognize, manipulate, name and describe circles, not how a computer, properly wired to a peripheral device, can generate something that looks like a circle to you.
SMIRSKY: “The letters I see aren’t the series of signals that convert the lighted areas on my screen into what are, for me, so many words. And yet they do it and produce something in which I can find meaning even if the digital signals themselves carry no such meaning.”
The shape you see on a screen is indeed generated by a computer. But neither what you can do with that shape (detect, recognize, manipulate, name and describe it: T3) — nor what it *feels like* to be able to see and do all that — are being done by the computer. This is just as true for being able to read and understand what words mean: T3 is miles apart from T2. And input to a T3 robot is not input to a computer.
SMIRSKY: “Why then should we abandon hope that feeling (qua awareness) could be causally explained.”
None of what you have said so far has had any bearing on whether (or how) feeling could be causally explained. It has only been about whether all of our cognitive know-how could be accomplished by computation alone. And the answer is: No. At the very least, sensorimotor grounding is needed, and sensorimotor transduction is not computation; and neither the input to a sensory transducer nor the output of a motor effector is input and output to or from a computer.
The problem of explaining how and why whatever system does successfully pass T3 (or T4) feels is a separate problem, and just as “hard” if the system is hybrid dynamic/computational as if it had been just computational.
SMIRSKY: “If semantics involves the complexity of a system, why wouldn’t that same complexity be enough to account for the sense of understanding that accompanies instances of understanding in us in cases like this?”
Because “complexity” itself explains nothing at all.
Bernie Ranson
I came across this conference by chance, I’m highly appreciative of the willingness of Professor Harnan and the other participants to make themselves available to the public in this way. Unlike W. S Robinson, I am unsympathetic to Professor Harnad’s emphasis on robots, and I hope to explain why. Professor Harnad doubts whether there is any way to give a credible description of what anything feels like, without ever having felt anything: and suggests that the way to overcome this is to augment the computer with sensorimotor apparatus. The mistake here, I suggest, is to imagine that a robot or a robot/computer combination would feel anything more than the computer would on its own. Professor Harnad suggests that our robots and computers are able to do a tiny fraction of what we do. My submission is that they are not able to do even that tiny fraction, that what they do do is not even related to what we can do, or at least, not to the relevant part of what we do. The crucial distinction between our capacities and those of computer/robots can be brought into focus by considering the distinction made by Searle between “observer-dependent” and “observer-independent” phenomena. Examples of observer-dependent phenomena include money and computers. Something is only money when identified as such by an observer. Something is only a computer when identified as such by an observer. Examples of observer-independent phenomena include the metals used to make the coins we use as money, the metals and plastics etc we use to make computers, the physical processes in our brains, and consciousness. All these exist in nature independently of whether they are identified as such by an observer or not. The robot’s feeling something is observer-dependent. The robot’s existence as a discrete entity is observer-dependent. The syntax in the associated computer is observer-dependent, but also the semantics and even the computer’s very status as a computer. This is not a matter of intuitions, it’s more in the nature of a mathematical truth. The whole of what (we say) makes up the computer/ robot is available for our inspection, and we can see that the “meaning” is something additional to that inventory. We know where the “meaning” is, and we know it doesn’t go into the computer, not in the necessary observer-independent way. And the same is true for the robot, so that a robot/computer combination is no improvement on a computer alone. The crucial, observer-independent features are still missing. This is a conference about consciousness, but computers and robots are products of consciousness, and not in any sense its source, and as such they can not be expected to tell us much about its source, or its nature.
NEITHER COGNITIVE CAPACITY NOR FEELING IS “OBSERVER-DEPENDENT”
(Reply to Bernie Ranson — with apologies for the delay: I missed your commentary!)
B.RANSON: “Professor Harnad doubts whether there is any way to give a credible description of what anything feels like, without ever having felt anything: and suggests that the way to overcome this is to augment the computer with sensorimotor apparatus.”
But note that that’s only about T2, and hence its about verbal *descriptions* of what things feel like. And the solution is T3, in which the robot can actual have sensorimotor interactions with the things described. That grounds T2 descriptions of things, but it most definitely doesn’t explain how and why the T3 robot feels anything at all (if it does): The robot’s descriptions would be grounded by the data from its sensorimotor interactions with things whether or not it really felt like something to undergo those sensorimotor interactions; if the T3 robot did not feel, then it would simply have the requisite sensorimotor means abd data to ground its verbal descriptions — means and data lacked by a T2 system, but possessed by a T3 robot.
B.RANSON: “The mistake here, I suggest, is to imagine that a robot or a robot/computer combination would feel anything more than the computer would on its own.”
It might or might not be a mistake to conclude that a T3 robot really feels. Turing’s point was that if the robot’s performance was indistinguishable from that of a real, feeling person, then it would be arbitrary to ask for more in the case of the T3 robot. (We could ask for T4, even though we don’t ask for it in real people when we’re doing everyday mind-reading, but even T4 would not explain how and why the brain feels: it would only — like T3 — explain how the brain can do what it can do.)
So whereas it may (or may not) be true that a T3 robot would no more feel than a T2 computer does, adding T4 would not explain how or why it feels either — and it would not even improve on T3′s performance power.
B.RANSON: “Professor Harnad suggests that our robots and computers are able to do a tiny fraction of what we do. My submission is that they are not able to do even that tiny fraction, that what they do do is not even related to what we can do, or at least, not to the relevant part of what we do.”
Let’s not quibble over how tiny is tiny. But I agree that generating fragments of our total performance capacity has far more degrees of freedom than generating all of our performance capacity. Toy fragments are more “underdetermined,” and hence less likely to be generating performance in the right way. (That was part of Turing’s reason for insisting on *total* indistinguishability.) So today’s robots are just arbitrary toys; but that does not mean that tomorrow’s T3 robots will be. (And let’s not forget that we, too, are T4 robots, and feeling ones, but engineered long ago by the “Blind Watchmaker” [i.e., evolution] rather than reverse-engineered by future cognitive scientists.)
B.RANSON: “The crucial distinction between our capacities and those of computer/robots can be brought into focus by considering the distinction made by Searle between “observer-dependent” and “observer-independent” phenomena… [observer-dependent:… money and computers… observer-independent:… metals and plastics… physical processes in our brains, and consciousness]”
Interesting way to put the artificial/natural kinds distinction (except for consciousness, i.e., feeling, which does not fit, and is the bone of contention here); but how does this help explain how to pass the Turing Test, let alone how and why T3 or T4 feels?
B.RANSON: “The robot’s feeling something is observer-dependent. The robot’s existence as a discrete entity is observer-dependent.”
If the robot feels, that’s no more “observer-dependent” than that I feel. And the robot is an artifact, but if it had grown, identically, from a tree, it would not be an artifact. Either way, it can do what it can do; and if it feels, we have no idea how or why (just as we have no idea how or why we feel).
B.RANSON: “The syntax in the associated computer is observer-dependent, but also the semantics and even the computer’s very status as a computer.”
Syntax is syntax. The fact that it was designed by a person is not particularly relevant. What is definitely observer-dependent is the syntax of an ungrounded computer program. To free it from dependence on an external observer, it has to be grounded in the capacity for T3 sensorimotor robotic interactions with the referents of its internal symbols. Then the connection between its internal symbols and their external referents is no longer observer-dependent.
But that still leaves untouched the question of whether or not it feels; and it gives no hint of how and why it feels, if it does. Whether or not something or someone feels is certainly not (external) observer-dependent…
B.RANSON: “This is not a matter of intuitions, it’s more in the nature of a mathematical truth. The whole of what (we say) makes up the computer/robot is available for our inspection, and we can see that the “meaning” is something additional to that inventory. We know where the “meaning” is, and we know it doesn’t go into the computer, not in the necessary observer-independent way.”
You seem to think that if we build it, and know how it works, that somehow rules out the fact that it really feels. There’s no earthly reason that needs to be true.
Yes, the meaning of ungrounded symbols is in the head of an external observer/interpreter. But cognitive science is about trying to reverse-engineer what is going on in the heads of observers!
B.RANSON: “a robot/computer combination is no improvement on a computer alone. The crucial, observer-independent features are still missing.”
It’s certainly an improvement in terms of what it can do (e.g., T2 vs. T3). And since, because of the other-minds problem, whether it feels is not observable, it’s certainly not obserever-dependent.
B.RANSON: “This is a conference about consciousness, but computers and robots are products of consciousness, and not in any sense its source, and as such they can not be expected to tell us much about its source, or its nature.”
And if this were a conference about life, rather than consciousness, it could well be talking about how one synthesizes and explains life. There’s no earthly reason the same should not be true about cognition — except that feeling (which is surely as real and “observer-independent” property as living is) cannot be causally explained in the same way that everything else — living and nonliving, artificial and natural, observer-independent and observer-dependent — can be.
Stuart W. Mirsky
Professor Harnad writes in his response to JStern: “Searle’s Chinese Room Argument has been generating fun as well as insights for over 30 years now, but it is rather late in the day to resurrect the hoary old ‘System Reply’ without at least a new twist!” I want to be cognizant of this concern and keep the discussion focused on Professor Harnad’s paper in accordance with his understandable desire for that. But I also think it would be a mistake to let the above pass uncommented on. The so-called System Reply fails if one takes the Chinese Room scenario to represent ANY possible system consisting of the same constituents (computational processes running on computers) as those that make up the CR. But there is no reason to think that this failure can be generalized to all possible systems of this type. Just as certain biological systems have capacities which others lack (humans have understanding but it’s not clear that any and all living organisms do, though there are certainly debates about that), so the fact that the CR fails says nothing about other qualitatively equivalent systems. The CR may be seen to fail because it is inadequately specked, just as a microbe or a jellyfish or an arthropod or a mouse or a horse lack the capacities to understand because they lack the specs we humans have. Just because the rote processes which the CR is capable of performing are 1) merely mechanical operations (not conscious, and thus capable of understanding, in and of themselves) and 2) provide nothing like what we would expect of a genuine understanding entity in the combination that is the CR, it doesn’t follow that this is generalizable to all systems of the same qualitative (computational) type. The issue, I think, lies in what we think understanding is. Searle’s argument treats it as a bottom line feature of minds, assuming that, if it is present or even potentially present, it should be found there as a feature of one or more of the contituent elements of the CR (e.g., that understanding is a process level feature). But, in fact, there is no reason to expect to find it at THAT level when you think about it. If understanding is a function of process complexity, as it well may be, then it would not be surprising that an underspecked system (lacking the scope and complexity needed) would lack it — or that a more fully specked system would not. The real point to the System Reply is not that the CR understands even if Searle, as a system component doesn’t (because Searle’s CR doesn’t understand anymore than Searle, the component, does). It’s that understanding may well be best understood as a system level feature (the result of the interplay of many processes, none of which are, themselves, instances of understanding) rather than as a feature of any of the particular processes that make up the system itself. This is why I have suggested that we need to look more closely at the matter of meaning (understanding the semantics of the symbols). And here we come back to the paper under discussion. Professor Harnad strikes me as correct when he focuses on the issue of feeling (meant here as a term for the kind of awareness we have when we know or understand anything). But it seems wrong to me to think that the other aspect of this, meaning, is a matter of a referential relation between symbol and object in the world (finally achieved through the addition of a dynamical physical relation between the CR and the environment external to it). There is no reason to think that all meaning consists of this relation, even if some meanings (in some sense of the word “meaning”) happen to. Moreover, as we are always imprisoned in our own organically based mechanical contraptions, there can finally be no difference between the robot CR’s connections between symbol and object and that of the non-robot CR. In both cases, the referencing that’s going on involves representation to representation. That the representations may in whole or part trace their genesis back to external stimuli doesn’t alter the fact that the external stimuli we can talk (and think) about are whatever it is we hold in in our heads at any given moment. Even the external world, with all its complexity and detail, when understood as the knowledge we have of it, consists of the layered mapping we develop over the course of the natural history of our individual lives. Here is an example I have often used to show this: Driving up the east coast on a road trip with my wife some time back I saw a road sign in South Carolina (I believe that was the state) which read “Burn lights with wipers”. I did a double take and was momentarily confused. I had images in my head of a big bonfire with people tossing old wiper blades and light bulbs onto it. Then, in an instant, everything shifted and I got a different picture. I realized that what the sign meant was that motorists should turn on their headlights when running their windshield wipers in inclement weather, that the imperative to “burn” meant turn on, not set on fire. This recognition was accompanied by a different set of mental images where I saw myself leaning over to turn on the headlights on the dashboard as the wipers rythmically stroked the windshield in a heavy downpour. I even had a brief flash of myself failing to do that and suddenly crashing into another vehicle for lack of visibility. Lots of images passed through my head, replacing the one of the bonfire of wiper blades and light bulbs. But my actions never changed. It wasn’t raining after all and I had no need to act on the sign’s instruction. It’s just that I suddenly had images that made sense of the words. No overt behavior on my part followed the change in my mental imagery. I want to suggest that the meaning I found in the words lay in the complex of images which, I expect, are not shared in precise detail by any two people on earth. But the fact that MY images (the dashboard or the interior of the vehicle I visualized, or the exterior environment I “saw” or the way in which I visualized myself leaning toward the dashboard, etc.,) were unique to me didn’t prevent me from getting the sign’s meaning as it was intended by those who had posted the sign. It seemed to me at that moment that what constituted the meaning was not the references in any particular, nor was it my behavior (as there was no change in mine) nor was it something found in the sign outside my car with which I abruptly connected. Rather it was a certain critical mass of imagery which, however different from person to person, had enough in common with the signmakers, because of shared associations, etc., to provide enough common touchstones for an exchange of information on a symbol/reference level.
The meaning of the words did not lie in some physical grounding but in the web of mental images I had accumulated in my head over a lifetime up to that point. Now it’s certainly true that this accumulation was driven in large part by the kind of referential relations implied by the grounding thesis offered by Professor Harnad. But it was the images and their relations to one another that constituted my understanding, not the grounding of the words to anything in particular in the world at that instance of observation. There seems to be no reason, in principle, that the images could not have found their way into a system like me in some other way, i.e., by being put there as part of a data dump. This may not be how we get our mental images, of course, as there is every reason to believe we build up our stored information piecemeal and cumulatively over the course of our lives. But the way WE get them seems to be less important here than their role in the occurrence of understanding in the system. As to the matter of feeling, I want to suggest that insofar as meaning may be a function of complex associative operations within a given system utilizing retained and new inputs, why should we think it unlikely that being aware of what one is doing on a mental level (such as my introspective observation of a moment’s confusion on the road through South Carolina) is any less likely to be a function of a sufficiently complex system that can and does collect, retain and connect multiple representations on many levels over a lifetime? If so, there is no real explanatory gap to worry about once we recognize that the only point of explanation in this case is to explain what causes what. Why should an explanation that describers semantics as a function of a certain level of computational complexity not also suffice to account for the occurrences we think of as being aware of what is going on, of feeling what’s happening (in Antonio Damasio’s interesting way of putting this)?
ON WHAT IT FEELS LIKE TO MEAN
(Reply to Stuart W. Mirsky)
SMIRSKY: “[From the fact that passing T2 through computation alone] fails [to generate understanding] it doesn’t follow that this is generalizable to all systems of the same qualitative (computational) type.”
It generalizes to all computational systems.
(I don’t know what you mean by computational systems of “qualitatively different type.” I only know one type of computation — the kind described by Turing, Church, Goedel and Kleene, all equivalently and equipotently: implementation-independent, rule-based symbol manipulation; the symbols are systematically interpretable as meaningful, but their meanings are not intrinsic to the system. The symbol-manipulation rules (algorithms) are based on the symbols’ shapes (which are arbitrary), not their meaning. Syntax, not semantics. See, for example, the Turing Machine.)
What Searle shows is that cognition cannot be just computation. And that means any computation.
SMIRSKY: “If understanding is a function of process complexity, as it well may be, then it would not be surprising that an underspecked system (lacking the scope and complexity needed) would lack it — or that a more fully specked system would not.”
The reference to complexity is exceedingly vague. I know of no qualitative difference between “types” of computation based on the complexity of the computation. Computation is computation. And cognition is not just computation, no matter how complex the computation.
SMIRSKY: “understanding may well be best understood as a system level feature (the result of the interplay of many processes, none of which are, themselves, instances of understanding)”
Perhaps, but this too is exceedingly vague. I can construe it as being somehow true of whatever it will prove to take in order to pass T3. But it certainly does not rescue T2, nor the thesis that cognition is just computation.
SMIRSKY: “it seems wrong.. that… meaning, is a matter of a referential relation between symbol and object in the world (finally achieved through the addition of a dynamical physical relation between the CR and the environment external to it).”
It seems wrong because it *is* wrong! Connecting the internal symbols of a T3 system to their external referents gives the symbols grounding, not meaning.
Meaning is T3 grounding + what-it-feels-like-to-mean
SMIRSKY: “there can finally be no difference between the [T3] robot’s… connections between symbol and object and that of the [T2 system]. In both cases, the referencing… involves representation to representation.”
“Representation” covers a multiple of sins! If you mean computation, see above. If you mean non-computational processes, all bets are off. Now the trick is to find out what those “representations” turn out to be, by reverse-engineering them, T3-scale…
SMIRSKY: “the external stimuli we can talk (and think) about are whatever it is we hold in in our heads at any given moment.”
Yes, and then the question is: what has to be going on in our heads to give us the capacity to do that? The way to find out is to design a system that can pass T3. (We already know it can’t be done by computation alone.)
SMIRSKY: “I… had images that made sense of the words… the meaning was not the referen[ts]… nor… my behavior…it was a certain critical mass of imagery … not… some physical grounding but… the web of mental images… and their relations to one another that constituted my understanding, not the grounding of the words.”
No doubt, but the question remains: what has to be going on in our heads to give us the capacity to do that? The way to find out is to design a system that can pass T3. (We already know it can’t be done by computation alone.)
SMIRSKY: “There seems to be no reason… the images could not have found their way into a system like me in some other way… [e.g.] put there as part of a data dump.”
I agree that a real-time prior history — although it is probably necessary in practice — is not necessary in principle in order to have full T3 power. (E.g., to recognize, manipulate, name, describe, think about and imagine apples, you need not have had a real-time history of causal contact with apples in order to have developed the requisite apple-detectors and apple-know-how. They could have been built in by the engineer that built you — even, per impossibile, the Blind Watchmaker. But it is not very likely. And, either way, whether inborn or learned, that T3 power will not be just computational. Hence building it in in advance would entail more than a data-dump.)
SMIRSKY: “[just as] meaning may be a function of complex associative operations within a given system utilizing retained and new inputs… [so] being aware of what one is doing on a mental level… is… a function of a sufficiently complex system that can and does collect, retain and connect multiple representations on many levels over a lifetime?”
“Computational complexity” is already insufficient to constitute sensorimotor grounding. It is a fortiori insufficient to explain feeling (hence meaning).
But let me clarify one thing: when I substitute “feeling” for all those other weasel words for “conscious” and “consciousness” — “intentionality,” “subjectivity,” “mental,” etc., I am not talking particularly about the quality of the feeling (what it feels like), just the fact that it is felt (i.e., the fact that it feels like something).
As I think I noted in the target essay, I may feel a toothache even though I don’t have a tooth, and even though it is in reality referred pain from an eye infection. So it is not my awareness that my tooth is injured that is at issue (it may or may not be injured, and I may or may not have a tooth, or even a body!). What is at issue is that there is something it feels like to have a toothache. And I’m feeling something like that, whether or not I’m right about my having an injured tooth, or a tooth at all.
By exactly the same token, when I suggest that meaning = T3 capacity + what-it-feels-like to mean, what I mean is that I may be able to use a word in a way that is T2- and T3-indistinguishable from the way anyone else uses it, but, in addition, I (and presumably everyone else who knows the meaning of the word) also know what that word means, and that means we all have the feeling that we know what it means. I may be wrong (just as I was about my tooth); I (or maybe everyone) may be misusing the word, or may have wrong beliefs about its referent (as confirmed by T3). But if there is nothing it feels like to be the T3 system using that word, then the word merely has grounding, not meaning.
SMIRSKY: “there is no real explanatory gap to worry about once we recognize that the only point of explanation in this case is to explain what causes what. Why should an explanation that describes semantics as a function of a certain level of computational complexity not also suffice to account for the occurrences we think of as being aware of what is going on, of feeling what’s happening”
Because computational complexity can neither produce grounding nor explain feeling.
On the Systems Reply (response to Stevan Harnad): There seems to be a significant miscommunication here. In the classic Systems Reply, I have always taken “the system” to include all of the sensorimotor components that are needed for the system to work in the real world. However, what you express as a paraphrase of Searle, seems to quite deliberately exclude that part of the system.
THE RIGHT ROBOT/SYSTEM REPLY
(Reply to Neil Rickert)
RICKERT: “In the classic Systems Reply, I have always taken “the system” to include all of the sensorimotor components that are needed for the system to work in the real world. However, what you express as a paraphrase of Searle, seems to quite deliberately exclude that part of the system.”
Quite deliberately.
In T2, the peripherals by which the Chinese symbol input are received from the pen-pal are trivial. So are the peripherals by which Searle’s Chinese symbol output are transmitted to the pen-pal. Symbols in, symbols out, and nothing but symbol-manipulation in between. It think it would stretch credulity to the point of absurdity to say that even if Searle memorized all the algorithms and executed them in his head, he still wouldn’t be the whole system, because the whole system consists of that plus the input and output itself.
That’s like saying I don’t understand English. It’s the system consisting of me and the words that understands.
No, the only point at which the peripherals become important is in the “Robot Reply,” which (in the original Searle paper and accompanying commentaries in BBS, which I umpired!) was a defective objection. The original robot reply ran along the lines of saying that T2 needed peripherals in order to connect with the world robotically. Searle responded, fine, I’ll still do the computational part, and I still won’t understand.
The right Robot Reply would have been to demand not just T2 power plus robotic peripherals, but T3 power, because in that case Searle really would not be, and could not be, the entire System, just its (implementation-independent) computational component.
That would be a valid “System Reply,” but it would be a reply about T3 power and not just T2 power, and it would purchase its immunity to Searle’s Chinese Room Argument (that cognition is not just computation) at the price of conceding that cognition is indeed not just computation! The “System” that understands has to include the sensorimotor dynamics (at least).
Harnad: "In T2, the peripherals by which the Chinese symbol input are received from the pen-pal are trivial."
Agreed. However, I don’t see that as contradicting the “systems reply”.
According to the “systems reply”, if AI is successful, then the intentionality will be in the system. As I see it, if the peripheral system is trivial, then AI will not be successful. And that is consistent with the “systems reply.” I expect that the kind of system that Searle was envisaging would fail a rigorous Turing test. We wouldn’t need to go to a TTT to see that.
When asked about the source of intentionality, Searle says it is due to the causal powers of the brain. That’s pretty much a restatement of what I take the “systems reply” to say.
Harnad: "That’s like saying I don’t understand English. It’s the system consisting of me and the words that understands.|
I am left wondering what you mean by that “I”. Surely, it includes the system.
“AI” VERSUS COMPUTATIONALISM
(Reply to Neil Rickert)
NRICKERT: “According to the ‘systems reply’, if AI is successful, then the intentionality will be in the system. As I see it, if the peripheral system is trivial, then AI will not be successful. And that is consistent with the ‘systems reply.’”
“AI” is a loosely defined field. Who knows what is true or untrue according to a field.
In contrast, “computationalism” is a thesis (somewhat close to what Searle called “Strong AI”).
According to computationalism, cognition is just computation.
Searle’s Chinese Room Argument shows that for T2, passed by computation alone, computationlism is false.
T3 cannot be passed by computation alone. Passing it any which way may vindicate “AI,” but it does not resurrect computationalism.
Response to Stuart W. Mirsky on 2/18 at 16:36 Mirsky: “The use [of ‘feeling’] you allude to is the emotional aspect of the word ‘feeling’ . . .” Robinson: No, I was thinking of an alleged *cognitive* feeling – the alleged feeling of understanding. (Besides what Prof. Harnad says, you can find claims to such feelings in, e.g., D. Pitt, G. Strawson, C. Siewert, etc.) But I think we can cut through some of this by observing that *feeling* in the target paper, in the other authors I just mentioned, in my own work, and in commonsense, requires consciousness. Robots (I stipulate, but I believe agreeably to others in this discussion) have no consciousness. Therefore, they have no feelings. Mirsky: “. . . intelligent in the way we use the term for ourselves much of the time (i.e., conscious), . . . .” Robinson: I don’t think we use “intelligent” to include consciousness by analytic implication. For, I think most readers of this discussion think it’s a non-trivial question whether a computer, or a robot, both of which they take to lack consciousness, could be intelligent – which they couldn’t do if they were including consiousness as partially constitutive of intelligence. (It’s arguable, I realize, but I think Turing agreed; i.e., he did not think that a machine’s doing well on the imitation game would show that it was conscious.) At any rate, I am holding out for a tripartite distinction: Intelligence (showable by success on T2), understanding (showable by success on T3) and consciousness which requires more than success on T3 (but it’s in dispute exactly what *would* show it). I’ve no wish to dispute about words, and I realize that all of the terms in this discussion are sometimes used in ways other than I’ve just described. But we need *some* words to mark important distinctions. I’ve used “intelligence” because that was Turing’s term, and “understanding” because that was Searle’s. The reason that’s implicit in Searle’s paper (though, as I said, he mixes it in with other points) for why he doesn’t understand is that he can’t do anything non-verbal with Chinese symbols. His program enables him to connect words with words, but there’s nothing in the situation that lets him connect words to non-words, e.g., to his hunger or to hamburgers. In Harnad’s terms, T3 is not satisfied. So Searle in the CR doesn’t have what he called “understanding”. One of my leading points was that failure on T3 is sufficient for lack of understanding. Absence of consciousness just doesn’t come into Searle’s reasoning. Neither does lack of a cognitive feeling (which he could have with or without actual understanding). And since feeling doesn’t come into the matter, absence of feeling does not show that a robot – a device that *does* make word-tothings connections – does not have understanding (again, taken to be the property that Searle is arguing about in “Minds, Brains and Programs”). Paraplegics can’t do much, but they can issue requests that get others to act on their behalf. Searle in the CR can’t do that. Even though he knows the people outside can read Chinese, he has no reason to write any particular collection of symbols in order to get himself a hamburger; he can’t connect his hunger to the symbols. It’s all just words to words, never words to things or things to words. Mirsky: “[We need an account that] adequately captures what happens in us when we have understanding.” Robinson: That’s too broad. Lots of things happen in us when we have understanding, but that doesn’t show that all of them are constitutive of what understanding is; it doesn’t show that they are necessary to understanding rather than its normal accompaniments in us. Searle’s “understanding” was denied to the man in the CR (and even after the internalizing of the scripts and program in memory) because of lack of word-world connectability – in Harnad’s terms, lack of success at T3. Consciousness doesn’t come into it. Searle was *not* arguing that formal symbol manipulation as such could not yield understanding on the ground that it could not yield consciousness. It was a bad argument, but it wasn’t *that* bad,
PLEASE LET’S MOVE ON TO THE “HARD” PROBLEM AND THE EXPLANATORY GAP!
(Reply to William S. Robinson)
WROBINSON: “Robots (I stipulate, but I believe agreeably to others in this discussion) have no…feelings”
Not agreeable to me! For me, robots are simply causal systems. *We* are feeling robots; and the task of cognitive science is to reverse-engineer our cognitive capacity, i.e., find out what kind of robots we are, and how we work (by designing a robot that can pass T3).
Trouble is that this will explain our cognitive capacities, but it will not explain how and why we feel.
Moreover, I think it’s claiming far too much to say that robots are, by stipulation, or by definition, or of necessity, unfeeling. That would rather prejudge what is surely a factual question. (I would have said an empirical question, but because of the other-minds problem, it is in fact an undecidable empirical question.)
WROBINSON: “it’s a non-trivial question whether a computer, or a robot, both of which [we] take to lack [feelings], could be intelligent”
It’s trivial if we’re just legislating what we choose to call “intelligent” or “understanding” or whatever. It is nontrivial if we are concerned about whether and why our reverse-engineering of cognitive capacity fails to explain the fact that it feels like something to have and use our T3 capacity.
WROBINSON: “failure on T3 is sufficient for lack of understanding. Absence of consciousness just doesn’t come into Searle’s reasoning.”
Searle only appealed to the unexamined, a-theoretical notion of “understanding Chinese” in order to conclude (correctly) that if he executed the T2-passing computations, he would not be understanding Chinese. That’s all.
But although it might be worthwhile querying Searle about this, I don’t think that the basis on which he was concluding that we would not be understanding in the Chinese room was that he would fail T3 in Chinese! (That’s the symbol grounding problem, not the Chinese Room argument: let’s not mix them up!)
Searle’s basis for concluding that he didn’t understand Chinese would be the same as mine: When I hear Chinese (or see it written) I have no idea what it means. It feels like I’m listening to meaningless vocalizations, or looking at meaningless squiggles and squoggles. That feels very different from what it feels like to hear, speak or write English, which I do understand.
And that would be true even if a Chinese pen-pal told Searle, truly: no, you’ve been communicating coherently with me in Chinese for 40 years. Searle would still reply (truly, and simply by consulting what it feels like to understand and not-understand) that, no, he had simply been doing meaningless symbol manipulations according to memorized rules for 40 years, and, no, he could not understand a word of Chinese.
Of course Searle knows that not understanding Chinese means, among other things, having no idea what Chinese symbols refer to in the world. But he doesn’t have to go that deep to say he doesn’t understand Chinese. Besides, it wasn’t robotic capacity in the world (T3) that was on trial in the Chinese room, it was T2.
WROBINSON: “Searle was *not* arguing that formal symbol manipulation as such could not yield understanding on the ground that it could not yield consciousness.”
Nor am I. But he was arguing that under such conditions he would not be understanding Chinese, and that on that topic he was the sole authority, not the Chinese pen-pals who insisted he did understand Chinese (because they could not distinguish his letters from those of a real pen-pal), nor the dedicated computationalists, who likewise insisted that he (or “The System”) did understand Chinese, because he had passed T2. Searle, the sole authority, made the privileged 1st-person judgment that only he was in a position to make: “I know what it feels like to understand Chinese, and I don’t understand Chinese.”
He could say that with as much Cartesian authority and certainty as he could say that he had a toothache (though not that he had a tooth injury, or even a tooth).
T3 incapacity had nothing to do with it. (And that was my point!)
WROBINSON: “what about visual object recognition? If that’s included, it doesn’t seem to be what Turing had in mind”
This is again about T3 vs. T2. And, yes, the punchline is that if Turing did just mean T2, and passing via computation alone, then Turing was mistaken; he ought to have included T3, and dynamical processes, if need be. (And I rather think he did mean to include the latter.)
WROBINSON: “Here’s a candidate for a definition of intelligence that seems to fit with Turing’s paper: Ability to respond appropriately to a wide range of novel circumstances.”
That would be just about as unrigorous and unacceptable as Turing’s own (inexplicably loose) suggestion that “in about fifty years’ time it will be possible, to programme computers… [to] play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning” — in stark contrast with his earlier suggestion in the same paper that “a statistical survey such as a Gallup poll [would be] absurd…”
I prefer to attribute this to the kind of loose speaking that great mathematicians often engage in, where they leave out the details that are intuitively obvious to them, but need to be rigorously proved by others in order to be understood and believed.
Turing was spot-on with his criterion of total indistinguishability in performance capacity (what I dubbed as “cognition is as cognition does”), but this does not entail a commitment to verbal performance capacity alone, nor to computation alone. On the contrary, the notion of Turing-indistinguishability immediately generates the Turing hierarchy, from arbitrary toy fragments of performance, to T2, T3 and T4, showing that the Turing performance hierarchy is actually an empirical observability hierarchy.
WROBINSON: “What about my shopping assistant robot, that gets around town under many variable and unforeseen circumstances”
The shopping assistant robot is a toy; it already fails T2 and T3, so you need inquire no further about whether or not it understands… (And grounding, too, is a T3-scale property, not a property of arbitrary fragments of robotic capacity.)
WROBINSON: “the + sign [in "understanding (etc.) *means* cognitive capacity + feeling"]… seems to indicate that you *do* make a distinction between cognitive capacity and feeling.
Yes of course. If not, then T3 (or T4) would be face-valid, and would constitute cognition by definition. There is (and always was) more at stake in this all along, no matter how coy or dismissive Turing affected to be on the question of the mind!
WROBINSON: “I’m trying to follow Searle’s usage, and his argument doesn’t bring up feeling”
Don’t get too hung up on Searle’s use of “understanding” in his argument. He really just meant the ordinary, everyday notion of understanding or not understanding a language. And he doesn’t bring up the fact that he is performing a simple introspection when he judges that he doesn’t understand Chinese because it’s just too obvious.
WROBINSON: “I hold that passing T3 could happen with cognitive capacity, period. You hold that it could be passed only with cognitive capacity + feeling.”
Of course passing T3 could happen with cognitive capacity alone. That’s true by definition, because T3 is supposed to be the generation of our cognitive capacity. The question is whether it would generate understanding (or seeing, or hearing, or meaning) or any other of the felt states that normally accompany the possession and exercise of our cognitive capacity.
And — if I could only wean this discussion away from the oh-so-seductive allures of the Chinese room — there’s still the problem of how and why cognitive states are felt that needs to be addressed…
Stuart W. Mirsky
The pre-posting review process in this discussion seems to have become backed up. As I am going off-line and may not return for awhile, I thought I would take the opportunity to offer some quick replies now to Professor Harnad’s latest response to some of my comments (received initially via e-mail but not yet appearing here). I fear this is a little unfair, as I cannot repeat his full text here and must confine myself to responding selectively which runs the risk of losing context or the full thrust of his remarks. I therefore hope this will not be taken the wrong way (and that the moderator(s) here will adjust the queue of responses so they follow the line of discussion if that’s needed). On the other hand, maybe this will also just go into the queue and come out in a more appropriate spot in the discussion). Professor Harnad writes: “. . . the most important point is that it is a big (and circular) mistake to assume that what goes on internally inside a T3 device (or inside any device, other than a computer) consists only of computations, plus whatever hardware is needed to implement them. Once we have left the domain of T2 — where it could, in principle, be just symbols in, symbols out, and nothing but symbol-manipulation (computation) in between — we have entered the world of dynamics, and not just external or peripheral dynamics, but also internal dynamics.” Responding: If the issue is what does a brain do to produce understanding (and all the other features we lump together under the term “consciousness” as applied to ourselves), then the idea that hardware or peripherals adds something which moves us out of the computationalist ballpark is misleading. No actual computer is ever pure theoretically (an abstract, uninstantiated model) and no AI researcher I know of thinks an AI application can be implemented without hardware. In and of itself, the hardware doesn’t matter, even if we always need whatever hardware is sufficient to run whatever programs are at issue. If that includes devices to provide inputs, well and good. But there’s no reason to think that inputs must enter the system in some particular way. As to what brains actually do, it’s at least possible that they don’t only compute (in whatever way brains might be said to do that) but what is at issue here is whether the computing part of what they do is the important part in producing understanding and other features of consciousness. Professor Harnad continues: “If by ‘CR’ you mean just a computer, then indeed any computer is just as vulnerable to Searle’s argument and to the symbol grounding problem, no matter what it is fed by way of input.” Responding: It isn’t at all clear to me that Searle’s argument succeeds, though perhaps some still think it does, in which case what vulnerability? (As I’ve written earlier, I think the Chinese Room argument fails on a number of counts which we have chosen not to discuss here because of the risk of distraction. But choosing not to discuss it is not to agree that it’s unchallengeable or that it’s even on the right track.) Professor Harnad: “Helen Keller was not a computer. She was a human with some sensory deficits — but enough intact sensorimotor capacity for normal human cognitive capacity.” Responding: Thus she had a source of inputs, albeit more limited than ours. But why should we think that how those inputs were delivered (so long as they were sufficient to convey information to her) mattered beyond the bare minimum of conveying information sufficient for her brain to do its cognitive work? If the mode of delivery doesn’t matter, of course, then we are back to the same question of whether her brain relied on computational type operations to perform the processing that made sense of the inputted signals. Professor Harnad: “We see, hear, touch and manipulate the things in the world. Computers just manipulate symbols. And tempting as it is to think of all of our “input” as being symbolic input to a symbol-manipulating computer, it’s not. It’s sensorimotor input to a hybrid analog/ digital dynamical system; no one knows how much of the structure and function of this T3 system is computational, but we can be sure that it’s not all computational.” Responding: I think it’s a fair point to note that we are more than just a digital operating system. But I don’t think it’s a telling one because no one denies it and denying it isn’t essential to a claim that the way brains produce understanding is in a computational way. Professor Harnad: “T2 can be made as ‘complex’ as verbal interaction can be made. So can the computations on the I/O. But as long they are just computations (symbol-manipulations), be they ever so complex, the result is the same: Symbols alone are ungrounded.” Responding: The question is what does it take to ground a symbol and thereby impute meaning to it? You suggest that the dynamical relation with the external world is the grounding and that this grounding establishes the meaning. But I’m suggesting that grounding is more likely an outcome of the process which establishes meaning, i.e., that grounding turns out to be one kind of meaning we find in our linguistic activity (as in the word and object references so common to so many of our words). I’ll add here that grounding seems to occur in at least one other, albeit somewhat different, sense: It is a way of describing the mechanism whereby we build up our representational mapping webs which picture the world. In that sense it is pre-linguistic though it does form the basis for the linguistic (and conceptual) capacities which follow it, among which are included the word-object referential relation which last seems to be what you mean by grounding-as-meaning — at least part of the time. But neither use of “grounding” effectively describes, I think, the actual process whereby meaning occurs. That, I suggest, happens within an associative process that the brain performs (among its other “duties”) as I’ve begun to try to elucidate nearby (see the anecdote of the road sign). Professor Harnad: “A person is looking at the shape on the screen. And the screen is not part of the computer (it is a peripheral device). If you see a circle on the screen, that does not mean the computer sees a circle. The objective of cognitive science is to explain how you are able to detect, recognize, manipulate, name and describe circles, not how a computer, properly wired to a peripheral device, can generate something that looks like a circle to you. “The shape you see on a screen is indeed generated by a computer. But neither what you can do with that shape (detect, recognize, manipulate, name and describe it: T3) — nor what it *feels like* to be able to see and do all that — are being done by the computer. This is just as true for being able to read and understand what words mean: T3 is miles apart from T2. And input to a T3 robot is not input to a computer.” Responding: My only point was to ask why we should think that a computational process, consisting of lots of smaller operations, none of them conscious themselves, should be incapable of working together in a larger system to produce a feature we recognize as “consciousness” (or its various components) at a “higher” level of operation? I was aware that my example of the computer screen and its images could be misread, but hoped my point would have come through nonetheless. It’s my error for not going into more detail there. Anyway, my main point is really twofold: 1) To question the assumption you apparently make that meaning is found in a dynamical relation with the environment (I think that certainly plays a part for us but that it is NOT the factor that accounts for meaning per se — see my comments nearby about the road sign again); and 2) To ask why, if complexity is the true issue for meaning (as I have suggested), it should not also be so for what you are calling the feeling of understanding that accompanies the kind of understanding we have? I have seen your comment that complexity does not explain understanding but that strikes me as mainly asserted, as of now. It hinges on another claim that the dynamical connectivity, as seen in your T-3 example, explains at least an aspect of meaning, i.e., meaningful responses without, but touching on the feeling part. My view differs in that I don’t see how robot-like dynamics can make much of a difference or account for the occurrence of meaning in symbols — nor do I think we must give up the expectation of finding a way to causally explain the occurrence of the feeling aspect of understanding as you argue for. But I have enjoyed reading (and listening to) your presentation and getting a better sense of where you’re coming from in this debate. I expect, however, that we will not find a lot of agreement, going forward, given where we now are.
Response to Stevan Harnad on 2/18 at 17:30 Harnad: “. . . having intelligence is synonymous with having indistinguishable cognitive capacities. (What else does it mean?) Robinson: It depends on what one includes under “cognitive capacities”. For example, what about visual object recognition? If that’s included, it doesn’t seem to be what Turing had in mind in his paper on Computing Machinery and Intelligence, because, manifestly, there was nothing in the imitation game to test for *that* capacity. Here’s a candidate for a definition of intelligence that seems to fit with Turing’s paper: Ability to respond appropriately to a wide range of novel circumstances. (Appropriate response goes with inability of interrogators to reliably distinguish. Wide range with absence of restriction on content of their questions. Novel with the fact that machine designers do not get the interrogators’ questions in advance.) No one has legislative power over the key terms in this discussion and not everyone has to mean my candidate by “intelligence”. But I think my suggestion at least describes an interesting property, and that it’s important to distinguish this property from some others (such as Searle’s *understanding*, and your *feeling*). Once distinguished, we can, of course, meaningfully ask whether there being an M with intelligence requires or does not require that it have one or another further property. Harnad: “. . . one question unanswered: Does the TT-passer feel . . .?” Robinson: I agree that passing T2 does not answer this question. I’d say: Passing T2 does not require feeling (though, of course, something might be a T2 passer and have feelings too!) I think Turing would give the same answer. What about my shopping assistant robot, that gets around town under many variable and unforeseen circumstances, describes what happens accurately, matches behavior to statements of what it’s about to do, etc.? That also does not imply that it feels anything at all. But its words are not just connected with other words: they are appropriately related to the things that are the causal inputs to its sensors and are manipulated by its effectors. That’s another important property that we should distinguish from others. Again, there is not and likely will not be a standard usage here; but I suggest “understanding” for this one, because it’s this property that Searle’s argument targeted when he argued that no system that was just a formal symbol manipulator could *understand* its words. Harnad: “I . . . believe . . . that it is . . . unlikely that anything could pass T3 unless it really felt, because, for me, understanding (etc.) *means* cognitive capacity + feeling. Robinson: I’m intrigued by the + sign. Because it seems to indicate that you *do* make a distinction between cognitive capacity and feeling. This distinction makes it possible to sharpen an issue, so long as we take some verbal care – needed, because I do not use “understanding” in such a way that it just *means* cognitive capacity + feeling (because I’m trying to follow Searle’s usage, and his argument doesn’t bring up feeling (i.e., feeling in your sense, i.e, consciousness)). So, I hold that passing T3 could happen with cognitive capacity, period. You hold that it could be passed only with cognitive capacity + feeling. The clarified issue is: Why? Why can’t my description of the feelingless shopping assistant robot coherently apply to some possible device? Harnad: “the hard problem of explaining how and why we feel is not just a moral matter.” Robinson: I couldn’t agree more. My remark about feelings and morality was about the importance of feelings, a certain difference they make. It was not directed at the question of explaining how or why we feel at all; it was in no way intended to be addressing the explanatory gap.
Stuart W. Mirsky
Quick Response to William Robinson: Okay, I see you are making the distinction I thought you were missing. Professor Harnad’s use of “feeling” in this context only refers to the sense of being aware of what we know, when we know it, the sense that accompanies our instances of understanding or knowing (at least much of the time). You do, in your comment to me, seem to be assuming that no robot could have such an awareness though, just because it’s a robot and I think that’s putting this cart before the horse here. In fact that is precisely what’s at issue (insofar as we presume the robot has a computational type processing platform in its “head” serving as its brain). I agree that Searle addresses what he terms “understanding” in his Chinese Room Argument (an argument he later replaced, but did not fully recant, with a later claim that computational understanding is simply unintelligible — I’m thinking here of The Mystery of Consciousness for starters). However, he has often invoked his CRA in the context of discussing consciousness so it is probably a mistake to make the kind of hard and fast distinctions you want to make between: intelligence understanding consciousness As you note, these terms all admit of multiple and somewhat varied applications. We can speak of intelligent toasters and dumb toasters and never mean that the toaster could follow a conversation or do a calculation or comment on Professor Harnad’s paper on a site like this. We speak of humans as more intelligent than horses and yet we also speak of intelligent and unintelligent humans where “intelligence” means something other than having more cognitive capacity than a horse. The same wide range of uses can be found with “understanding” and, certainly, with a term like “consciousness”. I think that Marvin Minsky’s point that “consciousness” is a “suitcase word” is correct though it is not an especially deep insight as it should be readily apparent when we look at how we use the term. What we have, whenever we start applying mental words, is increasing slipperiness of application and any discussion like this has to deal with it. That said, I think it’s wrong to take Searle as speaking only about understanding in his Chinese Room Argument. Aside from his raising his arguments in books like his Mystery of Consciousness (which specifically references consciousness!), he has often applied them to questions about consciousness. There are, as I recall, a number of on-line videos where he is to be seen discussing consciousness specifically in which he invokes his CRA. If understanding is not a feature of consciousness then what’s the point? We know that machines can be built to replicate many kinds of human judgments. But it is the thing Searle calls “strong AI”, the claim that computers can be programmed to have minds (meaning our kind of consciousness) that matters to him. And it is that that should finally be what matters here. The best way of seeing this is to recognize that the role played by “understanding” in the CRA is that of a proxy for consciousness and that what he there calls “understanding” is presented as one of many features we take to be parts or aspects of consciousness. Among others he has explicitly included things like intentionality, awareness and feeling (sometimes called “qualia” though he isn’t apparently happy with the term, thinking it redundant). As Professor Harnad notes in his video lecture associated with this paper, all these words tend to run together and overlap in their meanings. It’s a function, as I’ve said, of words about our mental lives. We can and should try to nail down what we mean precisely in discussions like these but there is something about these words, and their realm of application, that militates against that and constantly pushes us into ambiguity. Try as I have for many years to hammer out a clear and precise jargon for such discussions as this, it has never quite worked. I always find myself being misunderstood or being accused of having misunderstood others. Most times, on examination, the problem seems to boil down to the inherent willow-the-wisp quality of our mental words.
Hi everyone! Great discussion going on in here! Several times Stevan says that there is more to thinking than computation. there is also feeling. He compares this to Dave’s Hard Problem and argues that “Turing Machine’s can explain all that is explainable but do not explain everything there is.” Yet, at the same time he claims that he is not a dualist, and not even a “naturalistic” property dualist like Dave. He even rejects zombies and conceivability arguments in general. So, I am curious as to what Stevan’s response to the the zombie arguments are. The view as put forth so far sounds just like property dualism. In fact Dave does speculate that perhaps all of reality is computational and that information has dual aspects, one physical one qualitative. How could feeling be non-computational and yet also physical? For my part, I think that as science approaches the limit we will be able to make deductions from facts specified solely in microphysical terms to facts about feeling (given that we have actually felt the things in question. That is, given that I have had a conscious experience of red I will be able to deduce from some physical description that it is like seeing red for the experiencer that has it) and this makes qualitative facts physical. As Stevan says, we are feeling robots, but that we feel can be explained physically. That we can’t yet see how this can be done is not a very convincing argument.
ON ZOMBIES AND OTHER COUNTERFACTUAL FANTASIES
(Reply to Richard Brown)
RBROWN: “Several times Stevan says that there is more to thinking than computation.”
Yea, more e’en than computation + dynamics…!
RBROWN: “there is also feeling. He compares this to Dave’s Hard Problem”
Indeed, it *is* Dave Chalmers’s “Hard Problem,” except I am arguing that it is an explanatory rather than an ontic problem, and that it is insoluble, for reasons that have to do with the nature of both feeling and causal explanation.
RBROWN: “and argues that ‘Turing Machines can explain all that is explainable but do not explain everything there is.’”
That’s not a quote! Turing machines (i.e., computation) can explain part of cognition, but not all of our cognitive capacity.
Computation + dynamics can explain all of our cognitive capacity — verbal (T2), sensorimotor (T3) and even neurobehavioral (T4) — but they cannot explain how and why cognitive states are felt states: They cannot explain how and why we feel.
RBROWN: “Yet, at the same time [Stevan] claims that he is not a dualist, and not even a “naturalistic” property dualist like Dave. He even rejects zombies and conceivability arguments in general. So, I am curious as to what Stevan’s response to the the zombie arguments are. The view as put forth so far sounds just like property dualism. In fact Dave does speculate that perhaps all of reality is computational and that information has dual aspects, one physical one qualitative. How could feeling be non-computational and yet also physical?”
Yes indeed. I am not a dualist, not even a “property dualist.” My gap is truly just epistemic (i.e., explanatory), not ontic. My question is: How and why do we feel? It does not even represent an epsilon of explanatory inroad to reply to me by saying: “Listen, my son, there are many properties under the sun: bigness, littleness, redness, mass, energy, spin, parity, being a prime number… but one out of them — namely, feeling — is different: it can’t be explained in the way all the others are.”
I take that statement to be true, just so, but empty, and just a confession that my question “How and why do we feel?” is indeed different, hard, possibly unanswerable.
Well I already knew that, and telling me there is a “duality” among properties informs me nary a whit further! (Information is the reduction of uncertainty among alternatives. “Property dualism” does not reduce uncertainty, it just re-asserts it.)
Ditto for “zombies”. The flip side of the “mind/body problem” (which I prefer to call the “feeling/doing problem”) is the “other minds problem” (you cannot know for sure whether others feel). Another way of putting this is that you cannot know for sure that others are not “zombies.”
Now the other-minds problem can be construed in three different ways. Two are the usual sceptical ways: (1) soft scepticism and (2) hard scepticism. Soft scepticism is just a Cartesian/Humean admission that there are some true things that are true, but you can’t know for sure that they are true: this includes the truth of scientific laws, the future reliability of past regularities, the existence of the outside world, and the existence of other minds. According to soft scepticism, it’s not that any of those things are untrue; it’s just that — unlike the necessary truths of mathematics or the face-valid truth of the Cogito (namely, the fact that feeling is being felt when feeling is being felt) — these other truths can only be known with high probability rather than certainty.
In contrast, hard scepticism (ontic rather than just epistemic) countenances the possibility that these other truths are false. In particular, the conjecture that there could really be zombies is an instance of such hard scepticism about other minds. I don’t go there, for much the same reason I don’t bother with dualism. I find them uninformative and question-begging, when the question is simple, and straightforward: “how and why do we feel?” — not “What scope for sci-fi does that explanatory gap leave for us?” But the third way the other-minds problem can be construed is (3) as a somewhat more profound uncertainty than the one that attends the rest of the objects of scepticism. This is already evident in the case of fellow creatures other than our own species: Our degree of uncertainty about whether other people really have minds is pretty much the same as our degree of uncertainty about empirical regularity, the existence of the external world, etc., namely, tiny, and tractable. But our degree of uncertainty about other minds grows as we move to species other than our own, and becomes really quite sizeable when it comes down to simple invertebrates, unicellular creatures, and plants (an especially vexed question for vegans like me!).
That extra continuum of uncertainty is special, and sets the other-minds problem apart, because, unlike, say, uncertainty about our current best explanation in physics, which can improve with time and further data and ideas, our uncertainty about whether microorganisms or plants feel remains forever unresolvable, and its an outsize uncertainty, compared to the others: plants might or might not really be “zombies.” (I certainly hope they are!)
That does not make the possibility that other humans could be zombies one whit more worthy of serious consideration, except in one perhaps not entirely verbal sense. The question “How and why do we feel?” can be seen to be equivalent to the question “How and why are we not “zombies”? This does not give zombies any more credence or substance; it simply points out an “epistemic” equivalence of the two questions: they are simply flip sides of the same coin. We know of ourselves that we are not zombies. The worry about whether everyone else might be a zombie is just ordinary soft scepticism (also known as “empirical risk”). It does not make zombies any more real. Let’s say bets are off with other species. But the special status of the other-minds problem does reassert itself in one other very real case, namely, Turing’s: man-made robots. (Extra-terrestrials simply fall under “other species.) For when it comes to reverse-engineering the mind, it is quite natural to ask under what conditions our uncertainty about whether robots feel becomes as small and negligible as our uncertainty about whether other people feel. And that necessarily raises questions about which level in the Turing hierarchy (verbal indistinguishability [T2], verbal and robotic indistinguishability [T3], or verbal and robotic and neurobehavioral indistinguishability [T4]) that convergence occurs at. For a feelingless T3 or T4 robot would indeed be a Zombie!
But, despite the high esteem in which philosophers of science hold counterfactual conditionals, I prefer future conditionals. Rather than committing myself to epistemic contingency plans on what might prove to be the equivalent of the possibility of squaring a circle, I’ll worry about crossing the zombie road only if and when we ever get to it, noting only that the explanatory gap remains the same: How and why do feeling entities feel?
One last point on this long digression into epistemic risk: The underdetermination of theories by data (i.e., uncertainty about whether your complete causal explanation is the *right* causal explanation) converges on the number of utopian theories (complete causal explanations) that can successfully explain the totality of the data, at the end of the day, if we come up with more than one. The usual stance is to trust that the constraint of having to scale up to accounting for all data will minimize this underdetermination, and hence minimize the number of viable theories left — and then to bet with Occam (on the theories with the fewest parameters), hoping that they are in some sense notational variants of one another.
Well, I hope it’s by now obvious that there’s a strict counterpart to all this in scaling up the Turing Hierarchy: The possibility of “zombies” is maximal at the “toy” level of modelling arbitrary toy fragments of our cognitive capacity. But as we scale up toward T3, the degrees of freedom shrink to something closer to the size of ordinary scientific underdetermination and empirical risk, and with T4, “zombies” are no longer worth giving another thought…
Last point, on Dave’s computationalism: I of course think it’s wrong. Cognition is not just computational, it is hybrid: Computational and dynamical (and no one yet knows the right blend, since we’re nowhere near T2, let alone T3 or T4).
So the answer to Richard’s version of Dave’s question “How could feeling be non-computational and yet also physical?” is that T3 and T4 robots (including ourselves!) are not just computational. But that does not help, because even though we know the generators of T3 power will be hybrid — dynamical + computational — the “hard” question remains unanswered: “How and why are those dynamic/computational states felt states?”
Counting properties, or “kinds” of properties (“property dualism”) is just numerology or taxonomy. It does not answer that “hard” question. To talk about “dual aspects” (physical + “qualitative,” i.e., felt) is likewise simply to restate the very same question, without making the slightest inroad on answering it…
RBROWN: “For my part, I think that as science approaches the limit we will be able to make deductions from facts specified solely in microphysical terms to facts about feeling (given that we have actually felt the things in question. That is, given that I have had a conscious experience of red I will be able to deduce from some physical description that it is like seeing red for the experiencer that has it) and this makes qualitative facts physical.”
Successfully predicting *that* we feel, or even successfully predicting *what* we feel from its neurophysical correlates is a goal worthy of pursuing, but it is just weather-forecasting. What we need is not just correlations and prediction, but causal explanation.
And no matter how well you can read my mind and describe and predict my feelings, you have not explained how and why I feel — until/unless you have explained how and why I feel.
RBROWN: “As Stevan says, we are feeling robots, but that we feel can be explained physically. That we can’t yet see how this can be done is not a very convincing argument.”
That we are feeling robots is indisputable. That how-and-why we feel can be explained physically is highly disputable. In fact, I am disputing it.
And the argument is not just that we haven’t done it yet, wait-and-see, but that there is already a systematic and inescapable way to defeat any attempted causal explanation, showing that whatever causal role is being attributed to feeling in any Turing-scale mechanism, the mechanism and performance will be unaltered if feeling is not attributed.
Feeling will remain a take-it-or-leave-it property in any Turing explanation, and that means the explanation will never be causal — *except* if psychokinesis were to turn out to be true, and feeling (doing things because you feel like it) were to turn out to be an extra, independent causal force in the universe.
But, not being a dualist, and being cognizant of the fact that all physical evidence (conservation laws as well as an unending series of failed “parapsychological” experiments) goes contrary to the psychokinetic hypothesis, I give it as little credence as I give to the possibility of zombies and other counterfactual fantasies…
Stuart W. Mirsky
Reply to Professor Harnad I was about to go off-line and noted that you’ve already offered a response to the latest I’d written above, so, being rather obsessive about these things I’ve delayed signing off. I’ll try to deal with your reply as briefly as I can: SMIRSKY: “[From the fact that passing T2 through computation alone] fails [to generate understanding] it doesn’t follow that this is generalizable to all systems of the same qualitative (computational) type.” You wrote: “It generalizes to all computational systems. “(I don’t know what you mean by computational systems of “qualitatively different type.” I only know one type of computation . . . .)” Response: Sorry, I did not intend to suggest a reference to different kinds of computational systems. When I wrote “all systems of the same qualitative (computational) type” the point was to use “computational” to clarify what I had in mind by “qualitative” since not all systems are computational in nature. You wrote: “What Searle shows is that cognition cannot be just computation. And that means any computation.” Response: I don’t believe Searle shows that at all. What he shows is that the fundamental constituent elements of any computational system are not capable of cognition (understanding on his usage) in themselves and that, when combined in the system he specs as the CR, they do not succeed in producing cognition either. He then goes on to construct an argument that purports to generalize from what the CR cannot do to what any other possible configuration of the same constituent elements can do. I think his argument fails to provide support for that generalization. SMIRSKY: “If understanding is a function of process complexity, as it well may be, then it would not be surprising that an underspecked system (lacking the scope and complexity needed) would lack it — or that a more fully specked system would not.” You wrote: “The reference to complexity is exceedingly vague. I know of no qualitative difference between “types” of computation based on the complexity of the computation. Computation is computation. And cognition is not just computation, no matter how complex the computation.” Response: The mistake you’re making apparently hinges on the supposition that I am differentiating between types of computation. I am not. The reason the reference to complexity is vague is because I did not want to go too far afield here and turn this into a debate about Searle and his CRA (or my own views), in deference to your interests in this discussion. I certainly can offer a lot more in terms of detail but perhaps the quickest way to see where I’m going is to refer to Daniel Dennett’s views since there is not much daylight between my view and his — at least on this question. SMIRSKY: “understanding may well be best understood as a system level feature (the result of the interplay of many processes, none of which are, themselves, instances of understanding)” Your wrote: “Perhaps, but this too is exceedingly vague. I can construe it as being somehow true of whatever it will prove to take in order to pass T3. But it certainly does not rescue T2, nor the thesis that cognition is just computation.” Response: The issue really is what is consciousness (or cognition or understanding or whatever aspect of this we want to settle on)? If we expect to find understanding somewhere in the CR’s particular constituent processes we must end up disappointed. But if we are expecting to find it at a system level, in the way those constituent processes interact in a particular set-up (say the infamous Chinese Gym or, as I would prefer to put it, a Chinese city where each room in each building on each street is a processor doing some particular thing but interfacing with and affecting many others) then we may certainly get luckier. Then we have a system that’s more brain-like in its complexity. And why should we expect something less from a computational platform than we get from a brain here? Of course, the computational platform consists of so many computational processors, each doing its part, rather like the neurons and neuronal clusters work to do theirs in brains. And the way they are operating is in terms of passing signals around according to their received design and instructions — like computer chips and processors do. I can provide even more detail as to the kind of complexity I have in mind but it’s probably best addressed outside this discussion so as not to take us too far off point. SMIRSKY: “it seems wrong.. that… meaning, is a matter of a referential relation between symbol and object in the world (finally achieved through the addition of a dynamical physical relation between the CR and the environment external to it).” You wrote: “It seems wrong because it *is* wrong! Connecting the internal symbols of a T3 system to their external referents gives the symbols grounding, not meaning. “Meaning is T3 grounding + what-it-feels-like-to-mean” Response: Okay. As I now understand you, your position is that to get meaning we need both grounding, as you have described it, and feeling (being the sense of knowing what we know). Here your account reaches the unexplainable part then, right? But then it doesn’t help much except to reaffirm a mystery. But what if a different account of meaning (as in it being the outcome of a complex layered and interactive system that resolves into associative picturing and linking) can also tell us how we get feelings? Why wouldn’t such an account, if it covers the bases, be preferable to one that leaves something (feeling) out? SMIRSKY: “there can finally be no difference between the [T3] robot’s… connections between symbol and object and that of the [T2 system]. In both cases, the referencing… involves representation to representation.” You wrote: “’Representation’ covers a multiple of sins! If you mean computation, see above. If you mean non-computational processes, all bets are off. Now the trick is to find out what those “representations” turn out to be, by reverse-engineering them, T3-scale…” Response: Yes, the term is intrinsically problematic. When we have a mental image it’s a representation of sorts and when we use a symbol, it, too, represents something (if it’s meaningful). When the brain passes information along through its complex neuronal network, we think of the signals, at least in some cases, as representations, too. So it’s not entirely clear that “representations” is the best word. But I haven’t yet come up with a better one for expressing this particular point. My point above, though, was to note that when we find meaning we do so through an associative process which links different images in our minds (again see my road sign anecdote). So the representations in that case were images, but I admit that I don’t know how raw inputted signals entering and traversing brains transform into mental images, or why these seem different in important ways from actual sensory images we have. But all of this is important in any effort to synthesize what brains do, either computationally or in any other way, I expect. SMIRSKY: “the external stimuli we can talk (and think) about are whatever it is we hold in in our heads at any given moment.” You wrote: “Yes, and then the question is: what has to be going on in our heads to give us the capacity to do that? The way to find out is to design a system that can pass T3. (We already know it can’t be done by computation alone.)” Response: I agree with much of this but not your last statement though perhaps we mean different things by that? We certainly can’t just throw computational processes together and expect to turn out a mind. They need to be combined in the right way, doing the right things, i.e., if mind features are system level rather than process level, then you need the right system, don’t you? If your dismissal refers to your notion that T-3 goes beyond mere computation, I cannot agree entirely. It certainly does add an aspect to the model which is not computation, as you note, and that aspect could be essential for building up the contents of a synthetic mind without relying on a dump process. But I see no reason, from your paper or in these discussions, to think that it would be essential to the brain-like processes that perform the operations which, in the aggregate, would have the features we recognize as understanding and so forth. SMIRSKY: “I… had images that made sense of the words… the meaning was not the referen[ts]… nor… my behavior…it was a certain critical mass of imagery … not… some physical grounding but… the web of mental images… and their relations to one another that constituted my understanding, not the grounding of the words.” Your wrote: “No doubt, but the question remains: what has to be going on in our heads to give us the capacity to do that? The way to find out is to design a system that can pass T3. (We already know it can’t be done by computation alone.)” Response: As I’ve already said, I don’t necessarily agree with your last point. It depends on what you mean by “computation alone”. I don’t think any AI researcher thinks that there aren’t ancillary and undergirding elements to any computational system. There is the platform, of course, and then the avenues for feeding in information which could be sensory and motor devices (if we want to give the entity a semblance of real world existence a la what we experience) or data dumps and/or information fed in piecemeal over time. I don’t think there is any reason to think that adding robotic capacity does anything more than provide an alternative way in which such an entity can obtain/develop necessary content through which it can make “sense” of its inputs. SMIRSKY: “There seems to be no reason… the images could not have found their way into a system like me in some other way… [e.g.] put there as part of a data dump.” You wrote: “I agree that a real-time prior history — although it is probably necessary in practice — is not necessary in principle in order to have full T3 power. (E.g., to recognize, manipulate, name, describe, think about and imagine apples, you need not have had a real-time history of causal contact with apples in order to have developed the requisite apple-detectors and apple-know-how. They could have been built in by the engineer that built you — even, per impossibile, the Blind Watchmaker. But it is not very likely. And, either way, whether inborn or learned, that T3 power will not be just computational. Hence building it in in advance would entail more than a data-dump.)” Response: I question your insistence on it not being “just computational”, see above. SMIRSKY: “[just as] meaning may be a function of complex associative operations within a given system utilizing retained and new inputs… [so] being aware of what one is doing on a mental level… is… a function of a sufficiently complex system that can and does collect, retain and connect multiple representations on many levels over a lifetime?” You wrote: “’Computational complexity’ is already insufficient to constitute sensorimotor grounding. It is a fortiori insufficient to explain feeling (hence meaning).” Response: Yes, you have said this before. But I haven’t seen any reason in these discussions to take that as more than an assertion of a position. And if sensorimotor grounding isn’t essential to meaning, contra your view but as I maintain, then there is no reason to expect or require that “computational complexity” be essential to “sensorimotor grounding”. Here we seem to have a strong divide between our views since I am unconvinced as to the essential role you claim for this dynamical relation with the environment (and the apparatuses that make that possible). But what COULD “explain feeling and hence meaning”, as you put it? Until now I have shied away from offering too many specifics here. But perhaps I can no longer do that? On the view I have advanced, awareness, which is my take on your term “feeling”, involves the interplay of subsystems, specifically including an entity that has a sense of being what it is, a subsystem (or subsystems) within the larger system dedicated to differentiating between internal and external inputs and classifying/grouping certain internal inputs in a way that forms a picture of the entity FOR the entity. This subsystem, with its pictures, interacts on an ongoing basis with the subsystems which picture the external environment in various dimensions and aspects. Once you have the self subsystem (which works rather like the others but just involves attending to different inputs), you get awareness because the self subsystem, in interacting with the other subsystems, manifests those interactions as what we recognise as awareness. You wrote: “But let me clarify one thing: when I substitute ‘feeling’ for all those other weasel words for ‘conscious’ and ‘consciousness’ — ‘intentionality,’ ‘subjectivity,’ ‘mental,’ etc., I am not talking particularly about the quality of the feeling (what it feels like), just the fact that it is felt (i.e., the fact that it feels like something). Response: Understood and, if you recall my earlier remarks, understood from the start. Recall my mention of flatness as the absence of strong emotional content. I don’t agree about calling them “weasel words” though. Our language is not obviously equipped to handle mental words so we constantly fall into ambiguity in employing them. It just seems to be a hazard of the business. You wrote: “As I think I noted in the target essay, I may feel a toothache even though I don’t have a tooth, and even though it is in reality referred pain from an eye infection. So it is not my awareness that my tooth is injured that is at issue (it may or may not be injured, and I may or may not have a tooth, or even a body!). What is at issue is that there is something it feels like to have a toothache. And I’m feeling something like that, whether or not I’m right about my having an injured tooth, or a tooth at all.” Response: Yes, understood and recalled from your essay. You wrote: “By exactly the same token, when I suggest that meaning = T3 capacity + what-it-feels-like to mean, what I mean is that I may be able to use a word in a way that is T2- and T3-indistinguishable from the way anyone else uses it, but, in addition, I (and presumably everyone else who knows the meaning of the word) also know what that word means, and that means we all have the feeling that we know what it means. I may be wrong (just as I was about my tooth); I (or maybe everyone) may be misusing the word, or may have wrong beliefs about its referent (as confirmed by T3). But if there is nothing it feels like to be the T3 system using that word, then the word merely has grounding, not meaning.” Response; Again understood and agreed. SMIRSKY: “there is no real explanatory gap to worry about once we recognize that the only point of explanation in this case is to explain what causes what. Why should an explanation that describes semantics as a function of a certain level of computational complexity not also suffice to account for the occurrences we think of as being aware of what is going on, of feeling what’s happening” You wrote: “Because computational complexity can neither produce grounding nor explain feeling.” Response: You have said this before and I have answered it. Computational complexity need not produce grounding since grounding is more likely, on my view anyway, to be an outcome of the process that imputes meaning than it is to be the underpinning of that imputation. As to explaining feeling, as noted above, I think it does, indeed, provide a basis for explaining feeling in this sense, i.e., sufficient complexity allows for a system that includes subsystems capable of interacting in certain ways — one of which involves having a sense of being a something and of being affected by those inputs which are associated with other subsystems (those that capture and manifest more external inputs). Well this has been rather long but, given the points you have made, I thought that this time I ought to try to address as much as what you have written as possible so there would be little chance of misunderstanding. But the length will probably work against that, this time.
CAUSAL EXPLANATION VS HERMENEUTIC INTERPRETATION
(Reply to Stuart W. Mirsky)
SMIRSKY: “[Searle] shows… that the fundamental constituent elements of any computational system are not capable of cognition”
I agree: Computation alone (any computation) is not enough to generate cognition. In other words, cognition is not all computation.
SMIRSKY: “[Searle] then… purports to generalize from what the CR cannot do to what any other possible configuration of the same constituent elements can do. I think his argument fails to provide support for that generalization.”
I agree: Having shown that cognition is not all computation, Searle thinks he has shown cognition is not computation at all. He definitely has not shown that. (The T3-passing system could be — and indeed probably is — partly computational.)
SMIRSKY: “If we expect to find understanding somewhere in [T2 computation's] particular constituent processes we must end up disappointed. But if we are expecting to find it at a system level, in the way those constituent processes interact in a particular setup (say the infamous Chinese Gym or, as I would prefer to put it, a Chinese city where each room in each building on each street is a processor doing some particular thing but interfacing with and affecting many others) then we may certainly get luckier.”
I am afraid these analogies don’t help me at all, in explaining (or understanding the explanation) of how and why we feel!
[The "Chinese Gym" argument (CGA), by the way, was a dreadful anticlimax to the Chinese Room argument (CRA). The CRA, against pure computation, passing T2, is perfectly valid, and Searle easily rebuts the "System Reply" (that not Searle, but the system as a whole, is the one that is doing the understanding) by memorizing the computer program and doing all the computations in his head: He becomes the whole system, and can thereby say, truly, without remainder, that he does not understand (and there's nothing and no one else). But with the CGA, which is a variant of the CRA, meant to show that even if the T2-passing system is a neural network rather than computation, it still does not understand. A neural net is a parallel, distributed network of interconnected units, passing activation to one another. So Searle asks us to imagine a bunch of boys in a gymnasium playing the role of the distributed units, passing papers. Searle waves his hands and says there's obviously no understanding going on there. But that's not at all obvious, because here the "System Reply" would apply: The boys don't understand, but "the system" does. And, unlike in the CRA, Searle cannot himself become the system -- if the parallelness and distributedness of the units are essential to the neural net's success in passing T2, because those are not implementation-independent, purely computational properties. But there is a simple solution to show that the parallelness and distributedness are not essential to the neural net's success in passing T2. Simply simulate the neural net computationally. That's a symbol system that Searle can memorize, and thereby become the system. And it produces exactly the same T2 I/O performance as the neural net. And Searle -- now again become the whole system -- does not understand. QED. But only with the proviso that the noncomputational aspects -- parallelness and distributedness -- are inessential to the T2 success, and we stick to T2. As soon as we move up to T3, sensorimotor transduction becomes essential, and neither the computational component, nor Searle's simulation of it, can be the whole system. Hence T3 and the hybrid system that can pass it are immune to Searle's CRA.]
Harnad, Stevan (1993) Grounding Symbols in the Analog World with Neural. Think 2(1) 12-78 (Special Issue on “Connectionism versus Symbolism” D.M.W. Powers & P.A. Flach, eds.). Reprinted in Psycoloquy 12(034-063) Nets.
But I’m afraid that your “Chinese city” analogy is as unavailing as Searle’s Chinese Gym, for much the same reason. The problem, as always, is of course the “other minds problem”: The only way to know for sure whether a system has a mind (feels, understands, whatever) is to *be* the system. And normally, the only system you can be is yourself. The only exception is T2, when passed by computation alone. This is what I’ve called “Searle’s periscope,” penetrating the normally impenetrable other-minds barrier. And the reason Searle’s perisicope works in this one special case, is that computation is implementation-independent. Therefore, if something is really a purely computational property, then any and every implementation of that same computation will have that same computational property. That’s how Searle can “be” the TT-passing computational system, and thereby report back to us that it does not give rise to the property of understanding!
But that fails for the Chinese Gym. And it fails also for the “Chinese City”: We know no more about whether a Chinese City does (or doesn’t) understand than we know about whether Gaia or the solar system does or does not understand.
Besides, even if a Chinese city or the solar system were to understand — or even, for that matter, if the brain (which really does understand) or a T3-passing robot (which probably does understand) — is like a Chinese city, we still haven’t a clue of a clue as to why and how it understands. Not, that is, if we don’t lose sight of the fact that understanding is not just cognitive capacity that is Turing-indistinguishable from that of someone who really understands, but it also *feels like something* to understand. In other words: Why on earth would it feel like something to be like a Chinese city?
SMIRSKY: “Then we have a system that’s more brain-like in its complexity. And why should we expect something less from a computational platform than we get from a brain here?”
Whether the complexity is in the performance capacity (T2, T3) or in the internal processes (whether dynamic or computational), there is not a clue of a clue as to why or how any of it is felt, rather than just “functed.”
SMIRSKY: “your position is that to get meaning we need both grounding… and feeling (being the sense of knowing what we know).”
Actually, if you give me a T3 robot, it’s enough if it feels anything at all; we don’t even need to fuss about *what* it feels…
SMIRSKY: “Here your account reaches the unexplainable part then, right? But then it doesn’t help much except to reaffirm a mystery.”
Correct; it reaffirms the mystery. And pinpoints the mystery’s locus: explaining how and why a T3 (or T4) robot feels. And pinpoints also why it cannot be explained: Because there is no more causal room in the (already successful) T3 or T4 explanation (as long as one does not allow psychokinesis, for which all empirical evidence to date is resoundingly negative).
SMIRSKY: “But what if a different account of meaning (as in it being the outcome of a complex layered and interactive system that resolves into associative picturing and linking) can also tell us how we get feelings? Why wouldn’t such an account, if it covers the bases, be preferable to one that leaves something (feeling) out?”
Because unfortunately that would not be an explanation of how and why we feel: it would just be a Just-So story! What we would need would be a causal explanation — not mentalistic hermeneutics on top of a complete causal explanation of cognitive performance capacity.
SMIRSKY: “when we find meaning we do so through an associative process which links different images in our minds…”
But how and why would “an associative process which links different images” be a *felt* process?
SMIRSKY: “if mind features are system level rather than process level, then you need the right system, don’t you?”
Yes. But the way you find out that you have the right system is to make sure it generates T3 capacity: Then how do you make sure it feels? And what is the causal role of the fact that it is felt rather than just functed (if it is indeed felt)?
SMIRSKY: “[T3] certainly does add an aspect… which is not computation… and… could be essential for building up the contents of a synthetic mind without relying on a dump process. But I see no reason… it would be essential to the brain-like processes that perform the operations which, in the aggregate, would have the features we recognize as understanding and so forth.”
T3 adds essentially capacities if are trying to reverse-engineer a causal model that has all of our cognitive capacities, Turing-indistinguishable from our own. (The real objective is not to synthesize, but to explain causally: the synthesis is just to test the causal powers of the explanation.)
If we insist on brain-like processes, we can scale up to T4. But neither that — nor taking an aggregate or “system” view of it all — explains why and how the system (whether T3 or T4) feels (if it does), rather than just “functs,” Turing-indistinguishably from a system that feels.
SMIRSKY: “I don’t think any AI researcher thinks that there aren’t ancillary and undergirding elements to any computational system. There is the platform, of course, and then the avenues for feeding in information which could be sensory and motor devices (if we want to give the entity a semblance of real world existence a la what we experience) or data dumps and/or information fed in piecemeal over time.”
This all assumes that the lion’s share of the work is computational, and that the dynamic part is trivial I/O. I think there’s no particular reason to believe that that’s so. It more or less insists on computationalism despite the contrary evidence, and trivializes sensorimotor capacity.
But never mind! Suppose it’s *true* that T3 and T4 can be passed by a system that is mostly computational: The explanatory gap (how and why does it feel) remains just as gapingly wide.
SMIRSKY: “On the view I have advanced, awareness, which is my take on your term “feeling”, involves the interplay of subsystems, specifically including an entity that has a sense of being what it is”
Is that a felt “sense of being what it is” or just a functed sense? If felt, then the question is already begged, by supposing a felt component without explaining how or why it is felt.
SMIRSKY: “a subsystem (or subsystems) within the larger system dedicated to differentiating between internal and external inputs and classifying/grouping certain internal inputs in a way that forms a picture of the entity FOR the entity.”
Differentiation, classification, and even “pictures” (meaning: analogs of sensory projections) are fine. But why and how are any of them *felt* rather than merely functed? (And in what sense are they more “for” the entity — which is presumably the system itself — than are any of its other adaptive T3 or T4 functions?)
Do you see how the mentalism creeps in without any independent causal explanation? That’s what makes this decorative hermeneutics (even if it’s true!), rather than causal, functional explanation (of feeling).
And that’s what makes the “hard” problem hard (indeed, by my lights, insoluble).
SMIRSKY: “This subsystem, with its pictures, interacts on an ongoing basis with the subsystems which picture the external environment in various dimensions and aspects. Once you have the self subsystem (which works rather like the others but just involves attending to different inputs), you get awareness because the self subsystem, in interacting with the other subsystems, manifests those interactions as what we recognise as awareness.”
I am lost. The problem is not the “self-subsystem,” nor any useful information to which it is privy. The problem is with the fact (if it’s a fact) that any of this is *felt*: How? Why? It sounds like all the functions you describe would do their job just as well — in fact, Turing-indistinguishably well — if they were all just executed (“functed”) rather than felt. So how and why are they felt? (This is why “awareness” is a weasel-word: it conflates accessing information with feeling something whilst so doing. I think Ned Block has garbled this even further, by suggesting that there are two kinds of consciousness, when what’s really happening is that we’re conflating two things — only one of them a matter of consciousness — in one: access and felt access. One mind/body problem is surely enough!)
SMIRSKY: “Computational complexity need not produce grounding since grounding is more likely, on my view anyway, to be an outcome of the process that imputes meaning than it is to be the underpinning of that imputation.”
I’m not sure what “imputes meaning” means. (For me, “P” means something to a person if that person can uses “P” the way someone who understand “P” uses “P” *and* it feels like something for that person to mean that “P”.)
I don’t really know what computational complexity is (except in the formal complexity-theoretic sense). Whatever computation it takes to pass T3, that’s the requisite complexity; and in virtue of the fact that it passes T3, it is grounded. Now the problem is explain how and why it is felt (if it is indeed felt).
SMIRSKY: “I think it does, indeed, provide a basis for explaining feeling in this sense, i.e., sufficient complexity allows for a system that includes subsystems capable of interacting in certain ways — one of which involves having a sense of being a something and of being affected by those inputs which are associated with other subsystems (those that capture and manifest more external inputs).”
I understand the potential functional utility of having a subsystem with such privileged informational and causal role — but I do not understand how or why its functions are felt — or, if you really mean it a a homunculus, how and why it feels.
Response to Stuart W. Mirsky’s post on 2/20 at 13:59 Amen to your remarks at the end about the terminological difficulties in discussions like this one! And I accept that my suggestion about how to regiment “robot” was ill considered. However, I do think that the fact that a robot could connect its words both to words in conversation and to things it encounters in the world would not entail that it had any consciousness whatsoever. Since I think that’s a possible entity, I think it’s ok to introduce a term for it. As an abbreviation for “feelingless robot” I’ll introduce “Frobot”. Ability to do well in Turing’s imitation game is an interesting property. What I think Searle’s “Minds, Brains and Programs” showed is that having *this* interesting property is not sufficient for having another interesting property, namely, ability to connect words to things (either in detection or action). For that, you need at least a Frobot – and I’ve been proposing that a Frobot is sufficient for this latter ability. Ability to be conscious is a third, distinguishable ability. The fact that Searle himself is unclear about what his argument in MBP actually showed should not lead us to fail to make this important distinction. Of course, there might be some good argument that shows that a fully capable Frobot (like my shopping assistant) would *have* to also be conscious. Such an argument would have to show that in aligning inputs from detection devices with the program that makes good word-word connections (and so provides success at the imitation game) and in aligning both of these with outputs to effector devices, we would thereby, necessarily, produce the causes of consciousness. I don’t think we have any idea of how such an argument would go. I’ve avoided “intelligence” and “understanding” here. I hope the result will be clarifying, and that I haven’t inadvertently introduced some further verbal difficulty.
Response to Stevan Harnad’s post on 2/20 at 15:29 Harnad: “*We* are feeling robots . . . .” Robinson: I’ve conceded (in my reply to Mirsky’s post of 2/20, 13:59) that my suggestion about the regimentation of “robot” was ill considered. And I’ve offered “Frobot” as an abbreviation for “Feelingless robot” (since I think we need *some* term for a device that has certain interesting abilities, but is not conscious). But I think that to say that *we* are robots is to simply make “robot” a useless word. One may hold that we are nothing but material things, but that doesn’t make us robots. In philosophy and literature, robots (whether Frobots or robots that also have consciousness) have been understood to be electromechanical devices. Harnad: “Searle’s basis for concluding that he didn’t understand Chinese would be the same as mine: When I hear Chinese (or see it written) I have no idea what it means. It feels like I’m listening to meaningless vocalizations . . . .” Robinson: The feeling of not understanding might be illusory. Of course, that would be very strange! But so are reactive dissociation in pain phenomena, being under the illusion that one’s leg does not belong to one, Capgras syndrome, etc. The way Searle knows he doesn’t understand Chinese is that if he thinks “Maybe if I write ‘I’d like a hamburger, they’ll send me one” nothing happens – he has no reason (and knows he has no reason) to make one mark rather than another on his paper, and if he were to just write something by guessing, it would not produce the sending in of a hamburger. Harnad: “. . . if Turing did just mean T2, and passing via computation alone, then Turing was mistaken . . . .” Robinson: But Turing can be read as not having been mistaken about his own project. Instead, he can be taken to have been interested in the possibility of machine intelligence, which he conceived of as the property needed to successfully carry on an unrestricted conversation. I don’t see the point of the reference to Gallup polls. The imitation game is not remotely like polling. Of course, you and I are both interested in two additional properties: groundedness of symbols, and consciousness. I just think that progress with these depends on recognizing their distinctness, and keeping their distinctness clear as we go. Harnad: “The shopping assistant robot is a toy; it already fails T2 and T3 . . . .” Robinson: I don’t see the reason for this claim. As to T2, it converses intelligently; as to T3, it applies the right words to objects in its sensors and to actions and events that take place within its detector fields; and it suits its actions to its words. You can call it a toy, but it’s a toy with two really interesting properties (i.e., conversational ability and correct word-world connections). The aim to build a robot like my shopping assistant does not analytically entail the aim to build a device that instantiates the causes of consciousness. I don’t see any reason to think that satisfying the first aim would automatically guarantee that we’d satisfy the second. But I agree with you that our discussion would be more interesting and valuable if we could identify what the causes of consciousness are. I’ve argued in the reference below that they’re to be found in properties of patterns of neural events. So, my view is that it’s likely that a device like my shopping assistant could be made *without* also instantiating in it anything that has patterns of activity like those in our neural events. That is, it could do well in conversation and suit its actions to its words, and yet not have the causes of consciousness and, therefore, not be conscious. Reference: W. S. Robinson (2004) _Understanding Phenomenal Consciousness_ (Cambridge: Cambridge University Press).
Response to Richard Brown’s post of 2/20 at 20:05 Brown: “(. . . given that I have had a conscious experience of red I will be able to deduce from some physical description that it is like seeing red for the experiencer that has it) and this makes qualitative facts physical. Robinson: What we might (I’d even say, probably will) be able to do in the future is to make inferences of the following kind.
a. The K-ness of a neural event entails the occurrence of a red experience.
b. Jones is having a K event. Therefore,
c. Jones is having a red experience. But now we’d have to support a. That, however, is one way of posing the Hard Problem.
Thanks for the reply Stevan!
Harnad: “That’s not a quote! “
yes it is! From the video, but I suppose you asked us not to quote that. Sorry. It just came to me as I was writing. I thought that part of your thesis was that there is no true cognition without feeling and so a T3 system that lacked feeling would not really be thinking. Is this not the case?
Harnad: “Counting properties, or “kinds” of properties (“property dualism”) is just numerology or taxonomy.”
I don’t feel this way, but given that you shouldn’t mind being a property dualist. If one starts from a certain position, viz that it is reasonable to expect macro level truths to be entailed by some relatively small set of micro level truths, then a property dualist is just one who thinks that there are properties that exist but that are not entailed by a vocabulary restricted to that of an idealized physics. It seems to me that you are a property dualist of this sort, or at least you shouldn’t be bothered if people say that you are.
Harnad: “Feeling will remain a take-it-or-leave-it property in any Turing explanation, and that means the explanation will never be causal — *except* if psychokinesis were to turn out to be true, and feeling (doing things because you feel like it) were to turn out to be an extra, independent causal force in the universe.”
These are the kinds of things that make it sound like you do endorse property dualism. these two suggestions correspond to epiphenomenalism and interactive (property) dualism. Note also that zombies and teh like are not thought of as real possibilities for our world but as a test case for whether or not you think phenomenal consciousness is entailed by a completed physics. But, I guess the overall point I was trying to make is that you seem to be dismissive of conceivability arguments but then you seem to give one in support of your claim that there will ultimately be an epistemic gap. It seems at least conceivable to me that, say, they higher-order thought theory of conscious turns out to be the right account of phenomenal consciousness. Suppose, just for teh sake of argument, that it is conceivably the correct view. Then we may reason as follows. Pain is the painful stuff, the painful stuff is some higher-order thought, these higher-order thoughts just are states of the dorsal lateral PFC. If this was true then we could conclude that the painful stuff just was the dlpfc states. Why should those states be the painful stuff? This question is answered by the higher-order theory. Conscious pains just are the ones which are painful for me to have. I have a conscious pain when I am aware of myself as being in pain. That is what explains why it feels like something for me. Now, you may not like this theory or its proposed explanation but it is certainly not an objection to it to say that “Feeling will remain a take-it-or-leave-it property in any Turing explanation, and that means the explanation will never be causal,” since, at least on an account like David Rosenthal’s, consciousness doesn’t have much causal impact in the first place. He does not think that consciousness is epiphenomenal, quite, but most of the functioning of my mental life continues on without consciousness. So on the higher-order thought theory on can take away the feeling and leave the system (*mostly*) undisturbed. Given this the kind of argument you produce doesn’t show that consciousness can’t be computational. Rather it shows that you do not find the theories which would allow the kind of explanation you ask for to be very effective. But what we need are independent arguments against those theories of consciousness.
I just saw William’s comment. I agree that what we need is the second kind of argument. That is exactly where a theory of consciousness, like the higher-order theory, would come in. For instance premise A might be defended with ‘the appropriate higher-order state’ substituted for ‘Kness’…
FEELINGS ARE SUPERFLUOUS IN ANY CAUSAL THEORY
(Reply to Richard Brown)
R.BROWN: “Yes… ‘Turing Machines can explain all that is explainable but do not explain everything there is’… is [a quote] from the video”
I knew that couldn’t possibly have been a quote, because I argue the opposite, so I went back and checked the video, and what I said was words more like the following: “Turing’s contribution was that once you can explain everything you can *do* with your cognitive capacity, you have explained all that is explainable, but you have not explained everything there is…” Big difference. It was about whatever system successfully passes the Turing Test — which is not necessarily just a Turing Machine (computer). So I was talking about causal explanation, not necessarily just computational explanation. (I do agree with the strong Church/Turing thesis, though, according to which a universal Turing Machine can simulate just about any physical system and just about every physical process.)
R.BROWN: “I thought that part of your thesis was that there is no true cognition without feeling and so a T3 system that lacked feeling would not really be thinking. Is this not the case?”
Yes.
R.BROWN: “a property dualist is just one who thinks that there are properties that exist but that are not entailed by a vocabulary restricted to that of an idealized physics. It seems to me that you are a property dualist of this sort, or at least you shouldn’t be bothered if people say that you are.”
I’m not bothered, but all I’m really saying is that we can’t give a causal explanation of how and why we feel…
R.BROWN: “Note also that zombies… are not thought of as real possibilities for our world but as a test case for whether or not you think phenomenal consciousness is entailed by a completed physics.”
I seem to manage to make my small point without needing ‘em, though…
R.BROWN: “that you seem to be dismissive of conceivability arguments but then you seem to give one in support of your claim that there will ultimately be an epistemic gap.”
Where’s my conceivability argument? I’m saying we have no causal explanation of how and how we feel (or anything does), and that we are not likely to get one, because (except if psychokinetic forces existed, which they do not) feelings cannot be assigned any independent causal power in any causal explanation. Hence the causal explanation will alway works just as well with them or without them. Hence attributing feelings to whatever system we successfully reverse-engineer (T3 or T4) will just be a hermeneutic exercise, not an explanatory one — *even if the interpretation is true*!
R.BROWN: “It seems at least conceivable to me that, say, the higher-order thought theory of conscious turns out to be the right account of phenomenal consciousness.”
Remember I’ve deliberately dropped all vague, redundant and equivocal talk of “intentionality,” “subjectivity,” “awareness,” “access consciousness,” “phenomenal consciousness” etc. etc. in favor of one simple word that covers (rather than conceals!) it all — everything we need to account for: feeling.
And it certainly does not seem to me that any “higher-order thought theory of feeling turns out to be the right account of feeling.” That’s just (to me) trading on the (usual) equivocation between accessing and feeling: Till further notice, all accessing is unfelt accessing — until the causal role of feeling itself has been independently explained.
R.BROWN: “Pain is the painful stuff, the painful stuff is some higher-order thought, these higher-order thoughts just are states of the dorsal lateral PFC. If this was true then we could conclude that the painful stuff just was the dlpfc states. Why should those states be the painful stuff? This question is answered by the higher-order theory. Conscious pains just are the ones which are painful for me to have. I have a conscious pain when I am aware of myself as being in pain. That is what explains why it feels like something for me."
All I see is DLPFC activity (and whatever its actual causal role is in generating T3 or T4 performance capacity), plus a claim that DLPFC activity is felt (pain). Now I want to know how and why DLPFC activity (and all its causal doings) is felt, rather than just done. It’s correlated with and predictive of pain reports? Fine. But correlation is not causation, and certainly does not answer our question of why pain needs to be felt rather than just “functed” (executed). It fits with a “theory of higher order thought”? Then I ask the same question of the higher-order theory of thought: why does pain need to be felt rather than just functed, to do its job? (And, I’d add, be careful not to be too fancy about your higher-order thought theory of pain, lest you price out the amphyoxus and invertebrates, who seem to have their lowly but perhaps equally aversive “ouch”‘s…)
R.BROWN: “Now, you may not like this theory or its proposed explanation but it is certainly not an objection to it to say that “Feeling will remain a take-it-or-leave-it property in any Turing explanation, and that means the explanation will never be causal,” since, at least on an account like David Rosenthal’s, consciousness [feeling!] doesn’t have much causal impact in the first place.”
That may well be, but that just makes it all the harder to give a causal explanation of it…
R.BROWN: “He does not think that consciousness [feeling] is epiphenomenal, quite, but most of the functioning of my mental [internal] life continues on without consciousness [feeling]. So on the higher-order thought theory one can take away the feeling and leave the system (*mostly*) undisturbed.”
I would say that was (*mostly*) bad news for the higher-order thought theory — at least as an explanation of feeling. (And I would say the news was all bad, because I am sure that if the higher-order thought theory commits itself to any realistic adaptive contingencies, it will turn out that the feeling will prove not just mostly but entirely superfluous.)
R.BROWN: “Given this the kind of argument you produce doesn’t show that consciousness can’t be computational.”
I argued that (1) cognition (e.g., understanding) cannot be just computational (T2 passed by a computer program), but that (2) even for hybrid computational/dynamic systems that can pass T3 and T4, we cannot explain how and why they feel.
R.BROWN: “Rather it shows that you do not find the theories which would allow the kind of explanation you ask for to be very effective. But what we need are independent arguments against those theories of consciousness [feeling].”
It is not just that I find all theories I’ve heard to date ineffective: I find them *ineffectual*, because (if they are Turing-like performance-generating theories at all, rather than just hermeneutic hand-waving), they try to attribute causal power where it adds nothing, and where the causation works equally well with or without it.
Bernie Ranson
Professor Harnad, you have said that: “[T]he point of Searle’s Chinese room argument, and of the symbol grounding problem, is that without sensorimotor grounding (“physical embedding”), computation is nothing but the ruleful manipulation of meaningless symbols. Therefore cognition (not just “reasoning”: *cognition*) is not just computation.” However, isn’t it the case that Searle’s argument was rather about understanding, which by your definition above is “cognitive capacity + feeling”? The point would then be that feeling (or consciousness) is not just computation. I believe Searle would go further than that. He would say that, unlike consciousness, computation is an observer-dependent phenomenon, not something that one could discover in nature. That anything of sufficient complexity can be designated, interpreted and used as a computer. I understand in this context that any computer can be represented by a Universal Turing Machine, which is a notional machine, a read-write head, a tape, a program of instructions. And you I think have said that there is only one kind of computation, and I take it that that is it? In that case, as I understand it, the implication of Searle’s argument is that feeling or consciousness is not computation at all. I would like to ask therefore how we are to interpret computation in your sentences: “The brain and body certainly are not a computer plus peripherals; there’s not just [implemented] computations going on in the brain, but a lot of analog dynamics too.” and “We are dynamical systems — probably hybrid analog/computational.” I have to go now but I would like to come back to say something about sensori-motor grounding and grounding generally.
MOST DYNAMICAL SYSTEMS ARE NOT COMPUTERS: NEITHER ARE WE
(Reply to Bernie Ranson)
B.RANSON: “isn’t it the case that Searle’s argument was rather about understanding, which by your definition above is 'cognitive capacity + feeling'?"
Cognition is cognitive capacity + feeling.
Understanding, intelligence, meaning — are all instances of cognition.
B.RANSON: “The point would then be that feeling (or consciousness) is not just computation.” No, the points are (at least) two:
(1) Cognition is not just computation. (You can pass T2 with computation alone, without understanding.) Moreover, T2 is ungrounded. Solution: scale up to T3 (sensorimotor robot).
(2) Whether for T2 or T3 or T4, and whether for computation or hybrid dynamics/computation, there is no explanation of how and why the system feels (if it does).
B.RANSON: “I believe Searle would go further than that. He would say that, unlike consciousness, computation is an observer-dependent phenomenon, not something that one could discover in nature.”
That may well be, but it’s neither here nor there for the two points above (1) and (2).
B.RANSON: “That anything of sufficient complexity can be designated, interpreted and used as a computer.”
That is, I think, demonstrably false, if it means anything at all. Moreover, it is irrelevant to (1) and (2) above, which is what my essay was about.
B.RANSON: “I understand in this context that any computer can be represented by a Universal Turing Machine, which is a notional machine, a read-write head, a tape, a program of instructions. And you I think have said that there is only one kind of computation, and I take it that that is it?”
There is only one formalization of what computation is: Turing’s, Church’s, Kleene’s and Goedel’s all turned out to be equivalent.
B.RANSON: “In that case, as I understand it, the implication of Searle’s argument is that feeling or consciousness is not computation at all.”
Searle’s argument was about whether cognition (for example, understanding) was computational. Searle thought he had shown cognition was not computational *at all*, but all he really showed was that cognition was not *all computational*.
B.RANSON: “I would like to ask therefore how we are to interpret computation in your sentences: “The brain and body certainly are not a computer plus peripherals; there’s not just [implemented] computations going on in the brain, but a lot of analog dynamics too.” and “We are dynamical systems — probably hybrid analog/computational.” I have to go now but I would like to come back to say something about sensori-motor grounding and grounding generally.”
Interpret it as pointing out that (contrary to Searle!) (a) not every physical (i.e., dynamical) system is a computer (although the strong Church/Turing thesis is correct, that just about every dynamical system can be formally *simulated* by a computer) and (b) we — as well as all other T3 and T4 robots — are not just computers either: We are hybrid systems, partly dynamic, partly computational, and both are essential for cognition.
Stuart W. Mirsky
Thanks for the thoughtful response Professor Harnad. I will offer just a little more to further clarify our respective positions: SMIRSKY: “[Searle] shows… that the fundamental constituent elements of any computational system are not capable of cognition” S.HARNAD: I agree: Computation alone (any computation) is not enough to generate cognition. In other words, cognition is not all computation. SWM: I don’t think the agreement is as complete as suggested. My position is that the CR shows that the constituent elements (the computational processes) of the CR are not, in themselves, capable of cognition and they are not in the limited arrangement characterized by the CR’s specs. But where you and I seem to differ is that you’re saying (if I’ve got you right) that what needs to be added is something that isn’t computation, whereas I am saying more computational processes doing more things in an interactive way could conceivably work if they are doing the right things and arranged in the right way. Now it can be argued (as some have) that adding parallelism to the configuration (as I think is presumed by this position if we are to adequately mimic brains) is to go beyond pure computation, but I think that’s an odd claim (and hope it’s not yours). After all, parallel processing involves the arrangement and linkage of lots of processors doing serial processing in the way computational processes happen. So yes, there are some hardware requirements here which aren’t needed in straight serial processing but they don’t change the qualitative nature of the system. ******* SMIRSKY: “If we expect to find understanding somewhere in [T2 computation's] particular constituent processes we must end up disappointed. But if we are expecting to find it at a system level, in the way those constituent processes interact in a particular setup (say the infamous Chinese Gym or, as I would prefer to put it, a Chinese city where each room in each building on each street is a processor doing some particular thing but interfacing with and affecting many others) then we may certainly get luckier.” S.Harnad: I am afraid these analogies don’t help me at all, in explaining (or understanding the explanation) of how and why we feel! SWM: The reason it helps me to see what apparently it does not help others (including yourself) to see has to do with the analysis of semantics and the role of complexity. If semantics (meaning) includes the sense of being aware (your “feeling”) then awareness can, I believe, be fully explained by a model that includes a world picturing “machine” (like brains) which develop, maintain and use multiple pictures or mappings based on different classes of inputs. As I’ve mentioned elsewhere, if you have a self (or selves) consisting of one or several categories of inputs and a world consisting of several other categories or inputs, and the self subsystem represents the full entity while the world subystems represent the external environment and the two systems relate to one another in a way that involves the full entity operating in the environment, then you have the basic ingredients for being aware. Actual awarenesses will vary depending on the particular elements included (is it a vision dominated system, an auditory dominated system, and so forth?). Even the occurrence of emotions and physical sensation is probably optional on this view. But the awareness, I am suggesting, could be sufficiently explained in terms of the make-up and interaction of these two classes of subsystem within the overarching system. I agree that we typically look into ourselves and seem to see something else, something mysterious and different from the rest of what we experience. But what we seem to see doesn’t have to be what is actually there. S.Harnad: [The "Chinese Gym" argument (CGA), by the way, was a dreadful anticlimax to the Chinese Room argument (CRA). The CRA, against pure computation, passing T2, is perfectly valid, SWM: I disagree. The system argument is compelling once we get past the confusion of thinking that its claim depends on the CR being taken to have understanding. Its force lies in the argument for a generic system approach, not in arguing for the CR system to be what it manifestly is not. The Chinese Gym/City (the connectionist model) simply says that to get minds like our own on a computational platform, you need a system that is sufficiently large and complex and can mimic brain operations. But this still does not go beyond computation per se. A fellow over in England, Peter Brawley, suggested the best name for this version of the System Reply to me when, in a discussion, he came up with the point that expecting the CR to understand is like building a bicycle and then wondering why it doesn't fly. Since then I have tended to refer to this version of the system reply as the Bicycle Reply (a variant which gets at the need to think in terms of systems and not the CR system in particular). But I agree that the computational processes themselves do not understand and cannot be expected to evidence understanding and that's because the feature(s) we're looking for is a system level feature, not one that we should expect to see occur on the level of the system's constituents. S.Harnad (continuing): and Searle easily rebuts the "System Reply" (that not Searle, but the system as a whole, is the one that is doing the understanding) by memorizing the computer program and doing all the computations in his head: He becomes the whole system, and can thereby say, truly, without remainder, that he does not understand (and there's nothing and no one else). SWM: The problem with that is that on this view he is still not the system. Where, in the original CR, he was a component in the system (one of its constituents), in his rejoinder to the System Reply, the system has become a component within him. Just as we don't know what all our own component organs and systems are doing in our bodies (including our brains), why should Searle's ignorance of Chinese while running a system in himself argue against that system's capacity for comprehension? S.Harnad: But with the CGA, which is a variant of the CRA, meant to show that even if the T2-passing system is a neural network rather than computation, it still does not understand. A neural net is a parallel, distributed network of interconnected units, passing activation to one another. So Searle asks us to imagine a bunch of boys in a gymnasium playing the role of the distributed units, passing papers. Searle waves his hands and says there's obviously no understanding going on there. But that's not at all obvious, because here the "System Reply" would apply: The boys don't understand, but "the system" does. SWM: Yes.
S. Harnad: And, unlike in the CRA, Searle cannot himself become the system -- if the parallelness and distributedness of the units are essential to the neural net's success in passing T2, because those are not implementation-independent, purely computational properties. SWM: But they don't change the qualitative nature of the processes which are the same as in the CR, only now more is going on. If the CRA isn't about what those processes can do in terms of producing understanding, it's pointless. S.Harnad: But there is a simple solution to show that the parallelness and distributedness are not essential to the neural net's success in passing T2. Simply simulate the neural net computationally. That's a symbol system that Searle can memorize, and thereby become the system. And it produces exactly the same T2 I/O performance as the neural net. And Searle -- now again become the whole system -- does not understand. QED. SWM: If Searle contains the system it doesn't follow that he must also be co-extensive with the system. S.HARNAD: But only with the proviso that the noncomputational aspects -- parallelness and distributedness -- are inessential to the T2 success, and we stick to T2. As soon as we move up to T3, sensorimotor transduction becomes essential, and neither the computational component, nor Searle's simulation of it, can be the whole system. Hence T3 and the hybrid system that can pass it are immune to Searle's CRA.] Harnad, Stevan (1993) Grounding Symbols in the Analog World with Neural. Think 2(1) 12-78 (Special Issue on “Connectionism versus Symbolism” D.M.W. Powers & P.A. Flach, eds.). Reprinted in Psycoloquy 12(034-063) Nets SWM: I don’t think you need to go that far to show that Searle’s response to the System Reply is inadequate. It’s enough to show that containing a system is not the same as being co-extensive with it. Moreover, I don’t see a need to go beyond computational processes running on computers to achieve synthetic understanding. Whatever may actually be needed in fact to achieve a real world operating model (such as embedding in the world) is ancillary to the question of what it takes do produce understanding in principle. S.Harnad: But I’m afraid that your “Chinese city” analogy is as unavailing as Searle’s Chinese Gym, for much the same reason. The problem, as always, is of course the “other minds problem”: The only way to know for sure whether a system has a mind (feels, understands, whatever) is to *be* the system. And normally, the only system you can be is yourself. SWM: This mixes up what it would take to do it vs. what it would take to test for it. Of course, as I think you mentioned elsewhere, we never know in the usual sense of “know” that others have minds like we do or that they have minds at all. I favor the Wittgensteinian solution to this “problem” but in a testing regimen, it certainly remains an issue. That’s overcome, however, when we realize that all we are required to do is to test in the way we operate in the real world and that’s to interact with the tested entity in terms which match its capabilities. If it has only “e-mail” capabilities, then one type of test obtains. If it can move about in the world, act, and so forth, then another. The test would need to be open ended, of course, as you say. But it would demand no more of the subject entity than we demand of every other person on this planet with whom we may happen to come into contact. S.Harnad: The only exception is T2, when passed by computation alone. This is what I’ve called “Searle’s periscope,” penetrating the normally impenetrable other-minds barrier. And the reason Searle’s perisicope works in this one special case, is that computation is implementation-independent. Therefore, if something is really a purely computational property, then any and every implementation of that same computation will have that same computational property. That’s how Searle can “be” the TT-passing computational system, and thereby report back to us that it does not give rise to the property of understanding! SWM: Except that Searle’s argument doesn’t show that a sufficiently complex arrangement of CR constituents can’t do what the CR can’t do, thus it does not show that a understanding can’t be a “purely computational proerty”. The problem is that now we have hit a certain circularity in our discussion. You say that Searle’s argument proves that understanding isn’t a computational property but doesn’t work against a certain kind of System Reply which goes beyond pure computation, while I say that it doesn’t prove that and that the System Reply, itself, doesn’t go beyond pure computation. Our problem is that we both likely agree that something is required in the way of hardware to support a System Reply of the Chinese City variety that goes beyond the CR model. I take it that you think that what is required includes the mechanics that enable a dynamical relation with the world. I certainly do not. But I do agree that there would be a need for additional components (enabling linkages between processors and so forth) in a system conceived on the Chinese City model. S.Harnad: But that fails for the Chinese Gym. And it fails also for the “Chinese City”: We know no more about whether a Chinese City does (or doesn’t) understand than we know about whether Gaia or the solar system does or does not understand. SWM: Of course the only “Chinese City” we have is the brain and we know it does understand, at least in some cases! S.Harnad: Besides, even if a Chinese city or the solar system were to understand — or even, for that matter, if the brain (which really does understand) or a T3-passing robot (which probably does understand) — is like a Chinese city, we still haven’t a clue of a clue as to why and how it understands. SWM: We do if the thesis I’ve proposed, that it’s a matter of a certain kind of complexity in arrangement is right. But how do we determine if it is right? On a purely philosophical level the issue only requires that we look for an explanation which best accords with other things we know and which adequately accounts for all the things that need to be accounted for. But that still doesn’t establish the truth of such a thesis. For THAT we need to implement and test in the real world. And such testing need only be designed to 1) accord with the subject system’s capabilities and 2) look for responses that are evidence of the presence of comprehension (or other features if we think these may/should also be present). S.Harnad: Not, that is, if we don’t lose sight of the fact that understanding is not just cognitive capacity that is Turing-indistinguishable from that of someone who really understands, but it also *feels like something* to understand. In other words: Why on earth would it feel like something to be like a Chinese city? SWM: Why on earth does it feel like something to be us? Do we feel like our brains? If the Chinese City is an analogue for a brain, then it may or may not be the entire entity, in which case what it feels like to be it is best asked of it. Of course, the “Chinese City” is just a metaphor for a model of what a brain is so there’s no reason to think it feels like anything at all because metaphors don’t do that. But genuine subjects would, of course. ******* SMIRSKY: “Then we have a system that’s more brain-like in its complexity. And why should we expect something less from a computational platform than we get from a brain here?” S.HARNAD: Whether the complexity is in the performance capacity (T2, T3) or in the internal processes (whether dynamic or computational), there is not a clue of a clue as to why or how any of it is felt, rather than just “functed.” SWM: I don’t quite know what that means. Any subjective entity with the capacity of responding to this question ought to be able to say this is what it’s like to be me under the right circumstances. If we’re hypothesizing such an entity, whatever its physical platform, why assume it would be unable to respond from the get-go? Doesn’t that prejudge the answer? I think the problem is that you are focused on the mysteriousness of being a subject. But that may only be an artifact of language, that we are equipped to speak about certain kinds of experiences but not others. If you ask me what is it like to be me, what kind of an answer would I give? I could say, well I see this or that, remember this or that, have this or that emotion at this moment, grew up here, went to school there, lived here, work there. You might then say, no, I meant to be you instead of a bat or a rock or your pc. How would I answer such a thing? Isn’t it enough that I CAN answer while the rock and pc can’t? The bat, presumably, has something like what I have, a subjective life, however different than mine, and in that sense I am prepared to grant that it, too, is aware at some level and can imagine what kind of life it might live, how it would experience the world. But then I can do that for a computer that has elements of subjectiveness in its behaviors as well, can’t I? ****** SMIRSKY: “your position is that to get meaning we need both grounding… and feeling (being the sense of knowing what we know).” S.HARNAD: Actually, if you give me a T3 robot, it’s enough if it feels anything at all; we don’t even need to fuss about *what* it feels… SWM: Right. Same for the non-robotic computer, no? SMIRSKY: “Here your account reaches the unexplainable part then, right? But then it doesn’t help much except to reaffirm a mystery.” S.HARNAD: Correct; it reaffirms the mystery. And pinpoints the mystery’s locus: explaining how and why a T3 (or T4) robot feels. And pinpoints also why it cannot be explained: Because there is no more causal room in the (already successful) T3 or T4 explanation (as long as one does not allow psychokinesis, for which all empirical evidence to date is resoundingly negative). SWM: But what if it can be explained (say as I have explained it)? Why affirm a mystery as a mystery without first ascertaining that it cannot be de-mystified? This, of course, brings us to the question of whether my approach works to explain the “mystery”. I am assuming that, since you have had a chance to read my explanations as to what I mean by “complexity” and how that would work to yield awareness, you don’t think it works. But then we would need to see if you have definitively undermined my claim or merely denied it in favor of yours. ******** SMIRSKY: “But what if a different account of meaning (as in it being the outcome of a complex layered and interactive system that resolves into associative picturing and linking) can also tell us how we get feelings? Why wouldn’t such an account, if it covers the bases, be preferable to one that leaves something (feeling) out?” S.HARNAD: Because unfortunately that would not be an explanation of how and why we feel: it would just be a Just-So story! What we would need would be a causal explanation — not mentalistic hermeneutics on top of a complete causal explanation of cognitive performance capacity. SWM: If by “causal” you mean an explanation of what brings what about, then an explanation in terms of how computer processes produce features that look like the features we have which constitute our mental lives, then why isn’t this causal in the way we need? If you mean some set of laws a la the laws of physics, this is not excluded though perhaps not essential since this isn’t physics but a different phenomenon which could, conceivably, require something parallel to the already know physical laws or something that expands them. At the very least, the fact that subjectness is a different aspect of our experience than the forces and matter we encounter in the external world in which we exist may just be enough to explain why the “causal” answers should not be expected to be found in
chemistry or physics or astronomy and so forth. As our friend Searle likes to put it, consciousness is biologically based. Maybe what’s needed is an understanding of the dynamics of this particular aspect of biology, i.e., what brains do to make minds. But then that’s the point of cognitive science. *********** SMIRSKY: “when we find meaning we do so through an associative process which links different images in our minds…” S.HARNAD: But how and why would “an associative process which links different images” be a *felt* process? SWM: The feltness is a function, on this view, of the interplay of layered subsystems in the overarching operating system of the brain. SMIRSKY: “if mind features are system level rather than process level, then you need the right system, don’t you?” S.HARNAD: Yes. But the way you find out that you have the right system is to make sure it generates T3 capacity: Then how do you make sure it feels? SWM: You ask the right questions and do the right observations. How do you know that I feel or I know that you do? S.HARNAD: And what is the causal role of the fact that it is felt rather than just functed (if it is indeed felt)? SWM: Multiple layered, interactive subsystems within the overarching system. ****** S.HARNAD: T3 adds essentially capacities if are trying to reverse-engineer a causal model that has all of our cognitive capacities, Turing-indistinguishable from our own. (The real objective is not to synthesize, but to explain causally: the synthesis is just to test the causal powers of the explanation.) If we insist on brain-like processes, we can scale up to T4. But neither that — nor taking an aggregate or “system” view of it all — explains why and how the system (whether T3 or T4) feels (if it does), rather than just “functs,” Turing-indistinguishably from a system that feels. SWM: Complexity = multiple subsystems interactively operating in a layered way within a larger system. ****** SMIRSKY: “I don’t think any AI researcher thinks that there aren’t ancillary and undergirding elements to any computational system. There is the platform, of course, and then the avenues for feeding in information which could be sensory and motor devices (if we want to give the entity a semblance of real world existence a la what we experience) or data dumps and/or information fed in piecemeal over time.” S.HARNAD: This all assumes that the lion’s share of the work is computational, and that the dynamic part is trivial I/O. I think there’s no particular reason to believe that that’s so. SWM: And no particular reason to believe you have to add dynamic interactivity with the world (even you agree a data dump and individual feeds as needed might be enough). Moreover, it looks like adding on dynamic connections with the world is superfluous. One should do it only if what one already has under consideration can’t work. But Searle’s argument doesn’t demonstrate that it can’t. S.HARNAD: It more or less insists on computationalism despite the contrary evidence, and trivializes sensorimotor capacity. SWM: If computationalism covers everything (as I argue — not insist — that it does), why add more to the explanation? S.HARNAD: But never mind! Suppose it’s *true* that T3 and T4 can be passed by a system that is mostly computational: The explanatory gap (how and why does it feel) remains just as gapingly wide. SWM: I don’t see any explanatory gap if the account of a certain kind of complex system covers everything else and I think it does. ******** SMIRSKY: “On the view I have advanced, awareness, which is my take on your term “feeling”, involves the interplay of subsystems, specifically including an entity that has a sense of being what it is” S.HARNAD: Is that a felt “sense of being what it is” or just a functed sense? If felt, then the question is already begged, by supposing a felt component without explaining how or why it is felt. SWM: Depends on what “felt” is. Let’s look inside ourselves. Discount emotional and sensory impressions right off because they’re not what you have said you’re talking about. What’s left is awareness of this or that. What does this awareness consist of? Well it looks like its a capacity in us to associate inputs with an array of stored representations/images. These are arranged in different groupings. Some involve our sense of our physical selves, some involve our individual histories (how we were raised, what experiences occurred to us, how we were schooled, what kinds of work we’ve done, etc.). Some involve the things we know about the world, both in our immediate locations, our general locations, the places in which we live, work, etc., and about the organisation and geographical information about our planet. The subsystems having to do with our persons connect in one way, those dealing with the world in another and all the groupings interact with one another. What you call feeling and I call awareness can be explained, looks like the occurrence of instances of recognizing an input in relation to these subsystems. Not only do we get understanding this way but along with it, when the self subsystems are pulled into the interaction, we get impacts on ourselves. And here is all you need to find to account for awareness. ****** SMIRSKY: “a subsystem (or subsystems) within the larger system dedicated to differentiating between internal and external inputs and classifying/grouping certain internal inputs in a way that forms a picture of the entity FOR the entity.” S.HARNAD: Differentiation, classification, and even “pictures” (meaning: analogs of sensory projections) are fine. But why and how are any of them *felt* rather than merely functed? (And in what sense are they more “for” the entity — which is presumably the system itself — than are any of its other adaptive T3 or T4 functions?) SWM: The entity has various pictures, many of which relate to a larger subsystem which it recognizes as itself while others relate to the subsystems that constitute its mapping of the world outside its “self”. S.HARNAD: Do you see how the mentalism creeps in without any independent causal explanation? That’s what makes this decorative hermeneutics (even if it’s true!), rather than causal, functional explanation (of feeling). SWM: That we have to refer to the mental when explaining the mental should not be surprising and is not creeping “mentalism”. If
the point of an explanation is to say how something happens, is brought about, is caused, then an account that explains how a sense of awareness occurs is what’s needed. Otherwise all we are doing is holding out for the mystery which feels nice (it makes us seem special in the universe) but is not necessarily the way things really are. S.HARNAD: And that’s what makes the “hard” problem hard (indeed, by my lights, insoluble). SWM: By my lights the “Hard Problem” is an illusion, a function of our desire not to reduce the phenomenon in question to physicalistic terms. ****** SMIRSKY: “This subsystem, with its pictures, interacts on an ongoing basis with the subsystems which picture the external environment in various dimensions and aspects. Once you have the self subsystem (which works rather like the others but just involves attending to different inputs), you get awareness because the self subsystem, in interacting with the other subsystems, manifests those interactions as what we recognise as awareness.” S.HARNAD: I am lost. The problem is not the “self-subsystem,” nor any useful information to which it is privy. The problem is with the fact (if it’s a fact) that any of this is *felt*: How? Why? SWM: The issue is what is it to be “felt”? What is feeling in this way? Above I have endeavored to explain it as the recognition of a relation between external non-self and internal self-related representations. Of course there is no phenomenon of relation but there is the recognition of how one thing affects another. Insofar as comprehension is seen to occur with the occurrence of various associations between inputs and stored representations, the feeling you are focused on occurs when CERTAIN relations (those between some of the external and some of the internal subsystems) occur. They are just THOSE associations. S.HARNAD: It sounds like all the functions you describe would do their job just as well — in fact, Turing-indistinguishably well — if they were all just executed (“functed”) rather than felt. So how and why are they felt? (This is why “awareness” is a weasel-word: it conflates accessing information with feeling something whilst so doing. SWM: It’s a word we have in ordinary language. But why should it be more weasely than “feeling” in this context? Both have other meanings associated with them. ***** SMIRSKY: “Computational complexity need not produce grounding since grounding is more likely, on my view anyway, to be an outcome of the process that imputes meaning than it is to be the underpinning of that imputation.” S.HARNAD: I’m not sure what “imputes meaning” means. SWM: If there is meaning and not everything has it and not all the time then somehow what has it must get it. If we agree that a thing has meaning when that meaning is assigned to it by a subject, then imputing is as good a term as any for this operation. S.HARNAD: (For me, “P” means something to a person if that person can uses “P” the way someone who understand “P” uses “P” *and* it feels like something for that person to mean that “P”.) SWM: So meaning is grounded + feeling like, and something means something when it is grounded and someone feels (what? recognition?) that it is so grounded? S.HARNAD: I don’t really know what computational complexity is (except in the formal complexity-theoretic sense). SWM: I’ve already described it. It’s more things going on in interactive ways. But only certain complex systems are doing it in the right way. S.HARNAD: Whatever computation it takes to pass T3, that’s the requisite complexity; and in virtue of the fact that it passes T3, it is grounded. Now the problem is explain how and why it is felt (if it is indeed felt). SWM: On my view, passing T-3 includes feltness but the testing of course must look for that as well. How? By formulating questions or challenges that require a sense of self (with the consequent awareness of differences and impacts on the self). **** SMIRSKY: “I think it does, indeed, provide a basis for explaining feeling in this sense, i.e., sufficient complexity allows for a system that includes subsystems capable of interacting in certain ways — one of which involves having a sense of being a something and of being affected by those inputs which are associated with other subsystems (those that capture and manifest more external inputs).” S.HARNAD: I understand the potential functional utility of having a subsystem with such privileged informational and causal role — but I do not understand how or why its functions are felt — or, if you really mean it a a homunculus, how and why it feels. SWM: As noted, this depends on what we mean by “felt”. I think that if we examine any instance of feltness in ourselves, we will see that there is no special feltness that is felt, that is occurring. It’s just a matter of relations between some aspect of the self-picture and aspects of the external pictures. What particular images/representations/sensations are kicked up in the course of the given relations may and likely will differ. The fact that some ARE kicked up leads us to think there is one special image/ representation/sensation there among all these others. But why should there be? It’s enough that some ARE kicked up in the course of these mental events. It’s the set of these relations with their ever varying content that constitutes the occurrence of feltness/awareness on this view. We don’t need to invoke the notion of mystery at all.
Stuart W. Mirsky
I put in a little time today watching the video presentation (new version) of this paper’s argument for a second time. Near the end Professor Harnad poses his challenge to his critics — provide an account of what he calls the “feeling” part of understanding. He grants up-front that we can explain, in fact or at least in principle, the causal factors underlying human and other thinking entities’ actions, up to and including how their brains do what they do (though full understanding of this aspect is still a ways off). But, he concludes, there is no hope for an explanation of the way something physical (whether brain or computer) produces that sense of awareness he calls “feeling”. (I’m presuming that if I’ve misstated this– not an impossibility — he’ll correct as needed.) We’ve seen in these exchanges that this aspect of his position continuously comes to the fore in his responses although I think he’s a little less clear on what he means by “grounding” (it’s not just connecting symbols by reference to physical elements in the world but also operating effectively in the world on those elements). But the crux of his argument seems to be that understanding (getting the meanings of words and other references) requires the occurrence of the right kind of responses through the entity’s engagement in the physical world PLUS having the feeling of doing or at least getting it (which is especially relevant in cases like my road trip story recounts). It’s the “feeling” issue that seems to be the critical one here. William Robinson has argued for the possibility of intelligent zombies qua robots (I think that’s a fair formulation, anyway) and Josh Stern, early in the discussion, has argued that there’s no reason to look for something extra called “feeling” because a comprehensively physical account will suffice. My own view, which has been stated here over the discussion’s course in increasing detail (but perhaps, as yet, without adequate clarity) is that the idea of Stevan Harnad’s “feeling”, which I prefer to call “awareness” (that “weasely” word!) is about right, i.e., that John Searle was correct in focusing our attention on the role of actually experiencing the understanding when we understand. As I have already indicated, however, I think Searle draws a mistaken conclusion from his CR thought experiment, because you cannot generalize from the failure of an underspecked system to the failure of all possible systems of the same type. But the Harnad challenge remains outstanding after all the back and forth here: How to account for the occurrence of the subjective aspect of understanding which seems to be outside the realm of objective observation? If we reject a view like Stern’s, that there is nothing left out in a causal account which ignores subjectiveness (note that Stern doesn’t, as far as I know, deny subjectiveness, only the need to account for it separately from the physical description of what brains and any equivalents do), then we do have to say how feelingness occurs as Harnad demands of us. I’d like to suggest at this point that my response, which invokes the complexity of a brain-like operating system to account for both meaning and the feeling of understanding that accompanies instances of human comprehension, does answer his challenge though I expect he will not. At the least the account I’ve offered is rather complex and abstract and so hard to get hold of conceptually. After all, how can we suppose that a computer, however massive and however sophisticated its operations, ever really can understand as we do? Aren’t all such entities going to end up as nothing more than Robinson’s robots? Frobots? And if they weren’t how, in Harnad’s way of looking this could we hope to tell?
So near the end of this discussion, what have we got from it? And has the Harnad challenge been met?
CAUSAL EXPLANATION
(Reply to last two postings by Stuart W. Mirsky)
I think we have probably reached an impasse, and our exchanges are getting rather too long and repetitive, so I won’t use quote/ commentary here, even though I prefer it.
I think we have one point of disagreement, and this will not be resolved by further iterations: You feel that a (“complex”) higher-order componential schema somehow explains how and why we feel, whereas I cannot see that at all: It seems to me just to be hermeneutics on top of an (unspecified) causal theory of function.
I think there is also a point of misunderstanding: You seem to think that the problem of feeling has something to do with cognizers telling us what things *feel like* (or telling us that they feel like something). But that’s not it at all. It’s about explaining (causally) how and why they feel anything at all (if they do).
Joshua Stern
SWM @ 2/21/2011 17:01 SWM: Josh Stern, early in the discussion, has argued that there’s no reason to look for something extra called “feeling” because a comprehensively physical account will suffice. JS: I hope to post again soon, decided that I should do a little actual composition offline, and am finding the time. Just a quick note here regarding the “feeling” issues, that apparently SH believes we can do without in the filling of some gap that “grounding” fulfills. I also hold that any “feelings” issue is at least methodologically separable if not moot, but the feelings (eg qualia) issue yet remains to be explained. But most of all I revert to Stuart’s note that we have to clarify whatever it is we are actually talking about, or saying. My major and no doubt most annoying thesis is that much of the traditional discussion of these topics suffers from “The myth of the given”, and tries to answer questions that are not coherent. Yes this is hardly new and very Douglas Adams (“42″), but hardly less valid for being a classic complaint. Disclaimer: Stuart and I are Internet friends of old, and share many sympathies on these issues. Hi, Stuart! And others here who I know from other Internet fora. More ASAP.
Stuart W. Mirsky
Reply to Stevan Harnad @ 2/21 18:06 PM Yes, I think you’re right. At some point discussions like this reach bedrock and I think ours is here. I have no problem seeing how a computer system (given adequacy of the platform to run the full panoply of interrelated programs) could, at least in theory, produce a thinking/feeling synthetic mind that’s not different, in terms of functionalities, from our own. On the other hand you seem to draw the line there. As I read it, the idea just makes no sense to you. I have discussed this with many who are in agreement with that view and with some who aren’t and it always seems to come down to that, a matter of seeing the possibility or not. Note that my view is not that a machine mind would necessarily be just like ours as there could be any number of variations in terms of differences in the medium, the platforms, capacities, etc. But I think it quite conceivable that a sufficiently complex/ sophisticated system of the type I’ve described could have the kind of feeling you talk about. On the “misunderstanding”: I don’t think you’re right about that. I have never said here that the capacity to report or describe particular feelings was the issue though I have occasionally taken some of your points to allude to that (as when you questioned the value of the Chinese City model on the grounds that we could never know how it felt to be a Chinese City). I take it from your response that you would not agree that I have offered a way that meets the challenge you laid out in your video. While I didn’t think you would (and ascribe that to this fundamental difference in how we imagine or conceptualize consciousness), I wonder what it would take, on your view, to meet that challenge. Perhaps my response doesn’t work, but then there must be some sort of account that might, on your view. I’d be interested to know what the parameters might be. To offer the challenge after all, you must believe there’s some way to meet it, at least in principle — even if no one could ever actually do so in fact. As you also note, our exchanges have become much too long so I will withold anymore extensive detail in any further comments I post unless specifically requested to provide that. Thanks for a good discussion and I would certainly like to learn more about what you would count, on your view, as explaining how “feeling” (as you have used the term) is caused.
HOW TO MEET THE CHALLENGE OF EXPLAINING HOW AND WHY WE FEEL
(Reply to Stuart W. Mirsky)
S.MIRSKY: “I wonder what it would take, on your view, to meet… the challenge you laid out in your video. Perhaps my response doesn’t work, but then there must be some sort of account that might, on your view. I’d be interested to know what the parameters might be…”
Happy to oblige (though please note that my own view happens to be that it is impossible to explain how and why we feel, for the reasons I have given, but am happy to repeat when asked).
To meet the challenge of giving a causal explanation of how and why we feel, you would either have to:
(1) show that psychokinesis (“mind over matter”) is possible after all, despite all the contrary evidence (i.e., show that feeling is a 5th causal force in the universe); or
(2) show that some of the functions in a successful T3 or T4 causal mechanism could not do what they do if they were unfelt functions rather than felt functions; or
(3) show that (for some reason that needs to be described and supported) there is no need to give a causal explanation of the fact that we feel.
But what will not work is hermeneutics, namely, *interpreting* some components or aspects of the function of a causal mechanism as being felt, and then dubbing that interpretation as an explanation.
Nor will “wait-and-see”: There are some challenges on the table that suggest the wait is destined to be unrewarded. I especially like challenges in category (2), because it is usually illuminating to show how the alleged need for and role of feeling in particulal cases always turns out to be defeasible.
But I’m happy to take on challenges in category (1) or (3) as well.
What I can’t do anything about is hermeneutic hypotheses, because, like all interpretations, they are irrefutable, being merely matters of taste. I can only repeat that the challenge concerns objective explanation — not subjective exegesis — of the irrefutable fact that we feel.
Hi Istvan, it’s good to see you here (and see you discuss consciousness with smart philosophers, as opposed to discussing Hungarian politics with dumb right-wingers…). And thanks for the great talk. Let me try to give a reductio argument against your conclusion that even after having explained all that can be explained about cognition, there still is the residue of what you call ‘feeling’. Suppose that there is Bence*** who passes the relevant Turing Tests (whichever you pick – T3, T4…). But Bence*** lacks any phenomenal character every day between 12pm and 1pm. At 11.59.59am, it’s all feelings, but at 12.00.01pm, he feels nothing (but he is still behaviorally and neurally indistinguishable from me). If your conclusion is correct, Bence*** must be possible. I’m trying to show that it is not. Suppose that I eat mango pickle for the first time in my life at 12.30pm one day. I would clearly feel something at that time. Bence*** also tastes mango pickle for the first time. I then keep on eating mango pickle until 1.10pm. So does Bence***. Now what would Bence*** feel at 1pm, that is, when his feelings come back? He, presumably, would feel something new – something he’d never felt before. If you ask him: “do you feel something new, something you’ve never felt before?”, he’ll say (if he’s honest): “Hell, yeah”. But I will not feel anything new – I’d been eating mango pickle for half an hour by then. So there’s a behavioral (and, presumably, a neural) difference, which contradicts the original supposition that there isn’t any behavioral or neural difference betweeen Bence and Bence*** (we’ve both passed the relevant Turing Tests). But then your conclusion leads to a contraditction: it must be false… Which step of this argument do you disagree with?
TURING INDISTINGUISHABILITY IS NOT ABOUT INDIVIDUAL IDENTITY
(Reply to Bence Nanay)
B.Nanay: “Hi Istvan, it’s good to see you here (and see you discuss consciousness with smart philosophers, as opposed to discussing Hungarian politics with dumb right-wingers…).”
Bence, Szia! (For those who don’t know what Bence was alluding to: something much more important than puzzles in the philosophy of mind is happening to smart left-liberal philosophers in Hungary: the “dumb right-wingers” are the increasingly partisan and authoritarian government of Hungary: See this link.)
B.Nanay: “Suppose that there is Bence*** who passes the relevant Turing Tests (whichever you pick – T3, T4…). But Bence*** lacks any phenomenal character every day between 12pm and 1pm. At 11.59.59am, it’s all feelings, but at 12.00.01pm, he feels nothing (but he is still behaviorally and neurally indistinguishable from me). If your conclusion is correct, Bence*** must be possible. I’m trying to show that it is not.”
(1) Presumably B* is an engineered robot.
(2) Let’s say B* is a T4 robot we’ve designed, and we know how he works because we designed him and understand the internal causality that generates his T4 success.
(3) We cannot, of course, know one way or the other whether B* feels (for the usual reasons).
(4) You are *stipulating* that he feels at some times and not at others.
(5) Presumably this is not something the designers have done (since they just worked to pass T4), so there’s no observable difference between the times B* is stipulated to feel and the times he is stipulated to not-feel.
Well then I suggest we are no longer talking about the causal basis of T4 performance; we are talking about what omniscient stipulations entail. As such, they have no bearing at all on the point I am making. What you get out of an omniscient stipulation is what you put into it…
B.Nanay: “Suppose that I eat mango pickle for the first time in my life at 12.30pm one day. I would clearly feel something at that time. Bence*** also tastes mango pickle for the first time. I then keep on eating mango pickle until 1.10pm. So does Bence***. Now what would Bence*** feel at 1pm, that is, when his feelings come back?”
Bence, I think you may be misunderstanding the Turing Test: B* is not your doppelganger or your bioengineered clone. He is just another generic entity capable of passing T4. If you happen to build him so that he is not only Turing-indistinguishable from a real person (which is all the TT calls for), but you also make him identical with you up to time T, then (just as would happen with your clone), he would begin to diverge from you as of time T.
There’s nothing relevant or new to be learnt from that. And it’s the same whether or not B* feels. (He would diverge behaviorally even if he did not feel.) And omnipotently turning the feeling on and off for an interval has nothing to do with anything either. B.Nanay: “He, presumably, would feel something new – something he’d never felt before. If you ask him: “do you feel something new, something you’ve never felt before?”, he’ll say (if he’s honest): “Hell, yeah”. But I will not feel anything new – I’d been eating mango pickle for half an hour by then. So there’s a behavioral (and, presumably, a neural) difference, which contradicts the original supposition that there isn’t any behavioral or neural difference betweeen Bence and Bence*** (we’ve both passed the relevant Turing Tests). But then your conclusion leads to a contradiction: it must be false… Which step of this argument do you disagree with?”
The very first step, where you misunderstand the Turing’s Test to be one of Turing-indistinguishability between two individuals, rather than what it really is meant to be, namely, Turing-indistinguishability in (generic) performance capacity from any real person…
Stuart W. Mirsky
Reply to Stevan Harnad’s post of 2/22/11 @ 9:39 AM Okay, thanks for responding. Let me ask this: Suppose the Star Trek character, Data, actually existed and behaved the way it’s portrayed in the show. Suppose the science were there to give us a genuine artificial mind of Data’s type. Now we know that, at least in the earlier stories, Data hasn’t got emotions (though the acting and storyline don’t always hew to that line because we often see Data with a puzzled look and that is a sort of emotion, isn’t it?). At the least we know that Data has awareness and it’s an awareness that goes beyond the responsiveness of thermostats. Now perhaps Data is just a zombie intelligence as William Robinson has proposed is possible (and why shouldn’t it be possible in this scenario, too?). But then, to what extent should we expect the other characters to treat Data like a toaster or other zombie machines instead of as a fellow person? And would we treat Data thus, if we were among those other characters, or if a Data was in our presence? (This is not about whether we would be justified in treating him thus but whether it would make sense to treat this particular entity in that way.) By stipulation, Data passes the open ended lifetime Turing test. Moreover Data not only acts and speaks intelligently, Data also reports experiences. We are led to believe, by these behaviors, that what Data sees, even if it doesn’t look quite like what we see (perhaps it’s more like what the blind character, Geordie, sees with his visor), is still somewhat like what we are seeing. However it looks to Data, it meshes sufficiently with the things the other, normal characters are seeing to be taken as real sight, too. The fact that we grant that a Data, behaving in this way, is seeing anything at all means we recognize some degree of perception in Data, no? And the fact that he can report on what he sees and can operate in terms of it, etc., suggests that he knows what he is seeing and understands something about it. But here’s the crux: If he has understanding, then he must have feeling on your view. If, however, we deny that the understanding behavior he manifests includes feeling (as you are using that term), for lack of direct access to it as we have it in ourselves, then we must also deny that he understands, despite all the behavioral evidence to the contrary. Such a Data is really just a very clever, automaton, an autonomous puppet, an intelligent zombie. But this flies in the face of all the behavioral evidence (as it would if we were applying the same standards to other human beings). So do you think there must be a test for Data’s feelingness beyond the ongoing Turing-type test of real life itself? Has something been left out in this account that would need to be included for us to be sure that we have, in Data, a real understanding entity? If not, then what you are calling mere “hermenutic hypothesis”, “irrefutable” because it is just “a matter of taste”, looks wrong, wouldn’t you agree?
EXPLAINING DATA
(Reply to Stuart W. Mirsky)
S.MIRSKY: “Suppose the Star Trek character, Data, actually existed and behaved the way it’s portrayed in the show… Now we know… Data hasn’t got emotions. At the least we know that Data has awareness… that goes beyond the responsiveness of thermostats.”
Data is a fiction, and one can suppose or stipulate whatever one likes about a fiction.
If there were really a robot that could do what Data can do, then it would pass T3.
You used one of the weasel-words in your description: “awareness.” Can we please go back to the only word that doesn’t equivocate between access to information and sentience (feeling)?
Either data feels or he doesn’t feel. That’s all there is to it; and there’s no way we can know one way or the other (because of the other-minds barrier), no matter what Data tells us (T2) or does (T3).
If Data feels anything at all, then he “goes beyond [mere] responsiveness.” If not, not. (All of T3 could be “mere responsiveness.”) But of course it has always been pop nonsense to depict (the fictional) Data as having no “emotion”: What on earth does that mean? (I’ve met a lot of people with apathy or “blunted affect” — and I’ve met a few sociopaths too. Their emotional make-up is different from mine. But presumably they still feel pain, and whatever the right word to describe their affective state, I assume they would not take to being held under water with great equanimity.
So, personality aside, it’s not really about *what* the T3 robot feels, but about whether it feels at all (and if so, how and why). It not only feels like something to be angry or afraid; it also feels like something to be in pain, to be overheated, to smell incense, to hear church-bells, to see a red triangle, to touch a rough surface, to raise your arm, to say (and mean) “the cat is on the mat,” and to understand Chinese. Take your pick. If T3 has any of them, T3 feels. If not, not.
Harnad, S. (2001) Spielberg’s AI: Another Cuddly No-Brainer.
S.MIRSKY: “[P]erhaps Data is just a zombie intelligence… why shouldn’t it be possible…?)”
A rather awkward way of saying that “If not, not.”
Yes, either outcome looks possible, for all we know. Or maybe it’s not possible to pass T3 without feeling — but if not, then how and why not?
Back to square one.
S.MIRSKY: “[W]ould we treat Data… like a toaster or other zombie machines instead of as a fellow person? (not about whether we would be justified… but whether it would make sense…)”
Well the moral question would certainly trouble a vegan like me: I definitely would not eat a T3 robot.
Strauss, S. (1990) Is it an ant? A cockroach? Or Simply a Squiggle? Toronto Globe & Mail
Would it make *practical* sense to treat a human-built, metal entity differently from any of the rest of us if it was otherwise Turing-indistinguishable from us? Only if you could get away with it, I suppose, as with all other forms of racism. Not if there were enough of them to defend themselves, just as any of us would.
Would it make “logical” sense to treat them otherwise? I’d say holding out for T4 — before being ready to give one’s logical assent to the fact that we can be no more or less certain that T3 robots feel than that any one of us feels — would be a rather subjective, ad-hoc stance rather than something dictated by logic.
S.MIRSKY: “that we grant that a Data, behaving in this way, is seeing anything at all means we recognize some degree of perception in Data, no?”
Actually, we are not in a position to grant or not grant anything since we have no idea whether or not Data feels when he is successfully displaying opto-motor performance capacities Turing-indistinguishable from our own. All we know is that he can do it, not whether it feels like anything to be able to do it.
S.MIRSKY: “And the fact that he can report on what he sees and can operate in terms of it, etc., suggests that he knows what he is seeing and understands something about it.”
If seeing that someone says — and acts as if — he feels something were as good as seeing that he feels something, T3 would already guarantee feeling by definition (possibly even T2 would).
But all T3 reverse-engineering does is generate someone about whom you have no better or worse reason to doubt that he really means what he says when he says he feels something than you have for doubting it about anyone else.
The fact that he’s made out of the wrong stuff? Well, move up to T4 if it really means that much to you. I confess that scaling up to T4 only interests me if there’s some T3 capacity that we are unable to generate, and T4 gives us a clue of how to do it. (Nothing like that has happened yet, but maybe something we learn from neuroscience will help roboticists, rather than the other way round.)
As for me, Data would be enough — not just to prevent me from eating him or beating him, but for according him the full rights and respect we owe to all feeling creatures (even the ones with blunted affect, and made of metal).
S.MIRSKY: “But here’s the crux: If he has understanding, then he must have feeling on your view.”
Indeed. But for me T3 is the best we can do for inferring that he has understanding, hence feeling. Trouble is, we can’t explain how or why he has feeling…
(You seem to think that if he can report Turing-indistinguishably on what he feels, then he must have understanding. Seems to me he either does or doesn’t; he’s just Turing-indistinguishable from someone who does. But that’s good enough for me — since it’s not really possible to ask for more. T4 is just for pedants and obsessive-compulsives… T3′s already done the job.)
S.MIRSKY: “If… we deny that the understanding behavior he manifests includes feeling… then we must also deny that he understands.”
We can neither affirm nor deny that he has feelings. And since “understands” (like any other cognitive state) means “T3indistinguishable cognitive capacity + feeling”, we can neither affirm nor deny that he understands either, just that he acts exactly as if he understands, just as any other person who understands does.
S.MIRSKY: “So do you think there must be a test for Data’s feelingness beyond the ongoing Turing-type test of real life itself?”
No. The other-minds barrier is impenetrable (except in the special case of the hypothesis that T2-passing computation alone can generate understanding, which “Searle’s periscope” shows to be false). And brain scans won’t penetrate the barrier either; all they can do is correlate and predict.
T3 is the best we can do.
S.MIRSKY: “Has something been left out in this account that would need to be included for us to be sure that we have, in Data, a real understanding entity?”
Some people think you need to scale up to T4, but I think that’s just superstition.
But since the name of the game is not mind-reading but explaining cognition, what has been left out in the Turing approach is feeling — not just how to test whether it’s there, but to explain how and why it’s there.
S.MIRSKY: “If not, then what you are calling mere “hermenutic hypothesis”, “irrefutable” because it is just “a matter of taste”, looks wrong, wouldn’t you agree?”
Hypotheses non fingo. I am not purporting to answer the how/why question about feeling. It is the hypotheses that purport to do so that are hermeneutical. If I say “feeling is explained by the fact that you have a subsytem with privileged access to analog images etc. etc.” how can anyone ever refute that? It’s just a decorative description you either like or don’t like.
Istvan, thanks. I do indeed talk about Turing indistinguishability, because I took you to deny it at the very beginning of your talk (the video, not the paper), when you say that your claim is that appealing to “what we can do with our cognitive capacities can explain everything that can be explained, but it will not explain everything”. Now you ask, rightly, to use the written text, so using the slogans in the written test, this amounts to saying that what “cognition does (or, more accurately, [what] cognition is capable of doing)” (p. 1) will not explain everything because it will leave out feeing. Now, it does follow from this claim that two individuals’ being exactly identical in terms of what their ‘cognition does/is capable of doing’ does not entail their being identical when it comes to feelings. In other words, it does seem to follow from your claim that Turing-indistinguishability (be it T3 or T4) does not imply ‘feeling’-indistinguisability.
And I was arguing against this last claim. If you don’t endorse this claim, then I was misreading you (or, rather, I was reading too much into what you said in the video).
February 23, 2011 at 09:17 GENERIC DOING CAPACITY VS. INDIVIDUAL FEELING STATES
(Reply to Bence Nanay)
B.NANAY: “It does seem to follow from your claim that Turing-indistinguishability (be it T3 or T4) does not imply ‘feeling’indistinguisability…”
Turing-indistinguishability means indistinguishability in (generic) performance capacity, not indistinguishability in individual state. Indistinguishability in performance capacity does not even imply feeling, let alone indistinguishability in individual feeling state. My paper is not on the ontic aspects of the mind/body problem, just the epistemic ones: Explaining how and why we can do what we can do does not explain how and why it feels like something to be able to do what we can do.
Stuart W. Mirsky
Thanks for the response Professor Harnad. I’ll try to be brief in my responses in deference to your earlier concerns. But I’ll continue to use the interspersed text approach since I also want to be sure I don’t misrepresent any of your statements. (I will cut where I can though): . . . SH wrote: Data is a fiction, and one can suppose or stipulate whatever one likes about a fiction. SWM: Yes, of course. So is the Chinese Room and your T-3 and T-4 entities. That’s the point. We are constructing hypothetical though at least possible scenarios in order to explore the concepts we apply to them. SH: If there were really a robot that could do what Data can do, then it would pass T3. SWM: Yes. SH: You used one of the weasel-words in your description: “awareness.” Can we please go back to the only word that doesn’t equivocate between access to information and sentience (feeling)? SWM: I thought I had made clear that I equate your “feeling” with “awareness”. However, I am no more comfortable with “feeling” than you seem to be with “awareness” but I am quite prepared to stipulate that by “awareness” I mean just what you have said you mean by “feeling”. However, perhaps “sentience” would be a good compromise? SH: Either data feels or he doesn’t feel. That’s all there is to it; and there’s no way we can know one way or the other (because of the other-minds barrier), no matter what Data tells us (T2) or does (T3). SWM: You pointed out above that the solution I offered to your challenge would not pass muster with you because it is just “hermeneutics”, as you put it, and a matter of subjective interpretation (not objectively observable). My response is to suggest that, since Data’s (observable) behavior meets all the usual criteria for understanding as seen in others, either that IS enough to sustain a judgment that a Data programmed with the kind of system I’ve presented has understanding OR your decision to equate understanding with “grounding + feeling” must be mistaken. It hinges, of course, on the fact that observable criteria are all we ever have for judging the presence of “feeling” in others. SH: If Data feels anything at all, then he “goes beyond [mere] responsiveness.” If not, not. (All of T3 could be “mere responsiveness.”) SWM: My point, of course, is that the observable evidence attests to the fact that he “goes beyond mere responsiveness.” SH: But of course it has always been pop nonsense to depict (the fictional) Data as having no “emotion”: What on earth does that mean? (I’ve met a lot of people with apathy or “blunted affect” — and I’ve met a few sociopaths too. Their emotional make-up is different from mine. But presumably they still feel pain, and whatever the right word to describe their affective state, I assume they would not take to being held under water with great equanimity. SWM: Yes, but this IS an exercise in hypotheticals. And it seems perfectly feasible that one of William Robinson’s intelligent robot zombies could do a lot of what Data does convincingly (though I doubt it could do it all). On the other hand, you’re arguing that there’s a missing link that must be present in any real instance of understanding but which will forever be excluded from discovery in others. If that’s so, then the Other Minds problem is resurrected though with no more impact on the actual research project of building conscious machines than it has on our daily contact with other human beings like ourselves. So your challenge will be unmet because you stipulate it so. SH: So, personality aside, it’s not really about *what* the T3 robot feels, but about whether it feels at all (and if so, how and why). SWM: Yes, and the Data of this scenario gives every indication of having feels — just as we do. SH: It not only feels like something to be angry or afraid; it also feels like something to be in pain, to be overheated, to smell incense, to hear church-bells, to see a red triangle, to touch a rough surface, to raise your arm, to say (and mean) “the cat is on the mat,” and to understand Chinese. Take your pick. If T3 has any of them, T3 feels. If not, not. Harnad, S. (2001) Spielberg’s AI: Another Cuddly No-Brainer. SWM: Yes, but you argued that my explanation of how feeling arises in computational terms fails because it’s based on a subjective interpretation of the available information. My response is that it’s no more subjective than our judgment of whether other people, like ourselves, feel. If understanding requires feeling, as you put it, then every judgment we make that anyone else but ourselves has understanding hinges on the same kind of assessment, i.e., that feeling is present. So unless you want to say there’s nothing objective about that kind of assessment, then why worry about whether a Data like entity has it if it fits the same bill? S.MIRSKY: “[P]erhaps Data is just a zombie intelligence… why shouldn’t it be possible…?)” SH: A rather awkward way of saying that “If not, not.” Yes, either outcome looks possible, for all we know. Or maybe it’s not possible to pass T3 without feeling — but if not, then how and why not? Back to square one. SWM: The issue, I think, is whether anything could meet the challenge you’ve presented and, as of now, I am inclined to think that the only reason it cannot is a stipulated one. But then it wouldn’t be a fair challenge. S.MIRSKY: “[W]ould we treat Data… like a toaster or other zombie machines instead of as a fellow person? (not about whether we would be justified… but whether it would make sense…)” Well the moral question would certainly trouble a vegan like me: I definitely would not eat a T3 robot. Strauss, S. (1990) Is it an ant? A cockroach? Or Simply a Squiggle? Toronto Globe & Mail
SWM: I excluded the moral issue above by differentiating between whether we are justified or whether it makes sense. SH: . . . Would it make “logical” sense to treat them otherwise? I’d say holding out for T4 — before being ready to give one’s logical
assent to the fact that we can be no more or less certain that T3 robots feel than that any one of us feels — would be a rather subjective, ad-hoc stance rather than something dictated by logic. SWM: If we assume that Data has a “positronic brain” that replicates the functionalities of a human brain (as the narrative demands), then we have a case of T-4 being passed, too. But that still leaves the problem that you insist there is something else, not accessible to us, which must be there and which, if it isn’t, undermines the validity of a claim that Data has understanding. Doesn’t your view essentially reduce to the Other Minds problem and, if it does, can there be real implications for the science of cognition? S.MIRSKY: “that we grant that a Data, behaving in this way, is seeing anything at all means we recognize some degree of perception in Data, no?” SH: Actually, we are not in a position to grant or not grant anything since we have no idea whether or not Data feels when he is successfully displaying opto-motor performance capacities Turing-indistinguishable from our own. All we know is that he can do it, not whether it feels like anything to be able to do it. SWM: But isn’t that to beg the question since the issue is whether the behavioral criteria provide us enough information so that we can have such an idea? S.MIRSKY: “And the fact that he can report on what he sees and can operate in terms of it, etc., suggests that he knows what he is seeing and understands something about it.” SH: If seeing that someone says — and acts as if — he feels something were as good as seeing that he feels something, T3 would already guarantee feeling by definition (possibly even T2 would). SWM: Yes. Why doesn’t it, other than the fact that you demand a test for computational devices (whether pure or hybrid) that you don’t demand for other beings like ourselves? SH: But all T3 reverse-engineering does is generate someone about whom you have no better or worse reason to doubt that he really means what he says when he says he feels something than you have for doubting it about anyone else. SWM: Yes. And why isn’t that enough, unless this is just the Other Minds problem imported into cognitive science? (As I’ve previously noted, I think Wittgenstein’s solution in the Investigations effectively undermines the issue of Other Minds as a real problem.) SH: The fact that he’s made out of the wrong stuff? Well, move up to T4 if it really means that much to you. I confess that scaling up to T4 only interests me if there’s some T3 capacity that we are unable to generate, and T4 gives us a clue of how to do it. (Nothing like that has happened yet, but maybe something we learn from neuroscience will help roboticists, rather than the other way round.) SWM: I think we effectively have but that even at T-3 we have enough. If a chair suddenly started behaving consciously (and we successfully discounted all other possibilities) then we would have no choice but to deal with it on those terms — and very likely change our ideas about what it takes to be conscious (brains no longer needed, etc.). SH: As for me, Data would be enough — not just to prevent me from eating him or beating him, but for according him the full rights and respect we owe to all feeling creatures (even the ones with blunted affect, and made of metal). SWM: So in terms of your own interactions you’re comfortable treating Data as a sentient entity but still want to hold out for something more in the field of cognitive science? Why? What can possibly be gained? S.MIRSKY: “But here’s the crux: If he has understanding, then he must have feeling on your view.” SH: Indeed. But for me T3 is the best we can do for inferring that he has understanding, hence feeling. Trouble is, we can’t explain how or why he has feeling… SWM: We can if a detailed analysis determines that it is this program that does these things run on his “positronic brain” which result in his behavior (including behavior that attests to the presence of feeling). SH: (You seem to think that if he can report Turing-indistinguishably on what he feels, then he must have understanding. Seems to me he either does or doesn’t; he’s just Turing-indistinguishable from someone who does. But that’s good enough for me — since it’s not really possible to ask for more. T4 is just for pedants and obsessive-compulsives… T3′s already done the job.) SWM: If it’s good enough in daily life, why shouldn’t it be good enough in terms of cognitive science? S.MIRSKY: “If… we deny that the understanding behavior he manifests includes feeling… then we must also deny that he understands.” SH: We can neither affirm nor deny that he has feelings. SWM: That, I take to be a mistake. You have already said that you would treat him as if he did, i.e., no differently than others like ourselves. That IS an affirmation he has feeling. SH: And since “understands” (like any other cognitive state) means “T3-indistinguishable cognitive capacity + feeling”, we can neither affirm nor deny that he understands either, just that he acts exactly as if he understands, just as any other person who understands does. SWM: Same response here. S.MIRSKY: “So do you think there must be a test for Data’s feelingness beyond the ongoing Turing-type test of real life itself?” SH: No. The other-minds barrier is impenetrable (except in the special case of the hypothesis that T2-passing computation alone can generate understanding, which “Searle’s periscope” shows to be false). And brain scans won’t penetrate the barrier either; all they can do is correlate and predict. T3 is the best we can do. SWM: Whether it is or isn’t, it is good enough because it’s no different than how we deal with others like ourselves. And that’s all that’s required to have objective observation of the presence of feeling in an entity. S.MIRSKY: “Has something been left out in this account that would need to be included for us to be sure that we have, in Data, a real understanding entity?” SH: Some people think you need to scale up to T4, but I think that’s just superstition. But since the name of the game is not mind-reading but explaining cognition, what has been left out in the Turing approach is feeling — not just how to test whether it’s there, but to explain how and why it’s there. SWM: I agree with you that we need to account for the presence of what you call feeling and I call awareness, but I disagree with your conclusion that we cannot discern its presence nor account for it. In fact I think the evidence is clear that we can do both. S.MIRSKY: “If not, then what you are calling mere “hermeneutic hypothesis”, “irrefutable” because it is just “a matter of taste”, looks wrong, wouldn’t you agree?” SH: Hypotheses non fingo. I am not purporting to answer the how/why question about feeling. It is the hypotheses that purport to do so that are hermeneutical. If I say “feeling is explained by the fact that you have a subsytem with privileged access to analog images etc. etc.” how can anyone ever refute that? It’s just a decorative description you either like or don’t like. SWM: Build the systems and put them through their paces. Thanks for a good discussion. I now believe I understand your position much better than at the start. I hope I have managed to also make my own clear enough as I have not always succeeded in doing that.
MIND-READING, PERFORMANCE MODELLING AND EXPLAINING FEELING
(Reply to Stuart W. Mirsky)
I think we know one another’s respective views now, and are unlikely to inspire any changes. I’d like to close by pointing out a recurrent misunderstanding that has dogged our exchanges from the outset. You, Stuart, seem to think that the task is to (1) design a robot that we can mind-read as reliably and effectively as we mind-read one another (and we do agree that T3 is that candidate),
(2) explain his performance capacity (and we do agree that whatever mechanism successfully passes T3 explains that) and then (3) explain how and why it feels: This is the point at which we disagree. You think certain interpretations of the mechanism that successfully generates T3′s performance capacity explains how and why he feels. I think what’s missing is a causal explanation of how and why he feels, along the same lines as the causal explanation of how and why he can do what he can do — and I give reasons (not stipulations!) why that causal explanation is not likely to be possible. You are satisfied that your mental interpretation provides that explanation, and your reason is that our mind-reading is right. I reply that whether or not we are right that T3 feels, the mechanism that generates T3 does not explain that fact that he feels, even if he does feel because that mechanism would (for all we know, or can know) produce the very same (performance) outcome even if he didn’t feel. (And if a feelingless T3 [or T4] is impossible, no one can explain how or why it’s impossible. Mentalistic interpretation of the T3/T4 mechanism certainly does not explain it.) So this is not just a manifestation of the other-minds problem, but of the anomalous causal, hence explanatory status of feeling.
Stuart W. Mirsky
Reply to Stevan Harnad’s remarks of 2/23/11 @ 6:52 AM Yes, we have done one of the only things discussions like this can do — we have reached the bottom line issue(s) that divide us. In this case your point that “whether or not we are right that T3 feels, the mechanism that generates T3 does not explain that fact that he feels” neatly sums it up I think. Why there is still a division between us, I believe, is because I think that a description of what kinds of processes performing what tasks produce feeling behavior DOES causally account for feeling. This hinges, of course, on my view that feeling is adequately revealed in behavior, as you note — something you apparently don’t share. And it looks unlikely I can bring you around to sharing it. But let me say a bit about why I think mine is the right view. I think (following Wittgenstein) that language has its genesis in a public environment and so our words (and thus our concepts) have their basis in publicly accessible criteria. Thus the ideas we have about mind (characterized by all those “weasely” words you bridle at) are necessarily grounded in public criteria. That’s WHY the words seem “weasely”. They are public phenomenon words at bottom, being applied in what is intrinsically a non-public context. As a result, they denote much more imprecisely than public criteria words do when applied in a public venue, and they often lead us into conceptualizing in inapplicable ways. Hence the notion that minds are a distinct kind of thing (often conceived of as souls or other non-physical entities, whatever that might be). I don’t want to introduce a debate here about language or Wittgenstein but I do want to make clear why what doesn’t seem reasonable to you seems to perfectly reasonable to me. On my view, when we speak of mental phenomena we’re really stepping outside the area in which language works most effectively. I don’t want to suggest that we cannot reference mental phenomena because we do all the time and sometimes to very good effect. But I want to say that, especially in doing philosophy, we often end up playing in the penalty zone without realizing it. Speaking about minds and consciousness and feeling and so forth IS, on this view, to speak about behavioral criteria, not about some ethereal entity hidden from view in others. But I agree with David Chalmers’ point that we often have private aspects of our own experience in mind, too, when using these words. There are public AND private referents involved in mental words, reflecting our own experience of our mental lives as well as our experience of the public zone we share with others. The problem, as I see it, lies in the fact of differentiation. Just because when I speak of “feeling” I have in mind a range of my own personal experiences (feelings of physical sensations, feelings of emotional events, feelings of being aware of the things I am aware of and so forth), doesn’t mean that that is the main use of the word “feeling” for me. Or the important one in a public context. If our Data creature acts in every way like us (meets your T3 test) then even you have agreed you would not eat him (assuming he were edible), chop him up, enslave him and so forth. You would act toward him as if he were a feeling entity. And you would be right to do so. So the question is not, on my view, whether any entity, Data or other human being, has a mental life that is just like mine but whether it, or he, has a mental life at all. And behavior tells us whether he does or not. It is the only thing that can tell us and we have no reason to expect anything more. But if there is no reason to expect direct access to another’s mind, then identifying the processes in a given operating platform (brain, computer or whatnot) which produce the requisite behaviors surely meets the standard of identifying a cause of those behaviors. Of course, the more finely grained the description, the better the understanding of the causative factors. What else could “cause” mean in this case than that? By the way, I see that John Searle has penned a piece in this morning’s Wall Street Journal concerning the recent Watson business on Jeopardy. He invokes the usual Chinese Room scenario to show that Watson, a computer, doesn’t understand. Of course he’s right in the sense that Watson lacks plenty, even at this stage. Watson is still underspecked like the Chinese Room. Searle’s assertion is that this shows that Watson-like entities lack the potential to cause understanding though, as usual, he does not say anything about what it takes to actually cause it except that “the brain is a causal mechanism that causes consciousness, understanding and all the rest” and Watson, being a computer, lacks that causal capacity. But not telling us what that capacity is becomes, I think, an easy out. What Searle’s argument continues to miss is that being a limited system, like Watson, doesn’t mean say anything compelling about other less-limited systems like Watson so long as there is no reason to believe that brains’ causal powers in this regard are necessarily different from what computers can do. (His later argument, after the CRA, asserts that computational processes can’t have any causal power at all because they are man-made and take their nature of being computational from their makers which denies them real world causal efficacy — but I think that’s a worse argument than the original CRA.) I’m thinking of penning a rebuttal if I have the time this morning but it’s unlikely to be brief enough for a letter to the editor. So perhaps I’ll just pass for now.
Stuart W. Mirsky
Addendum to recent reply to Stevan Harnad: In my effort to keep my text short, I neglected to make an important point in my last response and this could turn out to fuel further misunderstanings between us. So let me just add it now. You’ve indicated that you think the problem resides in the fact that a T3 entity could pass a T3 test with or without the mental stuff we have, i.e., that it could act as if it feels while, in fact, not feeling. My point above, that we ascribe feeling to entities that act in a feeling way, is meant to cover this case though I failed to be sufficiently explicit. That is, on the view I’m espousing, there can be no difference between two entities that pass T3 in the same way. This is partly premised on my earlier point that all that we mean by ascriptions of feeling in such cases is that they behave in a certain way. But it is also partly reflective of my view that the notion of philosophical zombiedom is untenable. In this (as in many things) I’m inclined to agree with Dennett. Given an entity that does everything we do in the same generic way (and Dennett takes this up to what you characterize as the T4 test), it can make no sense to suppose anything is missing. Your own acknowledgement that you would not, in your behavior toward such an entity, treat it as if anything WERE missing, implicitly confirms this. As I noted earlier on, if a chair or other inanimate object suddenly began behaving in a conscious way (speaking intelligently and autonomously to us, reacting to situations in apparent fear or concern), however hard this might be to imagine, we would be left with little choice but to treat it as another conscious entity — once we had successfully eliminated other possibilities (is there a hidden operator? are we under the influence, etc.?), of course. We would likely have to revise our understanding of the world though. Maybe brains would no longer be thought of as a necessary (or causal) condition for minds after all. But what we would not have to have, what we would never require, is access to the chair’s “mind”. We don’t need it in normal circumstances so why would we need it in abnormal ones? (Sorry for the extra text but I just wanted to be clear.)
NO: DOING ≠ FEELING, AND IF GENERATING DOING ALSO GENERATES FEELING, WE DON’T (AND CAN’T) EXPLAIN HOW AND WHY
(Reply to Stuart W. Mirsky)
Hi Stevan, I think I missed your response in all the hubbub. Let me try to sum up the disagreement here.
HOW/WHY ≠ WHETHER
(Reply to Richard Brown)
R.BROWN: “1. You claim that we will never be able to explain how and why we feel. So, even when we have a T3/T4 robot we will be unable to say whether it feels or not and that is because we can always ask ‘what causal power does feeling confer?’”
No, the causal question of how/why an entity feels is not the same as the factual question of *whether* it feels. Even if we had God’s guarantee that a T3 or T4 robot feels, we would only know, from the T3/T4 theory, how/why a T3/T4 system can do what it can do (its know-how), not how/why it feels (i.e., how/why it feels like something to be able to do what it can do, to be the T3/T4 system).
R.BROWN: “2. This looks to me just like saying ‘zombies are conceivable,’ yet you seem to deny that you are talking about, or even need to talk about, zombies.”
No, it’s not “zombies are conceivable,” it’s “feelings are inexplicable.” You don’t need zombies for that.
(But, if you like, they can be marshalled thus: It is inexplicable how and why we are not zombies; or how and why there cannot be zombies. Same thing as: It is inexplicable how and why we — or T3/T4 robots — feel.)
R.BROWN: “3. If feeling turns out to have relatively little causal role then it cannot be an argument against a theory that adding feeling doesn’t add any causal powers.”
We can’t even explain how/why feelings have relatively little causal role: We can’t explain how/why/whether they have *any* causal role.
And the only theory is the theory of how T3/T4 systems can do what they can do. That’s a causal theory, a causal explanation.
“Adding” something to it that has no causal role is adding nothing to the *theory*. It’s a datum — an unexplained datum. (We know T3/T4 feels, because God told us; and we know we feel, because of the Cogito. But, theoretically speaking — i.e., explanatorily speaking — that leaves us none the wiser than T3/T4′s causal powers and explanation alone already leave us.) R.BROWN: “It then becomes an open question whether adding feeling does add any causal powers, with whatever the answer turns out to be providing us with evidence for or against various theories of consciousness.”
You lost me. If we can’t give any causal explanation of feeling’s causal role, what kind of a “theory” of feeling (consciousness) is that?
R.BROWN: “At the first online consciousness conference David Rosenthal gave his argument that we have overwhelming evidence that adding feeling doesn’t add (significant) causal powers, which then turns out to be evidence for the higher-order theory.”
What is a “higher-order” theory? What does it explain? Does it tell us how or why we feel? If not, it is probably just hermeneutics (as I’ve had occasion to suggest more than once in this discussion!). The reverse engineering of T3/T4 is the explanatory theory. And it explains doing, but not feeling.
(Did David, by the way, mention in passing what “insignificant” causal powers feelings have? I’d settle for *any* causal power…)
Harnad: No, it’s not “zombies are conceivable,” it’s “feelings are inexplicable.” You don’t need zombies for that.
You can say that you don’t need them but every time someone asks why feeling can’t be explained you appeal to a zombie like intuition; we can imagine that being done in the absence of feeling. What is more zombie than that? But I guess ultimately this doesn’t matter
Harnad: You lost me. If we can’t give any causal explanation of feeling’s causal role, what kind of a “theory” of feeling (consciousness) is that?
It would be something like the higher-order thought theory of (feeling) consciousness.
Harnad: What is a “higher-order” theory? What does it explain? Does it tell us how or why we feel? If not, it is probably just hermeneutics (as I’ve had occasion to suggest more than once in this discussion!). The reverse engineering of T3/T4 is the explanatory theory. And it explains doing, but not feeling.
Well, I don’t want to rehearse the high-order theory here but the basic jist is that to feel pain is to be aware of myself as being in pain. This, to be very brief about it, explains why it feels painful for me. It does so because that is how my mental life appears to me and that is all that there is to feeling. This is definitely not hermeneutics, whatever that is. Now I get that you don’t accept this as a theory of feeling but the point is that your argument stands or falls with the success of actual theories that try to explain feeling and so has to be evaluated by how well these theories fare. Higher-order theories are particularly relevant since these theories predict that feeling will have little, if any, significant causal role to play in doing, just as you suggest. If you are going to claim that something is unexplainable you need to understand the various proposed explanations in order to say that they don’t work; especially if you are not trying to give an a priori conceivability argument.
EXPLAINING FEELING: A CHALLENGE
(Reply to Richard Brown)
R.BROWN: “every time someone asks why feeling can’t be explained you appeal to a zombie like intuition”
In saying: “Thank you (T3/T4) for explaining how and why we can do what we can do; now please explain to me why it feels like something to do all that…,” am I appealing to a zombie-like intuition? But I don’t even believe in zombies! I just want to know how and why we feel.
R.BROWN: “the high-order theory [of (feeling) consciousness]… is that to feel pain is to be aware of myself as being in pain. This… explains why it feels painful for me… because that is how my mental life appears to me and that is all that there is to feeling. This is definitely not hermeneutics, whatever that is”
This is one of the (many, many) reasons I urge dropping all the synonyms, euphemisms, paralogisms and redundancies and just call a spade a spade:
The question was: “How and why do we feel (anything)?”
The (“higher-order”) explanation (now transcribed here minus the synonyms, euphemisms, paralogisms and redundancies) seems to be:
“to feel (something) is to feel something. This… explains why it feels like something …because that is how my feeling feels, and that is all that there is to feeling.”
Pared down to this, it’s definitely not hermeneutics; it’s tautology. Put back the synonyms, euphemisms, paralogisms and redundancies and it becomes hermeneutics: a Just-So Story.
But it’s certainly not explanation!
R.BROWN: “Now I get that you don’t accept this as a theory of feeling but the point is that your argument stands or falls with the success of actual theories that try to explain feeling and so has to be evaluated by how well these theories fare.”
But my argument is that no one has proposed a (non-psychokinetic) causal theory of feeling, and I give reasons why I don’t think there can be one (no causal room, and everything works just as well without feeling). Such an argument does indeed stand or fall on the success or failure of actual causal explanations. But the theories (when they are causal theories at all) seem to fail; so the argument would seem to stand.
Hermeneutics is not a causal theory, but a Just-So story.
R.BROWN: “Higher-order theories are particularly relevant since these theories predict that feeling will have little, if any, significant causal role to play in doing, just as you suggest.”
Well that would be convenient: I am asking for a causal explanation — of something that it seemed perfectly reasonable to expect to be causally explained — and instead I am given a theory according to which feelings have “little, if any” causal power.
Let’s take these one at a time: If little, then what’s the causal theory of how and why feelings have this (little) causal power?
And if feelings have *no* causal power, for theoretical reasons — well that does seem to be a rather handy way of deflecting the call for a causal explanation, doesn’t it? Feelings are acausal ex hypothesi ergo propter hypothesem! (One wonders if there are other things one can get out of explaining causally by hypothesizing that they are acausal: or are feelings the only thing? To me that sounds more like explanatory shortfall than theoretical triumph, if a Just-So Story tells me that that’s just the way things are…) But does it really work, to say that feelings are acausal? Our intuitions prepare us, somewhat, to accept that passive feelings might be just decorative, not functional, and in that sense acausal — though one can’t help wanting to know why on earth they’re there at all, then — and not just there, but center-stage.
It’s not just passive feelings that are at issue, however. There’s also feeling like doing something; doing something because you felt like it, not because you were pushed. That’s a harder distinction to wave away as merely decorative.
Bref: I’d say that any “higher-order theory” that declared that one should not be troubling one’s head about how and why we feel — because feelings are acausal — was simply begging the question.
R.BROWN: “If you are going to claim that something is unexplainable you need to understand the various proposed explanations in order to say that they don’t work; especially if you are not trying to give an a priori conceivability argument.”
Definitely no a-priori conceivability arguments (or zombies!). So I’m all ears: According to the higher-order theories, how and why do we feel? I’m happy to consider each theory, one at a time. This, after all, was the challenge at the end of my video (though I think Richard’s edited version cut it out!). (All I ask is that the theory should reply without wrapping feeling inextricably into further synonyms, euphemisms, paralogisms and redundancies that might camouflage the fact that they don’t provide any answer.)
Stuart W. Mirsky
Adding my two cents, I have to say that it’s strange to end an argument of this sort simply by declaration. It smacks of an appeal to intuition, at the least. I think it’s reasonable to say, as Stevan does, that the concept of feeling is not the same as the concept of doing and that it follows from that that feeling does not equal doing. But that doesn’t imply the next statement in Stevan’s last response to me: “IF GENERATING DOING ALSO GENERATES FEELING, WE DON’T (AND CAN’T) EXPLAIN HOW AND WHY”. Just being different notions does not imply an incapacity to explain why the referent of one of those notions occurs. It’s a separate claim to insist that we can’t explain the how and why of feeling’s occurrence and has no apparent logical dependence on the prior statement that feeling doesn’t equal doing (or vice versa). At least not without some argument that leads to such a conclusion. Of course, I think I gave a fair account of how we can be perfectly comfortable that we do have a way of explaining the occurrence of feeling: 1) Feeling is revealed in certain behaviors. 2) We can test for and observe those behaviors. 3) Therefore we can test for and observe the occurrence of feeling in a behaving entity. The first premise stands on a Wittgensteinian analysis of the concept of feeling (the way we use the word) and on its implication, that it’s just unintelligible to imagine we could have perfect replicas of feeling entities (both in terms of actions and internal functionality) without the feeling part. The rest of the argument is relatively simple and seems to be self-explanatory; not in need of extensive support. Based on this, I suggest that an account which specifies the particular functions that need to be performed for feeling toi occur, coupled with the argument that these functions can be achieved computationally, can then provide a perfectly acceptable explanation of how and why feeling occurs, i.e., it is just this and this and this set of tasks being performed by the entity’s relevant components. This does not guarantee, of course, that a computationalist account is the right one. But it does provide an explanation that could be true.
CORRELATION ≠ CAUSATION (NOR EXPLANATION)
(Reply to Stuart W. Mirsky)
S.MIRSKY: “I think I gave a fair account… of feeling: 1) Feeling is revealed in certain behaviors. 2) We can test for and observe those behaviors. 3) Therefore we can test for and observe the occurrence of feeling in a behaving entity.”
(1) Feeling is correlated with certain behaviors (T3) (and with certain brain-states, T4).
(2) We can test for and observe those behaviors (and brain-states).
(3) We can test for and observe the correlation of those behaviors (and brain-states) with (reported) feeling.
(4) Therefore we can provide a causal explanation of those behaviors (T3) (and brain-states, T4).
(5) We cannot, however, provide a causal explanation of the feeling which is correlated with those behaviors (and brain-states). Correlation ≠ Causation (nor does correlation explain causation).
S.MIRSKY: “The first premise stands on a Wittgensteinian analysis of the concept of feeling (the way we use the word) and on its implication, that it’s just unintelligible to imagine we could have perfect replicas of feeling entities (both in terms of actions and internal functionality) without the feeling part.”
Whether a feelingless T3 (or T4) robot is imaginable or unimaginable, intelligible or unintelligible, possible or impossible is *irrelevant* to the question of whether we can explain how and why we (or a T3 or T4 robot) feel.
(By the way, if a feelingless T3 or T4 robot is indeed impossible — as I rather suspect it is — that is an undemonstrated ontic impossibility, not a formal proof of impossibility, nor even contradictory to an empirical (causal) natural-law. Hence it is an unexplained impossibility.)
Wittgenstein on private states and private language in no way resolves or even casts any light at all on the problem of explaining how and why we feel.
S.MIRSKY: “an account which specifies the particular functions that need to be performed for feeling to occur, coupled with the argument that these functions can be achieved computationally, can then provide a perfectly acceptable explanation of how and why feeling occurs, i.e., it is just this and this and this set of tasks being performed by the entity’s relevant components.”
And the reason the causal mechanism underlying “this and this and this set of tasks” needs to be a *felt* one, rather than just a “functed” (i.e., executed, implemented) one…?
S.MIRSKY: “This does not guarantee, of course, that a computationalist account is the right one. But it does provide an explanation that could be true.”
The problem is exactly the same for a computationlist, dynamicist, or hybrid computational/dynamic account: The causal account accounts for doing, not for feeling. The causal explanation that can generate the doing is hence true (and complete) for doing, but completely empty for the feeling that is piggy-backing on the doing (if it is).
Stuart W. Mirsky
Response to Stevan Harnad
SWM: On a Wittgensteinian analysis of the word “feeling”, when applied in a public context (i.e., to entities other than ourselves), the term denotes certain behaviors. The fact that we also use the word to refer to the vague (because impossible to particularize) sense of being aware that we have when we’re attending to (are aware of) things, is not intrinsic to the public usage. And it’s the public usage that’s at issue when we’re trying to determine if another entity has feeling (as in “is aware of” what’s going on around it, what’s happening to it, etc.) This issue, I expect, is the one that really divides us. Stevan seems bent on applying a vague term, for an indistinct referent, in a public usage venue where a different application makes sense. We never need access to other human minds to be assured that they have minds (i.e., that they are feeling, as in having experience) when they are in an awake state. And we don’t need to do anything different with regard to other kinds of entities, whether chairs, tables, computers, arthropods, cephalopods, mammals, or aliens from outer space. Demanding something more looks like a mixing of categories which effectively imports a metaphysical problem, that of Other Minds, into a scientific milieu. There is no question that, absent an analysis like Wittgenstein’s, the Other Minds problem appears intractable. But science isn’t about such concerns but about accounting for the phenomena we deal with in the public domain where things are observable (if not in fact, then at least in principle). SWM: Reports are only part of the story. Creatures that cannot report “I am feeling this, seeing that, etc.” can still be observed to behave in feeling ways which is why we cringe when we see an animal in pain. Our response is to their behavior, not their reports (though reports ARE a sub-class of behaviors). SWM: On the view I’ve offered, the behaviors are understood as expressive of feeling, not merely ancillary to them. A perfect imitation (right down to the internals) would not be conceivable, even if partial imitations are. The problem is that, because we can conceive of successful partial imitations (lifelike robotic models, convincing computational question answerers, like highly sophisticated Watsons), we think we can also conceive of the so-called philosophical zombie type imitation, the one that passes an open ended Turing test for a lifetime at all levels (verbal, behavioral, internal functionality). But, confronted with such an entity, even you have assured us you would not eat it, torture it, treat it like an inanimate object, etc. So on one level you recognize the inconceivability of feeling behavior without feeling, while on the other you hold out for the Other Minds solution, which is irrelevant to the scientific question of whether machines can have conscious minds (feeling). SWM: True. The two concepts are distinct and correlation doesn’t explain causation. What correlation does do is provide a tool for imputing causation. SWM: If a T3 or T4 robot is unintelligible then having the feeling behavior in the right context IS having the feeling, and any explanation for how that feeling behavior is generated explains how the feeling comes about. SWM: If Wittgenstein’s point about the publicness of language is right, it renders the question meaningless, hence it can neither be possible nor impossible. This is a different question, however, from whether it is possible that minds, consciousness (or what you want to just call feeling) exist outside a physical framework entirely. But for that kind of claim (dualism) to be upheld, I think we would need different information about the world than we currently have. Barring evidence of minds divorced from bodies, our current information seems to accord quite well with the way our language generally works re: questions of mental phenomena in other entities so there’s no reason to look for something non-physical in explaining feeling behavior. But I don’t think you are saying otherwise, which is why I’m surprised at your insistence on asserting the impossibility of causal explanation for feeling. SWM: It casts light on what we mean by “feel” in the different contexts. I don’t dispute with you that we have a private sphere of experience, nor did Wittgenstein (he often spoke of mental pictures). The issue revolves, rather, around the question of whether a theory that accounts for the causality of feeling behaviors in a machine (or any other entity) explains the occurrence of feeling itself (my reply to your challenge). I am arguing that it does, in the only meaningful sense of this question, and that to suppose otherwise is to shift the underlying ground from the scientific to the metaphysical. Since the questions of cognitive science are manifestly scientific, metaphysical concerns have no role here. Thus an explanation of the occurrence of feeling (your use of that term), in terms of system operations, could work in cognitive science. I agree, though, that if one doesn’t accept the Wittgensteinian solution to the Other Minds problem, THAT issue persists. But my point is that it isn’t a scientific problem any longer. SWM: “Needs to be”? The reason the entity achieves feeling is because, on this theory, feeling is that certain set of processes which, combining in a dynamic system, collect information at one level, transform it at other levels and link it with other information maintained by the system. If a machine run by such a system is seen to behave in ways that manifest feeling (expressing, by report or behavioral demonstration, an awareness of itself, of other entities, of what it is doing and thinking, etc.), then that is sufficient to tell us that feeling qua sentience is present. If we turn off aspects of the system and the behaviors cease, then we can say that those aspects (those processes performing those tasks) are the causal elements of the feeling and, if we turn them back on and the feeling behavior returns, we have even stronger evidence for this thesis. SWM: This brings us back to the original problem, i.e., does “feeling” denote only the vague and difficult-to-pick-out sense we have of being subjects, of experiencing, which you describe as “piggy-backing” (epiphenomenalism)? Or does the term denote something (s) in the shared public sphere of our experiences, i.e., the ongoing complex of behaviors we recognize as feeling in others? My view is that the subjective application of the term which you have made paramount is really secondary to (and derivable from) the public application. But if you don’t share that view, it’s not hard to see why we are at loggerheads. However, in that case, my argument comes down to this: Insofar as this is about questions of cognitive science, the metaphysical problem of Other Minds that you have raised hardly seems relevant.
CORRELATION VS CAUSATION, AGAIN
(Reply to Stuart W. Mirsky)
S.MIRSKY: “On a Wittgensteinian analysis… “feeling”… denotes certain behaviors.”
Feeling means feeling, whether I’m talking about him or about me. When I say I feel tired, I mean I feel tired. When I say he feels tired, I means he feels tired — not that he’s *behaving* tiredly, but that he’s *feeling* tired. By the same token, if I say he’s lying, I mean he’s lying. Not that he’s behaving mendaciously but that he’s lying. I may be mistaken. He may not be feeling tired. He also may not be lying. But I mean what I mean in both cases. The only difference is that when I say he’s lying, there’s a way to settle the matter for sure. When I say he feels tired there’s not.
There’s nothing vague in any of these four cases: feeling/lying, me/him.
S.MIRSKY: “We never need access to other human minds to be assured… they are feeling… Demanding… more… imports a metaphysical problem… Other Minds, into a scientific milieu.”
Good thing we don’t need access to other minds to infer they are feeling!
The other-minds problem is an epistemic, not an ontic problem, but never mind.
The scientific problem is not to be “assured” that others are feeling, but to explain how and why they feel.
S.MIRSKY: “the Other Minds solution, which is irrelevant to the scientific question of whether machines can [feel].”
The scientific question is not *whether* machines can feel but how and why…
S.MIRSKY: “If a [feelingless] T3 or T4 robot is unintelligible then having the feeling behavior in the right context IS having the feeling, and any explanation for how that feeling behavior is generated explains how the feeling comes about.”
No. If a feelingless T3 or T4 robot is impossible then if T3/T4 behaves as if it feels, it must (somehow) be feeling. Fine. Now we know it feels.
Now: How and why does it feel? (Please don’t reply that the answer is that it would be impossible for it not to! That’s not a causal explanation.)
S.MIRSKY: “a theory that accounts for the causality of feeling behaviors explains the occurrence of feeling itself”
No, a theory (T3/T4) that accounts for behavior accounts for behavior. Why and how the behavior is felt needs an account of its own.
S.MIRSKY: “to suppose otherwise is to shift the underlying ground from the scientific to the metaphysical….Thus an explanation of the occurrence of feeling in terms of system operations could work.”
How have I shifted from scientific to metaphysical in asking for a causal explanation of how and why we feel? If “system operations” theory answers the question, let’s hear the answer. The trick will be to show how it does the causal work that wouldn’t be identically done without feelings.
(Don’t remind me that it’s impossible for T3/T4 to be able to behave exactly as if it feels without feeling: explain to me why and how it feels — or, if you like, why and how it’s impossible. [Assume I already agree that it's impossible: No contest.] No other-minds problem-problem. T3/T4 is feeling. It is indeed impossible for it not to be feeling. I get a headache just thinking about it. Now, just explain, causally, how and why it’s impossible.)
S.MIRSKY: “The reason the entity achieves feeling is because, on this theory, feeling is that certain set of processes which, combining in a dynamic system, collect information at one level, transform it at other levels and link it with other information maintained by the system.”
And why and how is that processing, combining, collecting and linking felt?
S.MIRSKY: “If a machine run by such a system is seen to behave in ways that [behave as if] feeling… then that is sufficient to tell us that feeling… is present.”
Indeed. Agreed. And now, the explanation of how and why this definitely present feeling is caused…?
S.MIRSKY: “If we turn off aspects of the system and the behaviors cease, then we can say that those aspects… are the causal elements of the feeling and, if we turn them back on and the feeling behavior returns, we have even stronger evidence.”
Correlation does not explain causation. If you turn off the correlates of feelings, the feelings are gone: Now: how and why do the correlates of feeling cause feelings?
S.MIRSKY: “does “feeling” denote only the vague and difficult-to-pick-out sense we have of being subjects, of experiencing, which you describe as “piggy-backing” (epiphenomenalism)? Or does the term denote something(s) in the shared public sphere of our experiences, i.e., the ongoing complex of behaviors we recognize as feeling in others?”
There’s nothing public or social (or vague or difficult-to-pick-out) about feeling a migraine. And until it’s explained how and why it’s felt, the only way it can be described is as piggy-backing (somehow, inexplicably) on the causal mechanism of T3/T4 capacity.
Bernie Ranson
Professor Harnad, I have been thinking about your statement that cognitive capacity + feeling = cognition I wonder if you could define and exemplify “cognitive capacity”?
COGNITIVE CAPACITY
(Reply to Bernie Ranson)
B.RANSON: “could [you] define and exemplify ‘cognitive capacity’?”
Everything an organism can do. (One might want to subtract “vegetative capacity,” such as respiration, thermoregulation, posture…)
Examples: Detecting, identifying, and manipulating objects; category learning, language… Cognition is the mechanism of cognitive capacity. It is what cognitive science is trying to reverse-engineer.
Joshua Stern
SH: None of your comments (which are only about the Chinese room and symbol grounding) bear on the problem of feeling (“qualia”); but the target essay does.” JS: Yes sir, that is exactly right. But what this shows is that you have given up on the symbol grounding problem, and are now completely engaged in a qualia grounding problem. – OK, I said that earlier, and then said I would post again. I will clarify the above briefly, because I now realize it must be misread, then take a different cut at the whole problem. – First, when I said what I said above, what I meant was that Harnad has given up on grounding symbols, and is now trying to ground not feelings in general, but Searle’s claimed feeling that “something is missing”. What is it like to be John Searle? It must include a feeling that something is missing from the CR. Harnad takes this as worth addressing. I don’t. It is not an argument. However, there are other things that Searle says that constitute actual arguments that I believe are worth addressing, and in fact, that I believe Searle gets right, in spite of the fact that on the major question I believe the “systems argument” is 100% conclusive, always has been, always will be. OK, what could that possibly mean? I make one claim here, that Harnad has granted above – that computation is itself a physical and causal process. We will see that that is enough, that it justifies T2 as the proper test, and computation as the only issue in cognition. Searle’s CR is a lovely little intuition pump from back in the day when functionalism was supposed to be the foundation of AI, of cognitivism, and of philosophy of mind. “Anything that has the function of X is X”, it was claimed, giving us multiple realizability as a conjunct of positivism, of T2 as a behavioral test. But the functionalist claim was not good enough for Searle, and he protests it in a number of ways. As in the legal doctrine that it’s OK to claim in a single defense that, “I never borrowed the pot, it was broken when I got it, and it was intact when I returned it,” it is not necesessary that everything Searle ever claimed about the CR fits together. What I believe Searle got right is that the claims of functionalism are not fundamental, they are descriptive but not explanatory. The T2 test is descriptive but not explanatory, it does not give us a constructive solution, “Write this program and it will pass T2”. Until and unless, first, something actually *does* pass the T2 test, and second, that we are able to learn from this what it is that allows it to pass T2, then Searle has a point. Harnad shifts the argument in this paper from understanding generally to symbol grounding: “Let us say that unlike his Chinese symbols, Searle’s English symbols are “grounded” in his sensorimotor capacity to interact with the things in the world that his symbols refer to: they connect his words to their referents.” (p 4) Harnad then moves from this to arguments in favor of T3, and then against a T4. If you want to ground a symbol, these might be good moves. However, it is tendentious to assert that grounding symbols answers Searle’s complaint. I concede I have no real idea how to directly answer Searle’s complaint, which I do not consider valid in the first place. Ex Falso Quodlibet. But, besides his complaint, Searle makes an implicit demand – build me a CR! He simply assumes this in the paper, pretending to do a reductio ad absurdum on it. The reductio fails but the demand is valid. To repeat the previous point, the implicit demand is dual
– first, show me, second, explain yourself. Build a machine that passes T2 (or T3 or TX), and explain just how you did that, how it can be duplicated. Short of doing the empirical demonstration, we amuse ourselves with discussions of principles. What could allow a system to pass T2, or T3, or TX, or what prevents anything from ever passing T2, or T3, or TX? Searle is entirely reasonable on this. He agrees that machines can pass T2 (etc), because humans are machines. That is reasonable, but it is still not explanatory, it is an observation but still not constructive. Searle has no constructive insight into that. He observes that humans do (pass T2), and that computers do not (yet, empirically) pass T2. The questions of what principles we have, and what valid arguments have been made, and can be made, pro or con regarding whether digital electronic computers qua computers can ever pass T2 (etc), are what I believe need to be argued. I will sketch such an argument. Let us take a computer system and have it attempt T2. Let it fail.* We add sensorimotor capacities, and then attempt T3. We succeed. Hurrah! Now we turn off the computer. The robot’s battery is still charged and the power switch is on, but the robot doesn’t move and now fails T3. What has changed? Hasn’t Harnad said that it is the combination which is needed, and haven’t we broken the combination? Well, I claim that the critical element is the computational element, and the rest is gingerbread, and we have just shown this. Let’s unpack it further. What have we done, by turning off the computer? We have removed the ability to have causal interactions, textual or mechanical. Harnad claims that mere textual interactions can never convince anyone of anything, but this flies in the face of the medium you are reading this on right now, as it flies in the face of the original idea of the Turing Test. Just as we have granted that computation is itself a physical and causal process, so is any textual exchange necessarily a physical and causal process. There may be some difference in degree between T2 and T3, but not of kind, even the T2 test has sensorimotor aspects. And, it turns out, those aspects need embedding in a causal sequence. To summarize, computers can (in principle) pass a T2 because they are the proper sort of machine, that which can participate in physical, causal interactions. All the rest is detail. I suggest a separate issue is whether we ever wanted to ground symbols in the first place. Is even symbolic computation, really symbolic? Well, no, not really. That is, the only symbols that a computer might be said to crunch are ones and zeroes. And, truth be known, there are no ones and zeroes inside of your computer chip. There are (again) physical machineries, circuits, flipping around bags of electrons, again physical, in a sequential, causal manner. The symbols are as much a matter of convention and degree as arms and eyes might be. The T2 test does not suggest the computer sends only ones and zeroes in its messages, there is already some kind of arbitrage and minimal levels of interaction specified. It is a methodological convenience to say that computation is symbolic – although of course, the neural network folks might argue even that. To make a long story short, Turing got this all right seventy years ago. The Turing Machine reduction is a cannonical form of all such related problems. We don’t have to use the cannonical form, it is not generally convenient, but all roads lead to ones and zeroes. In granting that some machines pass the test of consciousness, Searle leaves open the idea that some causal machinery other than biological humans, may do so. I suggest the Turing Test (T2) is actually much stronger than it looks, what it tests is exactly what needs testing, and its main fault is only what we demand further of it that goes beyond testing, a constructive answer to “Just how is that done?” rather than just a descriptive declaration of, “Congratulations you passed!”. I have seen no result, and few arguments, that computation cannot pass T2. Most of all, I want to oppose the tendency of my compatriots in compsci to give up any philosophical argument in favor of engineering to say “OK, we’ll just imitate until you can’t tell and pass T2 that way, but sure, it’s not *real* intelligence”. Searle grants that, too, but I think we can do better. I see no issues of principle, having explained that the arguments of physicalism are moot since computation is already physical. Two things remain, a demonstration, and an explanation, and yes, the explanation had better come first, which is why we need to focus on the right issues and keep up the discussion. What is called for is “Do this and this, and the resulting program will be just as intelligent, for exactly the same reasons, as humans.” That was the original statement of the cognitivist, computational movement, that if a machine passed T2 it would be for good reason. Much more needs be said, but perhaps that is enough for now. – *Harnad expresses doubt that T2 can be passed, but he does so only by breaking the T2 paradigm, “In fact, without the sensorimotor capacities of a robot, it is not clear how even the email T2 could be passed successfully: Would it not arouse immediate suspicion if our pen pal was always mute about photos we sent via snail mail?” (p 2). Well, I agree this far, that one can escalate demands on T2 or the CR to any extreme. I’d suggest such questions as, “How are you today?”, or “Have you changed your mind yet about what we talked about yesterday?”, or imperatives like “Jump up!”, or statements of fact like, “Your shoe is untied”, are interesting tests, but do not change the minimalist validity of T2. In any case, until T2 is easily passed we already have enough to worry about.
UNDERSTANDING THOUGHT EXPERIMENTS
(Reply to Joshua Stern)
J.STERN: “Harnad… is now trying to ground… Searle’s claimed feeling that “something is missing”
Symbols (not “claims”) are the things that need to be grounded (causally connected) to the things in the world they refer to (e.g., via T3 robotic capacity).
Grounding symbols is not enough to give them meaning (nor to make them understood). For that you need grounding plus what it feels like to mean, or to understand.
That’s what Searle (correctly) says he would be missing with the Chinese squiggles and sqoggles.
You don’t have to have written a computer program that passes T2 in order to be able to see this.
J.STERN: “The T2 test is descriptive but not explanatory”
Correct. It would be the explanation of the causal mechanism of the system that could successfully pass T2 that would be explanatory.
There’s no such explanation today. There need not be, in order to discern that passing T2 via computation alone would not be enough.
J.STERN: “a computer… [plus] sensorimotor capacities [passes] T3…turn off the computer… The robot’s battery is still charged and the power switch is on, but the robot doesn’t move and now fails T3. What has changed?”
You no longer have a system that can pass T3. (That’s why we don’t try to interact with people while they’re in delta sleep, or a coma, or brain-dead…)
(BTW, I don’t think that the hybrid dynamical/computation system that passes T3 will just be a computer plus peripherals, but never mind.)
J.STERN: “by turning off the computer…[w]e have removed the ability to have causal interactions, textual or mechanical.”
Indeed. And your point is…?
J.STERN: “There may be some difference in degree between T2 and T3, but not of kind, even the T2 test has sensorimotor aspects.”
The kind of difference in degree that distinguishes apples from fruit: T3 (Turing-indistinguishable sensorimotor capacity) includes T2 (Turing-indistinguishable verbal capacity).
The sensorimotor I/O for T2 is trivial, and nothing substantive hangs on it one way or the other.
J.STERN: “T2 test does not suggest the computer sends only ones and zeroes in its messages”
The coding is arbitrary convention and the hardware is irrelevant. What matters is the algorithm, which is formal.
Nothing hangs on any of this, one way or the other, for the points I made about T2, T3, grounding, meaning and feeling.
J.STERN: “I have seen no result, and few arguments, that computation cannot pass T2.”
I think you’ve missed the point of the Chinese Room Argument. The point was not that computation alone could not pass T2. (I happen to think it can’t, and I give reasons; but neither Searle’s point about understanding, nor mine about either grounding or feeling depends on whether or not computation alone could pass T2.) The point was that if computation alone could pass T2, it would not be understanding — nor would it be grounded, nor would it be feeling.
Bernie Ranson
B.RANSON: “The crucial distinction between our capacities and those of computer/robots can be brought into focus by considering the distinction made by Searle between “observer-dependent” and “observer-independent” phenomena… [observerdependent:… money and computers… observer-independent:… metals and plastics… physical processes in our brains, and consciousness]” S HARNAD: “Interesting way to put the artificial/natural kinds distinction (except for consciousness, i.e., feeling, which does not fit, and is the bone of contention here); but how does this help explain how to pass the Turing Test, let alone how and why T3 or T4 feels?” And later B.RANSON: “a robot/computer combination is no improvement on a computer alone. The crucial, observer-independent features are still missing.” S HARNAD: “It’s certainly an improvement in terms of what it can do (e.g., T2 vs. T3). And since, because of the other-minds problem, whether it feels is not observable, it’s certainly not observer-dependent. Thank you very much for all your stimulating responses Professor Harnad. I do think that the observer-dependent/observer-independent distinction is important in clarifying this issue so I must try to explain why. In the last sentence I quoted I believe you are not using observer-dependent in quite the sense which I am borrowing from Searle. If we consider money (as contrasted with metal in coins) it is highly questionable whether that is observable, but its existence is observer-dependent in Searle’s sense. Money is only money when somebody says it is. Also, I don’t think the other-minds problem is much of a problem. We know what causes feeling, it’s our nervous system, and people and (some) animals with nervous systems feel. You would want really good reason to think that a walking, talking electrified pile of metal and plastic was also capable of feeling, without a nervous system, and in the case of computers and robots there isn’t one. I don’t understand why you think consciousness “doesn’t fit”; for me it does. The existence of consciousness is observer-independent; consciousness would be there (say in early mammals) whatever anybody might say about it. The existence of computation is observer-dependent. Something is only computation when somebody says it is, something is only a computer when somebody says it is. B.RANSON: “Professor Harnad suggests that our robots and computers are able to do a tiny fraction of what we do. My submission is that they are not able to do even that tiny fraction, that what they do do is not even related to what we can do, or at least, not to the relevant part of what we do.”
S. HARNAD “Let’s not quibble over how tiny is tiny.” But I wasn’t saying tiny, I was saying none. Computers don’t do any consciousness or feeling at all, what they do isn’t related to feeling. They are in the wrong ontological category to have that kind of causative power. Essentially, a computer is an idea, and a very complex and developed one, so it can hardly be the cause of simple, basic ideas, or feelings. B.RANSON: “The syntax in the associated computer is observer-dependent, but also the semantics and even the computer’s very status as a computer.”
S. HARNAD: “Syntax is syntax. The fact that it was designed by a person is not particularly relevant. What is definitely observer-dependent is the syntax of an ungrounded computer program. To free it from dependence on an external observer, it has to be grounded in the capacity for T3 sensorimotor robotic interactions with the referents of its internal symbols. Then the connection between its internal symbols and their external referents is no longer observer-dependent.” The grounding can’t take place because it doesn’t have any internal symbols, it doesn’t have any “internal” at all, except, again, observer-dependently. It’s the observer who decides what constitutes the computer, where it starts and finishes. Syntax isn’t syntax, not in itself, those symbols are only symbolic in the mind of the observer. The electrical currents and moving components are observer-independent, but their identification as symbols is observer-dependent. A feeling being does have an “internal”, there is the feeling, and the thing being felt. In the definition of “cognitive capacity” you kindly provided you say that this is what an “organism” can do. My Concise Oxford Dictionary defines “organism” as “an individual animal, plant, or single-celled life form, or, a whole with interdependent parts, compared to a living being. A computer or a computer plus a robot is not an organism, not because it doesn’t fit this definition, but because it isn’t an “individual”, or a “whole”, not in the way that matters. That is what is missing.
WHY AND HOW DOES IT FEEL LIKE SOMETHING TO OBSERVE?
(Reply to Bernie Ranson)
B.RANSON: “We know what causes feeling, it’s our nervous system”
Yes, but we don’t know how or why (and I’ve suggested reasons why we never will).
Our nervous system also causes our cognitive performance capacity, but there’s no reason to believe we won’t eventually be able to reverse-engineer that so as to give a causal explanation of how and why we can do what we can do.
The latter is called the “easy” problem; the former is called the “hard” problem (insoluble, by my lights).
B.RANSON: “[Feeling] is observer-independent… computation is observer-dependent”
Well, the former, hardly! But even if it were so, the problem is explaining how and why observers feel.
B.RANSON: “Computers don’t [feel] at all”
Agreed. So how and why do the kinds of things that *do* feel, feel?
B.RANSON: “symbols are only symbolic in the mind of the observer”
Agreed. And symbols are only grounded (i.e., causally connected to their referents) in a T3 robot (or higher); not in just a T2 computer. But that does not explain why and how it feels like something to have a symbol in mind.
B.RANSON: “A feeling being does have an “internal”, there is the feeling, and the thing being felt”
Internal to its body (or brain) is not quite the same sense of “internal” as in “having something in mind,” or as in “the thing being felt”…
B.RANSON: “A computer or a computer plus a robot is not an organism… because… it isn’t an “individual”, or a “whole” in the way that matters”
We agree about computers. Maybe we don’t agree about T3 robots. But the problem is not with being an “individual” or a “whole” but with explaining how and why an individual or a whole feels (if it does).
Stuart W. Mirsky
Response to Stevan Harnad’s Comments of 2/25/11 @18:32 We obviously disagree on the linguistic issue. If you think that words like “feeling” are used in precisely the same way when speaking of our experiences and of others’ experiences, then of course you will expect to be at a loss in ascertaining that what’s going on in me is what’s going on in you. Is this ontic or epistemic? The Other Minds problem puts us in an odd position of never being sure about a great deal of the world in which we find ourselves. But, of course, we ARE sure enough (as even you attest in your reports of how you would behave). So there is this great conundrum posed by the supposition that, to know that others have a subjective life like we do, we have to guess and guessing isn’t always reliable. Andl, yes, we can guess wrong — but to be able to guess wrong we have to know what counts as guessing right (or we cannot know when we have guessed wrong when we do it). So we’re stuck with a sense that we’re missing direct access to others’ minds which, if only we had, would make a difference. But, in fact, there is no difference needed. The issue of lacking direct access to other minds goes nowhere on a scientific level and Wittgenstein unpacks it philosophically so that it loses its force. If you don’t embrace that view, then I suppose it will continue to seem an obstacle. But it still has no more bearing on the scientific question of what it takes to produce a feeling mind than the absence of such indubitable certainty has on any other scientific question. Your wrote: “The scientific problem is not to be ‘assured’ that others are feeling, but to explain how and why they feel.” But a description of the processes that lead to the occurrence of feeling in another entity is enough to explain the how and the why (if the theory bears out empirically, of course); it’s no more a problem to test for feeling in a machine than in another human being. We look at the same kind of phenomena. Now, it just looks like we are each repeating ourselves though so perhaps we shall just have to accept that ours are markedly different understandings of what understanding is. You wrote: ” If a feelingless T3 or T4 robot is impossible then if T3/T4 behaves as if it feels, it must (somehow) be feeling. Fine. Now we know it feels. “Now: How and why does it feel? (Please don’t reply that the answer is that it would be impossible for it not to! That’s not a causal explanation.)” I have replied previously (in response to the challenge) that the how/why questions can be answered by a theory that proposes that feeling is just the occurrence of so many layered processes (of a computational type) performing certain functions in an interactive way — and I’ve spelled out the kinds of functions and layering I think such a system would need. Not in great detail, of course, because all one can do in discussions like this is speak in general terms. Actual specifics must come from actual system designers and implementers and, as there are many ways to do this, it follows that they might not all work (indeed, none may work). But here I am only required to say what WOULD suffice to provide an explanatory description (if it worked) and that is what the thesis I’ve proffered does. I think this all hinges on what may be the differing ways in which you and I are conceptualizing this thing you call “feeling”. On my view, when I consider my own subjective life introspectively, once I get past the obvious givens of being a self, of having awareness of things, of recognizing meaning and the like, I see nothing that cannot be broken down into more basic functions. That is, it seems to me this can be explained adequately in a systemic way, that “feeling” need not be presumed to be some special feature of the universe rather than the outcome of a lot of things a brain does. I have to conclude, given our failure to agree on this very basic issue that this is NOT how you see “feeling” at all. So our disagreement boils down to what may just be competing conceptions of mind. You wrote: “a theory (T3/T4) that accounts for behavior accounts for behavior. Why and how the behavior is felt needs an account of its own.” If feeling in others is always known through behavior and there are certain behaviors which, in the right context, represent or express feelings to us, then all we have to look for to test the theory is behavior in context, then a theory about what processes like those we run on a computer can do in the right configuration is testable. So one cannot plead non-testability. Of course, that’s a different issue from whether the theory can explain feeling as I think it can. You asked: “How have I shifted from scientific to metaphysical in asking for a causal explanation of how and why we feel? “How have I shifted from scientific to metaphysical in asking for a causal explanation of how and why we feel?” But I have already given such an explanation by proposing that mental features, say understanding, may just consist of certain processes (refer back to my road sign anecdote) which are plausibly performed by computational processes and brain processes. You have defined “understanding” as grounding + feeling requiring the robotic (T3) model. I have replied that, on my view, a T2 model can achieve sufficient grounding (an issue we’ve since left behind). But this doesn’t seem to be the real crux of our difference. Rather it’s that the feeling part in your definition, which Searle and you are (to my mind, rightly) insistent on, can be explained as the way(s) in which different subsystems within the overarching system interact, i.e., the self subsystem recognizes (through the same kinds of associative operations already alluded to) the occurrence of inputs captured and passed through various other distinct subsystems. This recognition is pictured as relational to the self subsystem, i.e., as with the other systems, it builds representations of its “observations”. I don’t argue that the brain isn’t complex or that a computer type platform wouldn’t have to achieve an equivalent complexity to do what brains do. I only argue that we can conceivably explain even the occurrence of feeling in this kind of systemic way. The reason I say you’ve shifted from science to metaphysics is because science depends on observations while you have imported an unsolvable metaphysical problem (because of the absence of observability) which only seems to muddy the waters. Of course, this doesn’t look like a real problem to me because of my Wittgensteinian orientation but, even if it does look like such a problem to others, it still has no bearing on what science needs to do in order to figure out what it is that brains do which produces consciousness. You added: “explain to me why and how it feels — or, if you like, why and how it’s impossible. [Assume I already agree that it's impossible: No contest.] No other-minds problem-problem. T3/T4 is feeling. It is indeed impossible for it not to be feeling. I get a headache just thinking about it. Now, just explain, causally, how and why it’s impossible.)” What I say is that it makes no sense to suppose that an entity behaving in a feeling way within a sufficient testing regimen isn’t feeling. I don’t deny we can have behaviors that fool us in limited situations. My denial is that it can be intelligible to suppose that an entity, ANY entity, that passes the lifetime Turing Test you have specified is missing the underlying feeling — because my position is that that’s all it means when we ascribe feelings to other entities. I agree that that ascription implies that we are presuming that the entities in question experience as we do. But that presumption is a valid one and poses no problem for cognitive science. Anyway, above I have offered a sketch of a theory for why and how an entity with feeling behaviors can be presumed to have feelings. My point is that the absence of direct access poses no problem for testing the theory.
You wrote: “There’s nothing public or social (or vague or difficult-to-pick-out) about feeling a migraine. And until it’s explained how and why it’s felt, the only way it can be described is as piggy-backing (somehow, inexplicably) on the causal mechanism of T3/T4 capacity.” There are different kinds of feeling, of course. A migraine is a physical sensation while the feeling connected to instances of understanding is something quite different. I have said previously that a machine mind might not feel as we do and by that I meant that it might not have physical sensation or, if it does, it might be nothing like our physical sensations. After all, its medium and sensory apparatuses will be of a very different type than ours. But the real issue you’ve raised is the one about feeling in instances of understanding. And that, as my experience on the road up from the Carolinas shows (I hope) is something quite different, i.e., it’s being aware of mental pictures and how they relate to other mental pictures. So I think we have to be very careful here, even with your word of choice, “feeling”, since it doesn’t always denote the same phenomenon. It’s clear we have very different views and I thank you for the chance you’ve provided for me to understand yours a little better — and the opportunity to raise some questions and offer my own view. I suspect that the reason we’re at loggerheads is that we have a very deep difference in how we think about consciousness.
FEELINGS ABOUT EXPLANATION
(Reply to Stuart W. Mirsky)
S.MIRSKY: “we’re missing direct access to others’ minds which, if only we had [it], would make a difference”
Not a difference to what I’m saying: I think readers will find this tedious, but all I can do is repeat it till it is taken on board: I am talking about explaining how and why we feel, not just how and why we do. And for that, it would not help even if we had a God’seye view of whether (or even what) others feel. As I said before: Knowing WHETHER (and WHAT) ≠ knowing HOW and WHY. So why are we again speaking about the other-minds problem — and whether?
S.MIRSKY: “a description of the processes that lead to the occurrence of feeling in another entity is enough to explain the how and the why”
Till further notice, the processes that lead to feelings happen to be processes that lead to doings (bodily doings in T3 and both bodily and brain doings in T4). To explain how and why we feel requires explaining how and why those doings are felt doings. Otherwise all you have is a correct, complete explanation of doings, with which the feelings are inexplicably correlated.
S.MIRSKY: “I have replied previously… that the how/why questions can be answered by a theory that proposes that feeling is just the occurrence of so many layered processes (of a computational type) performing certain functions in an interactive way — and I’ve spelled out the kinds of functions and layering I think such a system would need.”
And I’ve replied previously that this all sounds like a just-so story, not a causal theory explaining how and why we feel.
So, really, Stuart, we will need to stop repeating this. I understand that you feel you have a causal explanation, and I think you understand that I feel you do not.
S.MIRSKY: “I see nothing that cannot be broken down into more basic functions… “feeling” need not be presumed to be some special feature of the universe rather than the outcome of a lot of things a brain does. I have to conclude, given our failure to agree on this very basic issue that this is NOT how you see “feeling” at all.”
Correct.
S.MIRSKY: “feeling in others is always known through behavior”
Yes, feelings are known through behavior, but feelings are not behavior.
S.MIRSKY: “feeling… can be explained as the way(s) in which different subsystems within the overarching system interact…”
Nope, I don’t see this at all, and I really don’t think anything is gained by continuing to repeat it: There is no need to come back and revisit this point…
S.MIRSKY: “you’ve shifted from science to metaphysics… because science depends on observations while you have imported an unsolvable metaphysical problem (because of the absence of observability) which only seems to muddy the waters.”
And I thought all I’d said was that we clearly don’t just do but feel; but then it seems a perfectly reasonable and natural question to ask: how and why do we feel?
How is that a shift from science to metaphysics? Even my reasons for thinking the question will not be answerable do not invoke metaphysics. It just looks as if a causal theory of doing covers all the available empirical evidence, and yet it does not explain feeling. I’m not saying feeling is magic or voodoo: just that we can’t explain how and why we feel. We do feel. I’m sure our brains cause feeling, somehow: I just want to know how — and why (because otherwise feeling seems utterly superfluous, causally).
S.MIRSKY: “it makes no sense to suppose that an entity behaving in a feeling way… the lifetime Turing Test… isn’t feeling… because… that’s all it means when we ascribe feelings to other entities… I agree… we are presuming that [they feel] as we do… [A] machine mind might not feel as we do… it might not have physical sensation or… nothing like our[s]…”
This is beginning to sound a bit incoherent to me: We can tell that others feel. The lifetime TT is the way. But they may not feel as we do; they might not even have sensations at all. (Are we still within the realm of the “intelligible” here? I thought “behaving feelingly” was all it took, and all there was to it — on condition that the right internal subsystems interact, etc…)
S.MIRSKY: “But the real issue you’ve raised is the one about feeling in instances of understanding.”
No, the issue I raised was about feeling anything at all.
S.MIRSKY: “I suspect that the reason we’re at loggerheads is that we have a very deep difference in how we think about [feeling]…”
...and about explanation!
Bernie Ranson
B.RANSON: “[Feeling] is observer-independent… computation is observer-dependent”
S. HARNAD Well, the former, hardly! No really! That is an important part of the point I was trying to clarify. Feeling is observer-independent in the relevant sense, and if you think it isn’t then you are misunderstanding the intended meaning of “observer-dependent”. Yes, it is an inherent characteristic of feeling that it is felt by someone, but that is not observer-dependency in the relevant sense. [quote] But even if it were so, the problem is explaining how and why observers feel.[/quote] I don’t think there is a “why”. As for how, my point is that it isn’t anything to do with computation. That’s got to be a helpful contribution to the search for an explanation: if you are looking to explain the how of consciousness through computation, you are looking in the wrong place. B.RANSON: “symbols are only symbolic in the mind of the observer”
S. HARNAD “Agreed. And symbols are only grounded (i.e., causally connected to their referents) in a T3 robot (or higher); not in just a T2 computer. But that does not explain why and how it feels like something to have a symbol in mind.” But they are also only “grounded” in the mind of the observer. This grounding, the symbols, the connections with the referents and the referents themselves, all are observer-dependent. Wheras, any grounding that is involved in having a symbol in mind is not observer-dependent. B.RANSON: “A feeling being does have an “internal”, there is the feeling, and the thing being felt”
S. HARNAD: “Internal to its body (or brain) is not quite the same sense of “internal” as in “having something in mind,” or as in “the thing being felt”.” No, and I’m not talking about that sense, not talking about physical internality. I’m saying that the very occurrence of feeling creates the distinction we describe as “internal/external”. But because computers/robots don’t have any feeling at all, they don’t have the basis for that distinction.
B.RANSON: “A computer or a computer plus a robot is not an organism… because… it isn’t an “individual”, or a “whole” in the way that matters”
S. HARNAD: We agree about computers. Maybe we don’t agree about T3 robots. Perhaps we will, when you have taken the point of the observer-dependent/independent distinction. S HARNAD: But the problem is not with being an “individual” or a “whole” but with explaining how and why an individual or a whole feels (if it does). The point of nearly everything I’ve been saying, I think, is that it has to be an (observer-independent) individual or whole first. That is a major element of the answer to the “how” question. An organism that can be conscious somehow is the relevant (observerindependent) kind of “whole” or “individual”, and a computer/robot isn’t. That’s what is missing, that’s what we should be looking for. And it is missing in computers and robots, so we aren’t going to find it there.
“OBSERVATION”: FELT AND UNFELT
(Reply to Bernie Ranson)
This reply will illustrate yet another reason why it is so important to stick rigorously to the spare saxon word “feeling” — rather than any of the countless other elaborate and equivocal synonyms, euphemisms, paralogisms and redundant variants for “consciousness” — if we want to keep ourselves honest whilst trying to sort out what’s what, how and why.
In particular, one of the most common bits of self-delusion one falls into when one lets entity-names proliferate beyond necessity is the profound equivocation gap between “observation” and “*felt* observation.”
This is *exactly* the same thing as the (to my mind) nonsense that has built atop the incoherent distinction between (AC) “access consciousness” and (PC) “phenomenal consciousness.”
The difference between the two putative “consciousnesses,” AC/PC is immediately seen to be merely verbal (and vacuous), when one transcribes it thus: “access” vs. “felt access.” For unfelt access is no kind “consciousness” at all: it’s just data-flow — or, even more objectively, it’s just dynamics, event-flow.
You could say that a computer has access to signals from the Mount Palomar telescope; but that’s no kind of “consciousness” at all. And remains so until and unless the access becomes *felt* access.
So all you needed was the “felt”; the “access” has nothing to do with it. (So much for “AC”: and “PC” just becomes feeling!)
Ditto for “observation”: You can talk about “experimental observations,” but (as even the quantum foundationalists will tell you, in puzzling over their own “hard problem”), an experimental datum really only becomes an observation if/when a human observes it. And humans feel: It feels like something to observe. Otherwise there is only data-collection and data-processing going on — or, rather, just dynamics, events.
(I hope no one will take this as an invitation to digress into the alleged role of consciousness in the collapse of the quantum wave-packet: *Please* let’s not go there: We have enough koans of our own to contemplate!)
The reason for all this preamble is to prepare the way to pointing out that Bernie Ranson’s reliance on an “observer-dependent/ observer-independent” distinction is deeply equivocal when discussing feeling itself, and the causal status of feeling.
Yes, whether or not something is a chair or some other object we invent or use is observer-dependent; so is whether or not we agree to call it a chair. Yes, whether squiggles and squoggles mean my weekly salary, as calculated by a payroll program, is observer-dependent, and so is “salary.”
And, yes, whether or not “squiggle” refers to chairs (or anything at all) is observer-dependent, even when it’s grounded in a T3 robot.
But the buck stops with feeling itself. Feeling *is* observation. Whether what we feel is veridical is what we usually like to call an “observer-independent” truth. (Is it really raining outside, or am I misreading what I seem to see from my window? Do I really have something wrong with my tooth, or is my toothache just referred pain from conjunctivitis?)
But feeling itself is just about as observer-*dependent* as you can get, even though the fact that it *feels* like whatever it feels like is beyond any doubt (as canonized in the Cogito).
So with every feeling there are two questions we can ask: (1) Is there an external, observer-independent state of affairs that *is* like what this feeling makes it feel as if there is? (Never mind the problem of incommensurability here; interesting, but not really relevant; “reliably correlated” will do just as well as “resembles”; and, no, Wittgenstein on “private language” does not settle this matter either, because now we are talking about public language and public correlations.) That’s the thing we’re usually talking about when we talk about what we do or don’t have “access” to.
But then there’s the question: (2) Is there anything going on that feels like anything at all, whether veridical or not? That’s the real question of feeling. And it’s the causal basis of *that* that my persistent how/why questions keep insisting on. Not whether we have veridical access to facts about the world: The T3 story can take care of all that without any need of extra help from feeling. And that’s the point!
B.RANSON: “Feeling is observer-independent in the relevant sense, and if you think it isn’t then you are misunderstanding the intended meaning of ‘observer-dependent’”
Bernie, you seem to be getting a lot of intuitive mileage out of this notion of “observer-dependence/independence.” What’s at issue here, however, is not that, but whether the observation is felt or unfelt.
B.RANSON: “I don’t think there is a “why” [we feel]…
Well, it would be odd if something as ubiquitous as feeling did not have any neural or adaptive function, would it not? Is it not natural to ask why everything wouldn’t work just as well without it? (And by “work” I mean causality — everything for which T3 and T4 *can* explain how and why we do it. Not only does T3 and T4 explanation leave out the explanation of the fact that we feel, completely, but there does not seem to be any (independent, causal) room to include feeling, even if we wanted to.)
B.RANSON: “As for how… if you are looking to explain the how of [feeling] through computation, you are looking in the wrong place.”
We’ve already agreed on computation’s shortfall. Searle’s Chinese Room Argument shows (and the symbol grounding problem explains) that even if T2 could be passed by computation alone, cognition is not just computation, because computation alone leaves feeling out.
But even if you add dynamics and grounding (T3, T4), you still have not explained feeling. So if that’s the wrong place to look too, there’s no place else!
B.RANSON: “grounding… the symbols, the connections with the referents and the referents themselves… all are observer-dependent… [But symbols] are also only “grounded” in the mind of the observer… any grounding that is involved in having a symbol in mind is not observer-dependent.”
The capacity of a T3 robot to pick out apples when it tokens “apples” internally (along with all the other word/world interactions empowered by T3) is just the grounding of robot-internal symbols. Having a symbol “in mind,” in contrast, requires having a mind, i.e., feeling. That’s no longer just a matter of symbol-grounding.
And, frankly, the issue of “observer-dependence/independence” casts very little light on whether things are felt or unfelt (pace both Wittgenstein and Searle), let alone how or why.
(I think the misapprehension that the observer-dependence/independence distinctionmight somehow prove helpful here is, once again, a symptom of the free proliferation of equivocal synonyms, euphemisms, paralogisms and redundant variants for “consciousness” — in this case, “intentionality”: In a nut-shell: T3 grounding guarantees only unfelt “aboutness,” whereas having a mind requires felt “aboutness.” So, as usual, it is not intentional/nonintentional that is the “mark of the mental,” but felt/unfelt.)
B.RANSON: “the very occurrence of feeling creates the distinction we describe as “internal/external”. But because computers/ robots don’t have any feeling at all, they don’t have the basis for that distinction.”
There are, again, two (hence equivocal) “internal/external” distinctions. The unproblematic one (states occurring inside vs outside a computer, robot or person) and the problematic one (felt vs unfelt states).
And you prejudge the matter (and beg the question) if you assume that “robots” don’t feel, since robots are just autonomous causal systems, and hence surely feeling organisms are robots too.
(If by “robot,” however, you just mean a computer plus I/O peripherals, I agree that just a computer plus peripherals probably could not pass T3, let alone feel. The dynamics in the hybrid dynamic/computational T3 robot will probably have to be a lot deeper than just add-on peripheral I/O devices.)
B.RANSON: “The point of nearly everything I’ve been saying… is that it has to be an (observer-independent) individual or whole first. That is a major element of the answer to the “how” question. An organism that can [feel] somehow is the relevant (observerindependent) kind of “whole” or “individual”, and a computer/robot isn’t.”
I hear all the language about “observer-independence/dependence” but I am not getting any causal (how/why) insight (or even clues) from any of it…
Stuart W. Mirsky
Reply to Stevan Harnad’s response of 2/26/11 @ 22:02 S.MIRSKY: “we’re missing direct access to others’ minds which, if only we had [it], would make a difference” SH: Not a difference to what I’m saying: I think readers will find this tedious, but all I can do is repeat it till it is taken on board: I am talking about explaining how and why we feel, not just how and why we do. And for that, it would not help even if we had a God’seye view of whether (or even what) others feel. As I said before: Knowing WHETHER (and WHAT) ≠ knowing HOW and WHY. SWM: Then, if nothing but a God’s eye view would help, it’s not a scientific question and belongs in a different arena, namely that of metaphysics. Funny, though, but I have the same sense about this as you: that all I can do is repeat my point until it’s understood. But I’ve been down this path before and it doesn’t seem that repetition does any more good than argument. It has to do with our apparently holding very different conceptions of consciousness (or feeling or whatever term we choose to use for it). You speak of a difference between “knowing whether” and knowing why or how, as if these were simple distinctions. But there a lots of ways we can ask why and/or how. Why is the sky blue is a different kind of why question from, say, why is the earth the third planet from the sun or why do we love our parents or why did you drink the scotch instead of the bourbon or why do humans have the kinds of brains they do or opposable thumbs and so forth. So asking why certain kinds of processes produce a feeling is analyzable in more than one way. In the scientific way it’s very much answerable by hypothesizing that feeling itself is just processes doing certain things at bottom — and then determining which kinds of things must be done and what kinds of processes it takes to do them, how they need to be arranged — and then we implement the system that meets those criteria and test it. But if one starts with the assumption that feeling is not process, cannot be, then obviously THAT kind of hypothesis seems excluded and any explanation based on it seems like it must be the wrong kind. SH: So why are we again speaking about the other-minds problem — and whether? SWM: Because you bring it up elsewhere in relation to this (most recently in response to Mr. Ranson) and because the claim that we cannot see the other’s mind and therefore cannot explain it depends on that Other Minds problem presumption. But, of course, we don’t have to be able to directly access the minds of others to explain their occurrence (how they are produced and so forth). S.MIRSKY: “a description of the processes that lead to the occurrence of feeling in another entity is enough to explain the how and the why” SH: Till further notice, the processes that lead to feelings happen to be processes that lead to doings (bodily doings in T3 and both bodily and brain doings in T4). To explain how and why we feel requires explaining how and why those doings are felt doings. Otherwise all you have is a correct, complete explanation of doings, with which the feelings are inexplicably correlated. ~~~~~~~~~~~~~~~~~~ S.MIRSKY: “I have replied previously… that the how/why questions can be answered by a theory that proposes that feeling is just the occurrence of so many layered processes (of a computational type) performing certain functions in an interactive way — and I’ve spelled out the kinds of functions and layering I think such a system would need.” SH: And I’ve replied previously that this all sounds like a just-so story, not a causal theory explaining how and why we feel. SWM: I think this has to do with your conception of consciousness (though you call it “feeling”) which assumes it is a bottom line type of thing, not analyzable into the physical processes of the brains which produce consciousness. Because if you could see it as analyzable, could for a moment step away from the idea that it is some special feature of the world which requires a “God’s eye” view to access in all its manifestations, then I think you would have no problem with the idea that consciousness is unpackable into elements that are not, themselves, conscious. But I admit your view is very strong in all of us, a kind of basic intuition we have of ourselves. It is hard to shake off. SH: So, really, Stuart, we will need to stop repeating this. I understand that you feel you have a causal explanation, and I think you understand that I feel you do not. SWM: Yes. I don’t see how either of us can bring the other around. The difference in our views is deeply rooted in what I take to be competing intuitions. That is, we all of us have the intuition that consciousness is a unified irreducible something based on how we see our own experience. But I think there is another intuition at work, i.e., the one that tells us that the physical world is paramount, that we are elements in that world. What happens when these two intuitions meet, like opposing polarities that repel? Sometimes we get philosophy and usually one intuition proves dominant. The right move, though, is to find a way to balance both. S.MIRSKY: “feeling in others is always known through behavior” SH: Yes, feelings are known through behavior, but feelings are not behavior. SWM: I agree. But certain behaviors ARE consonant with what it is we mean we ascripe feelings to others in public space. S.MIRSKY: “feeling… can be explained as the way(s) in which different subsystems within the overarching system interact…” SH: Nope, I don’t see this at all, and I really don’t think anything is gained by continuing to repeat it: There is no need to come back and revisit this point… SWM: I repeated it because you repeated your challenge to me to offer an explanation even as you focused on my talking about how we recognize feeling. My view of how we recognize feeling (what we mean by its ascription) certainly informs my proposal as to how best to explain feeling. But they are not the same. Hence, I have found myself obliged to repeat my explanation periodically in response to your suggestion that I haven’t offered one. I grant that you do not find my explanation congenial and I have suggested why I think that is above. But my only point is that it IS an explanation and it is based on a series of points that can be elucidated here in a reasoned way. S.MIRSKY: “you’ve shifted from science to metaphysics… because science depends on observations while you have imported an unsolvable metaphysical problem (because of the absence of observability) which only seems to muddy the waters.” SH: And I thought all I’d said was that we clearly don’t just do but feel; but then it seems a perfectly reasonable and natural question to ask: how and why do we feel? SWM: A question to which I have proposed an answer (though my answer depends on understanding feeling as the workings of certain kinds of processes such as those found in brains — now if brain processes are capable of producing feeling it only remains for us to see how they do it; computational operations at some level are at least a reasonable possiblity). SH: How is that a shift from science to metaphysics? Even my reasons for thinking the question will not be answerable do not invoke metaphysics. It just looks as if a causal theory of doing covers all the available empirical evidence, and yet it does not explain feeling. I’m not saying feeling is magic or voodoo: just that we can’t explain how and why we feel. We do feel. I’m sure our brains cause feeling, somehow: I just want to know how — and why (because otherwise feeling seems utterly superfluous, causally). SWM: But if we can explain it physically, in terms of a given set of processes doing a given set of things, then we do know how and why. The only reason to think this kind of explanation couldn’t explain it is to suppose that feeling is inherently mysterious because it is a-physical. But there is no reason to think it is since it is never discovered apart from some physical platform (i.e., brains). Therefore there is good prima facie reason to look at the physical activities of brains. And if we do that, then unless we want to say that feeling springs like Athena from the head of Zeus into existence, the only other alternative is to suppose it is a composite of some things that are not, themselves, feeling. And what is more manifestly not-feeling than purely physical things? (Questions of panpsychism being placed on hold for the moment.) S.MIRSKY: “it makes no sense to suppose that an entity behaving in a feeling way… the lifetime Turing Test… isn’t feeling… because… that’s all it means when we ascribe feelings to other entities… I agree… we are presuming that [they feel] as we do… [A] machine mind might not feel as we do… it might not have physical sensation or… nothing like our[s]…” SH: This is beginning to sound a bit incoherent to me: We can tell that others feel. The lifetime TT is the way. But they may not feel as we do; they might not even have sensations at all. (Are we still within the realm of the “intelligible” here? I thought “behaving feelingly” was all it took, and all there was to it — on condition that the right internal subsystems interact, etc…) SWM: I carefully distinguished between feeling as sensation and feeling as awareness. This is where your insistence on the term “feeling” misleads. You have claimed that other words are “weasely” while “feeling” isn’t. But what makes them “weasely”, you have told us, is that they have different meanings (“awareness” suggests to you paying attention). But note that “feeling” doesn’t get off scot free in this analysis. A feeling can be a certain mood we find ourselves in, a particular physical sensation, or the awareness of things that comes with experiencing. “Feeling” has other uses, too, as in feeling our way in the dark, i.e., reaching out for familiar things, and so forth. The distinction I made in the text you quoted above from was feeling as in having blinding headaches (migraines) and feeling as in being aware of the information we had taken in (as in Searle’s sense of understanding what the symbols in the Chinese Room meant). S.MIRSKY: “But the real issue you’ve raised is the one about feeling in instances of understanding.” SH: No, the issue I raised was about feeling anything at all. SWM: As I point out above, there is feeling and there is feeling and the uses we put that word to don’t all mean the same thing. The feeling relevant to the Chinese Room scenario, which you have previously invoked, has to do with the sense of the man in the room that he gets it. Any of us may lack certain sensation capacities. Indeed, we may have some of those capacities deadened through anaesthesia, and yet still retain feeling qua awareness. I think it’s a mistake to fix to rigidly on a single term when dealing with this mental area because the referents are slippery here, because of their non-public provenance. We have to keep repeating and clarifying to avoid falling into confusion. S.MIRSKY: “I suspect that the reason we’re at loggerheads is that we have a very deep difference in how we think about [feeling]…” SH: and about explanation! SWM: That could be though I think it more unlikely. I suspect we would both agree on most uses of the term. But it’s clear we cannot find a lot of common ground on matters of mind.
Stuart W. Mirsky
Emendation: Ah, sorry, in the above I misread your reference to a “God’s eye view” so that part of my response should be disregarded. I won’t bother to re-write though as there is quite enough dialogue between us as of now. Unless it becomes an issue down the road I will leave this as the error it is and rely on the rest of what I said in my response to you. Thanks for the (hoped for) forebearance.
TO “BE AWARE” OF X IS TO FEEL X: THE REST IS JUST T3
(Reply to Stuart W. Mirsky)
(Stuart, could I ask you, please, if you continue responding, to pare down your quotes to just the necessary essentials, rather than quoting the entire verbatim dialogue each time, with a few small interpolated responses? Neither I nor ConOn readers can wade through all that text every time, especially after there has already been so much repetition. Quote/commentary is good, but please apply Occam’s Razor: It might even help focus thoughts…)
S.MIRSKY: “But if one starts with the assumption that feeling is not process, cannot be, then obviously THAT kind of hypothesis seems excluded and any explanation based on it seems like it must be the wrong kind.”
No causal hypothesis is excluded. “Feeling is process” is not a causal hypothesis — or if it is, it calls for a causal explanation: How/ why are some processes *felt* processes? (With your substitution, this would become “How/why are some processes *processed* processes?” That, I think lays the question-begging bare.)
S.MIRSKY: “I carefully distinguished between feeling as sensation and feeling as awareness. This is where your insistence on the term ‘feeling’ misleads.”
And, as I’ve replied many times, I think this is just equivocation on quasi-synonyms and the multiplication of pseudo-entities. Sensorimotor “processes” can be felt (seeing, hearing, smelling, tasting, touching) or merely “functed” (optical, acoustic, molecular, mechanical input processing), so can other internal states (emotion, motivation — which, if unfelt, are merely activations of various functional sorts and sign), including thoughts (e.g., what it feels like to think and mean that “the cat is on the mat”).
All feelings feel like something. If I am aware that I have a toothache or that the cat is on the mat, that just means I am feeling what it feels like to have a toothache or what it feels like to see or think or mean that the cat is on the mat.
It is not my insistence on the term “feeling” that misleads, but the long tradition of imagining (and talking as if) “feeling X” and “being aware of X” were two different kinds of things, rather than one and the same thing. (The rest is just about whether X is or is not true, and if so, how we access and act upon that datum; i.e., the rest is just T3!)
Stuart W. Mirsky
Reply to Stevan Harnad’s Comments of 2/28/11 @ 9:27 AM You have a point about paring down. I have been trying to walk a fine line, aware that failure to quote fully may leave something important you have said out — or lead to changes in meanings (as, I fear, I think some of your elided quotes from things I’ve said have sometimes done). But I am painfully aware of the downside of extensive quoting, especially in a forum like this, which you rightly point out. So I will try to be more selective in responding. I don’t get your point that it is “question begging” to suppose that feeling is explainable by treating it as processes and then to offer an explanation in terms of processes. After all, isn’t it equally question begging to suppose that it isn’t, and then denying any possibility of such an explanation on the grounds that it isn’t? In either case what’s happening is that an underlying assumption is being deployed to address a particular thesis, except that they are opposite assumptions. What’s really in play here is the disputed assumptions, no? Can a phenomenon like consciousness be explained mechanistically or must something else be invoked to explain its occurrence? The Challenge Redux Your challenge was for someone to explain how and why what your call “feeling” (and I prefer to call “awareness”) happens in a system like a T3. Your claim is that no one can do that. I responded by suggesting, first, that the distinction between the T2 and T3 standards is likely not essential to achieving the desired result because grounding can be done entirely within a symbols framework. Now we haven’t really dealt with that and probably don’t need to at this point because the main issue dividing us really is the one to do with “feeling” as you have rightly noted. So what’s really key here is the question of explaining feeling itself, the point of your challenge. I have argued that feeling can be explained computationally (whether that is the RIGHT explanation is a different question) since it can be determined to be present behaviorally, hence non-access to it in other entities poses no special problem in a testing regimen. Against my view, that computational explication is at least feasible, you have suggested that no amount of computational description can satisfy because all it can ever tell us is what observable events will result from whatever operations are performed by the computing (or any) mechanism in question. So here we have the sharp difference between us. I’ve said that observed operations (behaviors) are all that’s required to recognize that feeling is present in the behaving entity and therefore, on my view, a description of a process-based system that produces what we will both agree is feeling behavior, is enough to tell us how feeling occurs. But you have demurred, on the grounds (if I am reading you right) that feeling remains inaccessible to the outside observer so all we can ever know about the behaving entity is that it is operating LIKE a feeling entity. On your view, something IS still left out, namely the feeling of the feeling which only the particular observed entity itself can ever have — just as only we can feel our feelings. I think there is some confusion in this. If “feeling” is ascribed on the grounds of observed behaviors, then we know all we need to know. So nothing is left out. And if nothing is, then an explanation that describes how certain processes (whatever they are) combine to produce feeling behavior (meeting the lifetime TT standard) is enough. Such an explanation could be the wrong one. But the issue here is not whether it’s right or wrong but whether it could explain “feeling”. On the Matter of Terminology Here I think is where a lot of the difficulty in sorting out our opposing views may be found. I am aware that your claim that all words referring to the mental aspect of our subjective lives, except “feeling”, smack of equivocation. I profoundly disagree. As I’ve pointed out, your “feeling” works in the same way. We can agree to stipulate our meaning, whether for “feeling” or “awareness” or “consciousness” or “intentionality” and so on and so forth and this will surely help in any discussion, though it probably won’t prevent our slipping and sliding on the meanings nonetheless. That’s because words about the mental aspects of our lives, derived as they are from a public venue, are inevitably going to be slippery because they’re being used at a remove from that venue. Witness your own shifting between “feeling” meaning sensations (as in migraine headaches) and “feeling” meaning comprehending, as in the man in the Chinese Room having the feeling of knowing what the symbols he is reading mean. It is certainly fair to say both instances can be characterized as “feeling”, as you have done, but they are surely not the same thing. Moreover, I think it is a mistake to create a word like “functed” to substitute for “felt” when you want to avoid using the latter term. Insofar as “felt” seems to be called for, it’s arbitrary to make this substitution. And it makes little sense to deploy a word that has no real purpose of its own (if it had a real purpose our language would have already included it). If you just mean to substitute for words like “performed” or “did” or some other action word, then why not stick with those ordinary uses? You say “all feelings feel like something”. Well, that’s true by definition. Feelings are felt, of course. But the question isn’t whether it’s true (who denies it?) but whether any given system feels and that’s a different kind of question. We have seen that “feeling” can denote a wide array of features including the sensations we are accustomed to having (heat, cold, pain, pleasure, hard, soft, rough, smooth, bright, dark, dull, sharp, salty, sweet, sour, loud or barely audible, etc.). It also includes the states we find ourselves in (happy, sad, ebullient, morose, angry, desperate, etc.), and the senses we have of being awake (when we are) as well as a sense of understanding things such as geometry problems, English letters, Chinese ideograms and so forth. But just because we can refer to all of these phenomena as having feelings, broadly speaking, doesn’t mean they are the same thing. Indeed, I’ve suggested previously that any given entity that feels may have a wide variation in its feelings. That a T3-passing robot, which passes the T test on a comprehensive and open-ended (lifetime) basis, may lack certain equipment, or may have equipment of a different type than ours (and so have a different set of sensations or, perhaps, no physical sensation that is recognizable to us as such at all), says nothing about whether it will or will not have awareness in terms of understanding its various inputs. Feeling qua sensation is not the same as feeling qua understanding. So “feeling,” on examination, does not appear to be one simple thing at all but a wide array of different features that a system of a certain type may have. The word is fairly broadly deployed, as with most of our notational words so it can surely be as misleading as any of those other words you dismiss as “weasely”. What does it “feel like to think and mean that ‘the cat is on the mat’” as you have put it? Well what does it feel like to suddenly get the meaning of some words on a sign, as in my drive up from the Carolinas? I have already suggested that the latter event involved my suddenly having a range of mental pictures that differed from the previous ones I’d had. In recognizing those for what they were, I was feeling them, on your use.
This recognition was accompanied by another and different kind of feeling, in this case a sense of relaxation, i.e., I no longer had to keep rooting around for mental images that fit the words on the sign within the context and so I breathed a mental sigh of relief, you might say. I had a sense that something that had been unclear to me before was now clear, a feeling of cessation of perplexity — and, a kind of satisfaction in that cessation. As I had also recently been debating the implications of Kripke’s Meaning and Necessity for Wittgenstein’s meaning-as-use insight with a Direct Reference Theorist, my abrupt recognition of what that moment of understanding consisted of had the further effect of making clear to me why my DRT interlocutor was wrong about something he had said. So I had yet another moment of satisfaction, too, feeling that I had solved a difficult problem that was of more longstanding concern to me than the momentary confusion on the road. So lots of feelings occurred in me at the moment of sign recognition and, while related in various ways, they were not all the same, despite the use of a single word to name them. Your point has been that feeling is feeling and I want to suggest that it is not, that we use the word “feeling” for many things and so it’s no help to jettison the other mental words (to which you object) in favor of the one you don’t object to on the grounds that your choice alone is clear. It isn’t any clearer than its sibling words though our coming to some kind of stipulative agreement about using it (or one of its alternatives) can certainly aid us in this particular discussion. Conclusion My main point, then, is that we don’t really need anything more than a process-grounded account to explain the occurrence of feeling, even if we don’t yet have an entirely satisfactory one we can accept (the jury is surely still out on the kind of theory I’ve sketched here). And holding out for something more just moves us outside the scope of scientific inquiry. It sets up an insoluble problem for the scientific investigation of consciousness and thinking, a problem which there is no reason to expect science to solve — or to suffer from not solving.
AGREEING TO DISAGREE
(Reply to Stuart W. Mirsky)
S.MIRSKY: “Feeling behavior…”
No such thing. Just behavior, simpliciter (possibly correlated with what it feels like to do that). Your Carolina highway experience makes you feel you’ve explained that via “processes,” but alas it doesn’t make me feel that way. I think those processes just explain behavior and feeling remains to be explained (and probably can’t be, for the reasons I’ve suggested).
S.MIRSKY: “It is certainly fair to say both [headaches and understanding] can be characterized as “feeling”, as you have done, but they are surely not the same thing.”
I didn’t say all feelings were the same; I just said they were unexplained.
Our exchanges are really just repetition now. I apologize but I won’t be able to reply again unless something new and substantive is said. Thanks for the exchange.
IGNORANCE OF AN EXPLANATION DOES NOT ENTAIL LACK OF AN EXPLANATION (reply to Harnad)
SH: “But I don’t even believe in zombies!” If you think that you can have a T4 robot that lacks feeling then you do believe in (the conceivability of) zombies!
SH: The (“higher-order”) explanation (now transcribed here minus the synonyms, euphemisms, paralogisms and redundancies) seems to be: “to feel (something) is to feel something. This… explains why it feels like something …because that is how my feeling feels, and that is all that there is to feeling.” Pared down to this, it’s definitely not hermeneutics; it’s tautology. Put back the synonyms, euphemisms, paralogisms and redundancies and it becomes hermeneutics: a Just-So Story. But it’s certainly not explanation!”
If one were inclined to be uncharitable, one could parody your view along the lines of ‘feeling can’t be explained because feeling is unexplainable’. Luckily, I am not so inclined At any rate these are not the same thing! On the one hand we hand we have being conscious OF something in our environment and on the other hand we have feeling. You are right that being conscious of something in the relevant sense is something that can be functed and does not have to be felt. On the higher-order view one has feeling when one is conscious of oneself as being in a toothache state or thinking that the cat is on the mat. So, when one has those states without being conscious of them, there is no feeling. When one is conscious of being in those states there is feeling. This does not amount to ‘feeling my feeling’ or some other tautology since the relevant way of being conscious of is just the same one that figured in the story earlier. So, when you funct that you funct in the right way you get feeling is something like what the theory says. You keep reading things into ‘conscious of’ that aren’t there. At this gross level that doesn’t look like an explanation. The ‘a ha’ moment comes from the details, if it comes at all. If you want to see what I take to be the basic argument for the HO approach you can here: HOT Qualia Realism. I will just say here that we know that one can be conscious of something in the relevant functing sense without one being conscious of one’s so functing from subliminal perception, masked priming, etc. So the basic idea is when you have that kind of thing directed at one’s own mental functioning you get feeling. The argument for this starts by considering winetasting type examples. We know from these kinds of cases that acquiring new concepts leads to new qualities in one’s phenomenology. So it is not out of the question that we have a funct-thought to the effect that some mental quality is present and this results in our feeling. You may not think that this is true, but why not? It cannot come from the kind of considerations that you have developed so far. You must show that there is something wrong with the explanation. But what is it? Just so we are all on the same page here, I am not claiming that the HOT theory IS true or even that I THINK that it is true. Rather what I want to claim is that it COULD be true and IF IT WERE true it would provide an explanation for feeling. This is enough to show that the argument you have presented doesn’t work.
WHY “HOT” LEAVES ME COLD
(Reply to Richard Brown)
R.BROWN: “Ignorance of an explanation does not entail lack of an explanation”
True. But I did not just say we lack an explanation; I also gave some reasons why we lack one, and are unlikely to be able to find one. (T3/T4 explanations work identically well whether or not there is feeling; and there is no room for assigning feeling an independent role in any causal explanation — except through psychokinesis, which is false.)
R.BROWN: “If you think that you can have a T4 robot that lacks feeling then you do believe in (the conceivability of) zombies!”
I don’t believe in zombies, and both my suggestion that we do not have a causal explanation of how and why there cannot be zombies (which would be equivalent to a causal explanation of how and why we feel) and my suggestions as to why we will never have one are independent of whether I believe in zombies.
As to the “conceivability” of zombies: I’d rather not get into that, and I don’t think I need to.
I understand what empirical evidence is; I understand what deductive proof is; and I understand what plausibility arguments are. But I do not understand what a “conceivability” argument is. I can “conceive” Meinongian objects (my images of them are rather Escherian), but I cannot construct them, because they are logically impossible. That’s worse than whatever might be the reason there cannot be zombies: we certainly don’t know what that reason is, otherwise we’d have an explanation of why and how we feel! I also don’t believe in psychokinesis, but I’m not sure you would want to argue that it’s inconceivable. (In fact, I’ll wager that most people can not only conceive it, but believe it!) Now if psychokinesis had really existed (along with the requisite independent evidence for it — a detectable fifth causal force in the universe, empirically testable), then it would be easy to explain how and why we feel, as well as to explain how and why there cannot be zombies: because zombies would lack psychokinesis. There would even be ways of empirically distinguishing — through ordinary reverse-engineering (the “easy” problem) — the presence or absence of feeling on the basis of the presence or absence of the psychokinetic force. And hence it would be possible to explain, causally, why, if a robot lacked psychokinesis, it could not manage to pass T3: It could *do* this but not that.
But there is no psychokinesis. So whether or not it is “conceivable” that there could be zombies, we are powerless to say how or why not.
R.BROWN: “one could parody your view along the lines of ‘feeling can’t be explained because feeling is unexplainable’”
Nope. Feelings cannot be explained for the reasons I’ve already mentioned. No tautology. (T3/T4 explanations work identically whether or not there is feeling; and there is no room for assigning feeling an independent role in any causal explanation — except through psychokinesis, which is false.)
R.BROWN: “On the one hand we hand we have being conscious OF something in our environment and on the other hand we have feeling. You are right that being conscious of something in the relevant sense is something that can be functed and does not have to be felt.”
Consciousness is *felt* access. Without the feeling, access is not conscious. So drop the “conscious” and just worry about explaining feeling.
R.BROWN: “On the higher-order view one has feeling when one is conscious of oneself as being in a toothache state or thinking that the cat is on the mat. So, when one has those states without being conscious of them, there is no feeling. When one is conscious of being in those states there is feeling.”
What a complicated story! Is it not fair to say that I feel the states I feel, and I don’t feel the states I don’t feel? (The rest is just about what I feel.”
I can’t get my head around notions like “being in a toothache state… without [feeling that I'm in a toothache state]".
Despite the mirror-tricks and other fun it allows, I do not find higer-orderism any more useful than any of the other synonyms and paralogisms. Speech-act theorists may get some mileage out of “knowing that I know that you know that I know…” but insofar as feeling (and hence consciousness) is concerned, “feeling that I feel” is just feeling. Explain the bottom-order one, and you’ve explained it all; leave it out or take it for granted and you’ve explained nothing.
R.BROWN: “This does not amount to ‘feeling my feeling’ or some other tautology since… ‘when you funct that you funct in the right way you get feeling’ is… what the theory says…. At this gross level that doesn’t look like an explanation. The ‘aha’ moment comes from the details, if it comes at all.”
For me, it doesn’t come at all (and it *does* sound like tautology, dressed up in higher-order just-so hermeneutics…)
R.BROWN: “we know that one can be conscious of something in the relevant functing sense without one being conscious of one’s so functing from subliminal perception, masked priming, etc.”
That’s not consciousness, it’s access (detection), and it is indeed just functing, if it’s unconscious.
R.BROWN: “So the basic idea is when you have that kind of thing directed at one’s own mental functioning you get feeling.”
“That sort of thing” is so far just unconscious access to data. To make it mental, you have to make it felt. And then you have to explain how and why it is felt, not just say it is so.
R.BROWN: “consider… winetasting… acquiring new concepts leads to new qualities in one’s phenomenology.”
That’s my research area: category learning and categorical perception. Learning new categories (new ways to sort things) can alter their appearance: Within-category differences are compressed and between-category differences are enhanced.
But that’s all felt (though our models for it are just neural nets). Plenty of reason why it would be adaptive to make consequential differences more discriminable — but not a clue of a clue why any of that should be felt, rather than just functed.
R.BROWN: “So it is not out of the question that we have a funct-thought to the effect that some mental quality is present and this results in our feeling. You may not think that this is true, but why not? It cannot come from the kind of considerations that you have developed so far. You must show that there is something wrong with the explanation. But what is it?”
Shall I count the proliferating equivocations? What is “mental quality” as opposed to just a property? That it feels like something to perceive it (e.g., red, or the fruitiness of Beaujolais). Being able to recognize things as red, fruity or Beaujolais is one thing, and that’s just functing (T3). Feeling what it *feels like* to see red, to taste something fruity, or to recognize something as Beaujolais is another, and it’s not just that your explanation is wrong — it doesn’t even touch the problem!
R.BROWN: “I am not claiming that the HOT theory IS true or even that I THINK that it is true. Rather what I want to claim is that it COULD be true and IF IT WERE true it would provide an explanation for feeling. This is enough to show that the argument you have presented doesn’t work.”
I think not. I think the HOT theory is no explanation at all. It doesn’t explain how or why some processes are felt rather than just functed: It projects a mentalistic interpretation on them, and then is read off as if it had been an explanation.
Joshua Stern
SH (@February 26, 2011 at 20:43): The sensorimotor I/O for T2 is trivial, and nothing substantive hangs on it one way or the other. Now wait a minute! The classic argument against computation, that Searle makes and you repeat here in various ways, is that computation “is just formal”. The “trivial” aspect of T2, that it is in some modest way physical, sensorimotor, is thereby _hugely_ important.
I suggest it shows that the formality argument is moot. It also threatens the entire grounding argument, which is much stronger as a matter of type than a matter of degree. You might agree with that. But most of all, it suggests we have not yet located, enumerated, or addressed whatever prevents us from passing the modest T2 test. The CR _assumes_ something passes the T2 test. I suggest that, until that is actually done, we don’t know how to even begin to talk about T3 or T4 tests. And once it is done, we will find there is no need for T3 or T4 tests. – Various implications flow from this, some of which I have outlined. But would you not also agree, whatever its merits or faults, that the CR is a claim against type? – Just a note, that in all of my posts here I am emphasizing differences, because I believe they matter. There are many points at which any two out of three of Harnad, I, and Searle may agree, and even a few where all three may agree. For example, we all agree that a (computational) system may pass the T2 test without it representing real cognition, real consciousness. However, we reason very differently from that point on, and perhaps the greatest points of contention are directly adjacent to the points of agreement – as is so commonly the case. What if a system *does* pass T2? I suggest we all still have issues. First, we have the issue of other minds. Then we have the issue of Humean skepticism. Then we have various sorites arguments about whether it really passed, or not. Then we have issues of attribution versus realism, and on which side must mechanism fall? None of these go away with T3 or T4 tests. None of them are easy, or easily discussed in brief. Again, what if a system does pass T2? Some say that is not enough. Well, and yet, it seems a remarkable challenge that has not yet been met. My focus remains on considering how it might actually be done, what it must entail if it is to be done, and what it must imply if it were done – and when it is done.
UNFINISHED BUSINESS
(Reply to Joshua Stern)
J.STERN: “The ‘trivial’ aspect of T2, that it is in some modest way physical, sensorimotor, is thereby _hugely_ important. I suggest it shows that the formality argument is moot.”
I’m afraid I can’t agree. I/O is already part of the Turing machine, but not in the sense of robotic dynamics; just enough for 1/0 in and 1/0 out.
J.STERN: “What if a system *does* pass T2? I suggest we all still have issues… other minds…. Humean skepticism…. sorites arguments … attribution versus realism… None of these go away with T3 or T4 tests”
If a system passes T2, or T3, or T4, you still have no explanation of why and how it feels (if it does). And that’s without any recourse to the other-minds-problem, Humean skepticism, sorites, or realism…
But the problem of grounding, at least, goes away with T3…
“T3/T4 explanations work identically well whether or not there is feeling; and there is no room for assigning feeling an independent role in any causal explanation — except through psychokinesis, which is false” This is false as the fact that the HOT theory is possible shows but to see that you would have to come to understand the theory in detail. I recommend Rosenthal’s 2005 book Consciousness and Mind.
That sounds a bit harsher than I intended. I am not trying to be glib, the point is that there are answers to your questions but they can only be answered by taking the theory seriously and you seem set to dismiss it from the outset based on a priori considerations about what you think is causally superfluous in a T4 machine.
HERMENEUTIC SKYHOOKS
(Reply to Richard Brown)
R.BROWN: “there are answers to your questions but they can only be answered by taking the [HOT] theory seriously”
My questions are simple; if the answers are complicated and book-length, the questions have not been understood and the answers are answers to something else.
Besides, I’ve already understood enough of higher-order approaches to see that they beg the question: awareness of toothache, or awareness of awareness of toothache are merely felt states, like toothache itself. Account (“bottom-up”) for feeling, and you’ve accounted for it all: Try to account for it top-down, with a higher-order something taking a lower-order something as its object (of “awareness”), and declare that “consciousness” and you are hanging from a hermeneutic skyhook.
If the job can be done, it can be done speaking only about feelings — no synonyms, substitutes or skyhooks.
Stuart W. Mirsky
Reply to Stevan Harnad’s comments of 2/28/11 @ 23:09 Some behavior of things is feeling driven, some not. Thus it makes sense to distinguish between “feeling behavior” and behavior which isn’t. The Carolina experience involved no overt behavior on my part, just thoughts which, in your terminology, are felt. After consideration, it seemed to me perfectly reasonable to understand the thinking that occurred, in relation to that sign reading, in terms of what a computer could have done. From the connection of the words to particular representations, to the connection of those representations to other representations, all in terms of mental images (which are, when you think about it, rather different from visual images), everything that happened could be replicated in a computer format. Nowhere, on examination, do I see any special feature that somehow lies outside the kinds of physical processes computations instantiate. Thus, I concluded that my brain processes, which underlay the stream of experience I had at the moment of getting the meaning, could reasonably be replicated computationally, from the picture of the world in which the sign existed, to the historical associations in which the meaning of the sign’s words were found, to the sense of self that found and anchored the meanings in the array of images generated. But yes, this surely comes down to our differing ways of seeing our private mental lives. There is a sense in which one’s experiences look like the whole ball of wax and another in which they look veridical in some special way that sets them apart from everything else apprehended within those experiences. There is a sense that being aware (my version of your “feeling”) seems to be a special phenomenon among all the vast panoply of phenomena we experience. But should we take this at face value? Science has often taught us that things aren’t always what they seem and are often quite a bit more complicated and/or strange. I understand your feeling that we have reached the end of our exchange here. In fact, I think we reached it a while back so I understand your reluctance to continue. There comes a time when we have dug as deeply as we can, when we hit bedrock and can dig no further. I appreciate the time you’ve taken and the opportunity you afforded me here to discuss your thinking with you first hand and I certainly won’t be offended if you don’t reply further. We shall just have to agree to disagree.
Stuart W. Mirsky
Addendum to my last response to Stevan Harnad: I meant to address this but slipped up You wrote: “I didn’t say all feelings were the same; I just said they were unexplained.” What you said, which prompted my response delineating a wide array of feeling referents, was that “feeling” was a better (as in less confused) term for what we mean by consciousness than other so-called “weasely” words like “consciousness”, “intentionality”, “awareness”, and so forth. My reply that “feeling” refers to a wide range of different things was to show that your choice of THAT word is no better than others’ choices of words like “subjectness” or “awareness” or “intentionality”, etc. All have different uses and referents, which may prompt confusion. Thus none are less “weasely” in this sense than the others. By responding that you never said all feelings were the same (only that they were unexplained), I’m afraid you missed the point I was making. Given the alleged unexplainability you say characterizes what you call “feeling”, my point was that since feelings, given their range of types, COULD be explained in different ways, it was conceivable that a robot intelligence could have one kind of feeling but not another. That a robot intelligence might not have migraines (as you put it) would not then be evidence it did not have the kind of feeling(s) associated with understanding. If understanding can be explained computationally (which I’ve argued it can be), then the absense of physical sensation feeling in a robot does not pose a problem for this kind of explanation and the challenge you posed at the outset could be said to have been met. Of course we have already agreed that agreement that it has been met rides on a deeper agreement about what feeling, itself, is (to explain something we have to know what it is we’re explaining). And here we seem to be in very deep disagreement, i.e., I see “feeling” (awareness) as reducible to non-feeling constituents like physical processes whereas you see a gap that cannot be bridged. Note that given this deep level disagreement, it follows that getting clear on what your term “feeling” denotes was important enough to warrant some discussion. Anyway, as long as I’m adding this, I just want to say that I noticed a small error in my last post to you which I might as well correct here for the record. At the end of the third paragraph in my preceding response to you I wrote: “Nowhere, on examination, do I see any special feature that somehow lies outside the kinds of physical processes computations instantiate.” Of course, I should have written “outside the kinds of physical processes which instantiate computations” or some equivalent. I got careless. Otherwise I believe the points made above remain sound. Again, thanks for the opportunity to engage on this. I’ve found it helpful.
Joshua Stern
J.STERN: “The ‘trivial’ aspect of T2, that it is in some modest way physical, sensorimotor, is thereby _hugely_ important. I suggest it shows that the formality argument is moot.” SH (02/28/2011 16:17): I’m afraid I can’t agree. I/O is already part of the Turing machine, but not in the sense of robotic dynamics;
just enough for 1/0 in and 1/0 out. Well, I agree with that (I think), but you have to realize that many people say things like, “The Turing Machine is just an abstraction, it works by just syntax, it has no physical realization,” and the related argument, “Turing Machines have infinite tapes, you don’t, so obviously all computation is just an abstraction.” They would hold that I/O of 1/0 is just abstract or formal. Physical and causal are the points here. If you and I (at least) recognize computation as physical, we have departed from Searle in what I consider to be a hugely significant way, even at the T2 stage (even before the T2 stage!). You believe there is more to be said by expanding on the physical aspect to ground symbols, meaning, … something. I understand the move, but put it in a different game. If you want to understand why any string in a T2 has a certain meaning, you need to relate it to something in the real world. In most philosophy this is done by some variety of hand-waving about reference, correspondence, truth, rigid designators, acquaintance, etc. Your “grounding” combines many of these to a physical particular. I salute all of this. However, as I have (briefly) argued, I believe the crux of the matter stays within the computational aspect, once that is seen to be physical and causal. J.STERN: “What if a system *does* pass T2? I suggest we all still have issues… other minds…. Humean skepticism…. sorites arguments … attribution versus realism… None of these go away with T3 or T4 tests” SH: If a system passes T2, or T3, or T4, you still have no explanation of why and how it feels (if it does). And that’s without any recourse to the other-minds-problem, Humean skepticism, sorites, or realism… But the problem of grounding, at least, goes away with T3… I suggest T2 is necessary and not sufficient, but most of all I note that it is very difficult, at least circumstantial evidence that it may also be very significant. Let me see if I have your position right on this. You have posited that perhaps no computer could pass T2, but what about a human? If we judge that something has just passed T2 and it turns to be a human, you would say that is because the human was grounded, had the right stuff – even if all we ever knew of the contestant was the “ungrounded” symbols coming across the teletype. Yet I, at least, stipulate that some kind of really dumb computer system might pass T2 and still not be intelligent because it lacks the right stuff – though my right stuff is not your right stuff. Certainly *Searle* has conceded this, and the compsci/ AI contingent takes it as a matter of faith – “We can fake it well enough to pass your tests.” What is it that differentiates the fake that passes the test, from the “real” that passes the test, that is, what *could* it be that differentiates the one from the other? Won’t you simply say, “Oh, the programmers, who are human, privileged, and grounded, finally found a way to capture grounding and include it in the program”? One can continue along these lines, but the bottom line for me always comes down to the fact that Turing saw right through all of this in his combined 1937 and 1950 papers. Searle’s CR expressed some bogus doubts but sets up the one demand that Turing did not address, because neither of Turing’s two papers was constructive. The CR demands a *constructive* answer. Searle suggests there cannot be one, but that is fluff. Searle bases this heavily on the idea that computation is the wrong sort of stuff, not even physical. You and I have agreed (I think) that computation is physical – this might be a neo-computationalism, but if so, I’m happy with it, and have offered a few points of argument for it above. Let me finally offer a few notes of support for T3. There is an ancient question, which is, “We know there is nothing essential in this squoggle of ink on paper, so how is it that it comes to mean ‘cat’?” Some answer needs to be made to that, in simply linguistic terms, even if we are speaking only of humans, not of computers. Some empirical story needs to be told involving sensorimotor details, also computational details. Some number of agents have to have some number of conventions about interpreting sqoggles as some reference to distal objects. T3 respects *all* of that. Excellent. T4 would be another matter entirely, I will skip here rather than launch another topic. Anyway, I will address it from the other side right here. – SH: “T3/T4 explanations work identically well whether or not there is feeling; and there is no room for assigning feeling an independent role in any causal explanation — except through psychokinesis, which is false” RB (@ 2/28/2011 16:53): This is false as the fact that the HOT theory is possible shows but to see that you would have to come to understand the theory in detail. I recommend Rosenthal’s 2005 book Consciousness and Mind. I just want to put my vote in with RB here, and whether this is exactly his theory (or wildly at odds with it), I would assert that the feelings (qualia, phenomenalism) can fit within a physical and causal framework, something along the lines of a HOT in no way precludes physicality or causality individually or together, and indeed may teach us more about just how such physical and causal systems need to be constructed.
LOOSE ENDS
(Reply to Joshua Stern)
J.STERN: “I suggest T2 is necessary and not sufficient”
Right you are, since T2 is a subset of T3. (But I also think being able to pass all of T3 is necessary in order to be able to pass T2.)
J.STERN: “You have posited that perhaps no computer could pass T2, but what about a human?”
We’re looking for (causal) *explanations* of T2 capacity (reverse-engineering), not just examples of it. The model’s capacities have to be indistinguishable from those of a real human — but it has to be a model that we built so we can explain how it can do what it can do: Humans are “Turing-indstinguishable” from one another, but so what, since we have no idea how they pass the test?
J.STERN: “What… differentiates the fake that passes the test, from the ‘real’ that passes the test?”
Feeling. But the whole point of the test is that once you can’t tell them apart, you can’t tell them apart. (And I would not call any robot that can pass the T3 for a lifetime a “fake”!)
J.STERN: “[TT] demands a constructive answer.”
Yup, you’ve got to build it, so you know how it can do what it can do. Otherwise there was not point…
J.STERN: “You and I have agreed (I think) that computation is physical”
I don’t think anyone would seriously disagree that computation (software) has to be physically implemented in a dynamical system (hardware), otherwise it’s just inert code on paper. (But the code is still implementation-independent: that’s just the hardware/ software distinction.)
J.STERN: “I would assert that the feelings… can fit within a physical and causal framework, something along the lines of a HOT…”
Hermeneutics always fit; they just don’t explain…
Bernie Ranson
Hi Stevan, Unlike some of the other contributors I don’t feel we have drilled down to the bottom of our disagreement, it feels to me like we haven’t penetrated the surface. For example you say that “feeling itself is just about as observer-*dependent* as you can get”; but that is clearly not the case if we use the term “observer-dependent” in the way I have tried to explain and exemplify. It’s not fair to say that you aren’t getting anything out of the observer-dependency talk if you aren’t thinking in the same language it uses. The metal in coins and the paper in notes are observer-independent. They will continue to exist whatever anybody says or thinks about it. Consciousness is observer-independent in that sense. “Money” only exists when somebody says it does. That’s what observer-dependent means here, and consciousness isn’t like that. Yes, feeling is a special case because of the way we seem to observe our own consciousness, but that is really not important here, and it isn’t what “observer-dependent” means here. In the introduction to his book Mind (2004) Searle says “There are two distinctions that I want you to be clear about at the very beginning, because they are essential for the argument and because the failure to understand them has led to massive philosophical confusion. The first is the distinction between those features of a world that are observer independent and those that are observer dependent. … we also need a distinction between original or intrinsic intentionality on the one hand, and derived intentionality on the other. For example I have in my head information about how to get to San Jose. I have a set of true beliefs about the way to San Jose. This information and these beliefs are examples of original or intrinsic intentionality. The map in front of me also contains information about how to get to San Jose [...] but the sense in which the map contains intentionality [...] is derived from the original intentionality of the map makers and users. [...] These distinctions are systematically related: derived intentionality is always observer-dependent. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Steve, you said that the observer-dependent/independent stuff didn’t seem to cast much light on the question you are interested in. The thing is, it doesn’t cast a light, it casts a shadow, it shows you something negative, it shows you that a robot can’t have feeling any more than a computer can, because the things that you invoke that are going to give it feeling, its sensors and actuators say, fall into the observer-dependent category just as much as the parts of a computer do. Therefore the intentionality they have is derived intentionality.
S. HARNAD: “There are, again, two (hence equivocal) “internal/external” distinctions. The unproblematic one (states occurring inside vs outside a computer, robot or person) and the problematic one (felt vs unfelt states).” Well no, the “inside/outside a computer” is problematical. Because in that case “inside and outside” are observer-dependent. Whereas, with a conscious being what we are describing as “inside and outside” does have an observer-independent existence: “inside” refers to feeling.
S. HARNAD: “And you prejudge the matter (and beg the question) if you assume that “robots” don’t feel, since robots are just autonomous causal systems, and hence surely feeling organisms are robots too.” But robots aren’t autonomous systems. That’s entailed by their observer-dependent status. Something is only part of a robot when somebody says it is.
S. HARNAD: “If by “robot,” however, you just mean a computer plus I/O peripherals, I agree that just a computer plus peripherals probably could not pass T3, let alone feel. The dynamics in the hybrid dynamic/computational T3 robot will probably have to be a lot deeper than just add-on peripheral I/O devices.” The dynamics in the T3 robot will be observer-dependent, its intentionality derived. You aren’t adding the relevant thing when you add deeper dynamics, it’s just more of the same. The causes of feeling are observer-independent. You’ve been asking us to identify what is missing from a robot that stops it feeling:
well this is the answer. It’s the wrong category of phenomenon, it doesn’t have the appropriate causal properties. And this isn’t just intuition, it’s hard practical fact. That is shown by the multiple realisability of computation for example. Consciousness is caused by very specific physical processes. It isn’t reasonable to believe that it is also going to be caused by innumerable other quite different processes. Specific physical processes in the nervous system cause feeling. They are related to, and a development of, physical processes in the nervous system that cause unconscious behaviour. It seems quite plausible to me that both these levels build on a still lower level. What I mean is that perhaps the missing thing we are looking for is whatever makes an autonomous organism autonomous. Perhaps that thing comes along very early in the development of life, maybe bacteria have it. Or maybe the autonomy only comes along with, and as an inevitable consequence of, feeling.
ON INDEPENDENCE AND AUTONOMY
(Reply to Bernie Ranson)
B.RANSON: “‘Money’ only exists when somebody says it does…”
And feeling only exists when somebody feels it…
But, never mind; let’s see where you want to take the suggestion that feeling is observer-independent:
B.RANSON: “[Searle-Distinction 1] between those features of a world that are observer independent and those that are observer dependent… [Searle-Distinction 2] between original or intrinsic intentionality on the one hand, and derived intentionality on the other… derived intentionality is always observer-dependent”
“Intentionality” is another of the weasel-words. If “intrinisic” intentionality’s not just T3 grounding, then it’s “derived” if it is unfelt, and “intrinsic” if it is felt.
A book’s and a computer’s “intentionality” is not only unfelt (hence “derived”), but it is also ungrounded (no connection between internal symbols and the external things they are about). Sensorimotor grounding remedies that in a T3 robot — but it is still unknown whether (and if so, it is unexplained how and why) T3 grounding also generates feeling. That’s the “hard” problem.
B.RANSON: “a robot can’t have feeling any more than a computer can, because… its sensors and actuators…[are likewise] observer-dependent [like the] computer… Therefore… they have [only] derived intentionality”
Unfortunately these are just expressions of your beliefs about “robots.” It’s not even clear what you mean by a “robot.” What I mean is any organism-like autonomous causal physical system (i.e., the ones having T3 or T4 capacities like those of organisms): That includes both naturally evolved organism-like autonomous causal physical systems (i.e., organisms) and bioengineered or reverse-engineered ones (e.g., T3, T4).
The systems are autonomous, so observers’ interpretations have nothing to do with what they are and what they can and can’t do.
B.RANSON: “robots aren’t autonomous systems… [they are] observer-dependent… Something is only part of a robot when somebody says it is.”
I think you’re wrong on this one, on anyone’s sense of “autonomy.”
(And, as I said, Searle went too far when he thought everything was a computer: what he meant, or should have meant, was just about anything can be simulated by a computer. But that does not make them identical: A computationally simulated airplane cannot fly in the real world, and a computationally simulated organism or robot cannot move in the real world.)
I also think that your conception of a robot as having to be just a computer plus peripherals is unduly narrow. A robot (whether man-made or gowing on trees) is any autonomous organism-like dynamical system, and there may be a lot of dynamics going on inside it that are other than computation. How much of that could in principle be replaced by internal computer simulations is an empirical question, but I would be skeptical about, say, simulating pharmacological effects…
B.RANSON: “The causes of feeling are observer-independent… what is missing from a robot that stops it feeling [is that] it doesn’t have the appropriate causal properties… That is shown by the multiple realisability of computation… Consciousness is caused by very specific physical processes. It isn’t reasonable… that it [can] also… be caused by innumerable other quite different processes.”
There’s a distinction that has to be made between (1) “multiple realizability” in general (there are plenty of examples, both in engineering and in convergent evolution), in which the same I/O is generated with different mechanisms and the special case of multiple realizability, which is the (2) “implementation-independence” of computation (the hardware/software distinction).
You are conflating them. There may well be more than one way to pass T3, maybe even T4. Some of them may feel and others may not. But it won’t be because of the implementation-independence of computation (because T3 and T4 are not just computational). And, because of the other-minds problem, we’ll never know which ones do feel and which ones don’t (if some do and some don’t). Moreover, we won’t be able to explain the difference (if any)causally either (and that’s the problem!)
If you think feeling can only be generated by certain specific features of T4, please say which those are, and how and why they generate feeling…
B.RANSON: “Specific physical processes in the nervous system cause feeling. They are related to, and a development of, physical processes in the nervous system that cause unconscious behaviour.”
I’m listening…
Stuart W. Mirsky
Comments for Bernie Ranson: I know you want to get Stevan Harnad’s response but I couldn’t help wondering about a few things when reading your remarks. You wrote: “But robots aren’t autonomous systems. That’s entailed by their observer-dependent status. Something is only part of a robot when somebody says it is.” I know you’re taking a Searlean position here (one I think very much in error) so I was wondering if you could more clearly state why you think the statement above is correct? You say robots aren’t autonomous and, certainly, none that I know of at present are. But the issue isn’t just to do with what currently exists. It’s about what can be done with computers going forward and, of course, with robots. So just declaring that robots are unable to achieve what is being argued they might achieve can hardly be an argument against their achieving it. You say that something is only a part of a robot when somebody says it is. Well, yes and no. Certainly whatever we encounter in the world is what we believe about it in a certain sense. The coin is money because people believe that about it, treat it that way and so forth. But in another sense there is what exists independent of us. If we built a robot and endowed it with a million year energy supply and it then managed to remain operational for a million years, but mankind didn’t, would it cease to be what it is (an operating robot) once all the people who used to know it as a “robot”, capable of doing this and this and this, were no longer around to think of it in that way? In what sense is a robot (or a computer) just like money and in what sense is it like us? “Consciousness is caused by very specific physical processes. It isn’t reasonable to believe that it is also going to be caused by innumerable other quite different processes.” Why not? We don’t really know that it can only be caused by very specific processes such as those found in brains. We only know that brain processes do seem to most of us to be implicated causally in the occurrence of consciousness. But that brain processes seem to be causal of consciousness implies nothing about what other processes can accomplish. In fact, maybe it’s not the processes at all. Maybe it’s the functions, i.e., what the processes do. In that case any processes capable of doing the same things (and, presumably, not all processes will be) ought to be able to get us to the same result. Are computers just like read only devices? Merely books, capable of nothing more than an amazon Kindle in the end? All the evidence seems to militate against that presumption, even if it remains to be seen whether computers can be built and programmed to do what brains do.
Joshua Stern
J.STERN: “You have posited that perhaps no computer could pass T2, but what about a human?” SH (@ 3/1/2011 15:30): We’re looking for (causal) *explanations* of T2 capacity (reverse-engineering), not just examples of it. The model’s capacities have to be indistinguishable from those of a real human — but it has to be a model that we built so we can explain how it can do what it can do: Humans are “Turing-indistinguishable” from one another, but so what, since we have no idea how they pass the test? Well, I have to remind us, we *are* looking for examples of it. The test, whether it is T2 or T3, is never explanatory, that comes separately. Perhaps it comes after – we generate T2 candidates at random, and do not plan to even attempt to analyze them until they first succeed. J.STERN: “What… differentiates the fake that passes the test, from the ‘real’ that passes the test?” SH: Feeling. But the whole point of the test is that once you can’t tell them apart, you can’t tell them apart. (And I would not call any robot that can pass the T3 for a lifetime a “fake”!) Lost me again. Let me try to recover. So, T3 for a lifetime equals feeling? Don’t you disclaim that in the paper? It seems to me this skates the edge of my own position, that feeling results from causal systems. Come on over to the dark side! We may be evil, but at least we are consistent.
WHETHER VS HOW/WHY (YET AGAIN)
(Reply to Joshua Stern)
J.STERN “T3 for a lifetime equals feeling? Don’t you disclaim that in the paper?”
(1) Passing T3 certainly does not guarantee feeling.
(2) But it’s the best we can do (Turing Indistinguishable). (That’s why I wouldn’t eat or beat a T3 robot.)
(3) Even if T3 (somehow) guaranteed feeling, we would not know how, and T3 would not explain it.
Harnad: “I’m saying we have no causal explanation of how and how we feel (or anything does), and that we are not likely to get one, because (except if psychokinetic forces existed, which they do not) feelings cannot be assigned any independent causal power in any causal explanation. Hence the causal explanation will always work just as well with them or without them.” I’m sympathetic with this view, which involves two claims: that we’re not going to get a causal explanation of feelings (conscious qualitative phenomenal states, or qualia for short) and that feelings don’t contribute causally in 3rd person explanations of behavior. Re the second claim, I suggest that since feelings like pain aren’t public observables, they are logically barred from causal accounts that involve intersubjectively available objects, such as neurons, brains, bodies and environments. This solves the problem of mental, or more precisely phenomenal, causation: there is none. But of course feelings will still feel causal to the experiencing subject, a perfectly real phenomenal appearance that tracks 3rd person explanations well enough such that it’s very convenient (and harmless) to appeal to the causal role of feelings in folk explanations, even if it’s literally false to do so, see http:// www.naturalism.org/privacy.htm#fiction
Re the 1st claim, I suggest that qualia might be non-causally entailed by being a complex, free-standing, behavior-controlling representational system, of which we are examples. For a number of reasons, any such system will perforce end up with basic, irreducible, uninterpreted representational surds that function as the (privately available) vocabulary in terms of which the world gets represented to the system – qualia. Those reasons, admittedly sketchy and merely suggestive thus far, are presented at http:// www.naturalism.org/appearance.htm#part5
TANGLED WEBS (OR REDUCTIO AD SURDUM)
(Reply to Tom Clark)
T.CLARK: “since feelings like pain aren’t public observables, they are logically barred from causal accounts that involve intersubjectively available objects, such as neurons, brains, bodies and environments. This solves the problem of mental, or more precisely phenomenal, causation: there is none. But of course feelings will still feel causal to the experiencing subject, a perfectly real phenomenal appearance that tracks 3rd person explanations well enough such that it’s very convenient (and harmless) to appeal to the causal role of feelings in folk explanations, even if it’s literally false to do so, see http://www.naturalism.org/ privacy.htm#fiction”
All true, but not particularly helpful, if all one is doing is asking the perfectly innocent and natural question, for which it is perfectly natural to expect an answer (yet no answer turns out to be forthcoming, probably answering is not possible): Why and how do we feel (rather than just do)?
T.CLARK: “[feeling] might be non-causally entailed by being a complex, free-standing, behavior-controlling representational system… any such system will perforce end up with basic, irreducible, uninterpreted representational surds that function as the (privately available) vocabulary in terms of which the world gets represented to the system – qualia. Those reasons, admittedly sketchy and merely suggestive thus far, are presented at http://www.naturalism.org/appearance.htm#part5”
No, I’m afraid that doesn’t do the trick for me…
Harnad: “No, I’m afraid that doesn’t do the trick for me…” For me it at least gets us in the vicinity of qualia by non-causal means, but I’m not surprised you’re not impressed. What we all instinctively want, perhaps, is some sort of causal, mechanistic or even emergentist account of feelings (e.g., HOT-type internal observation) involving public objects, since that’s how 3rd person explanations go. Such explanations likely won’t be forthcoming since feelings (qualia) are categorically private, system-bound realities, intersubjectively unobservable (stronger: they don’t exist intersubjectively). I’m wondering if you have even the glimmer of an idea of what *would* do the trick for you if you rule out as nonstarters the sorts of non-causal entailments I suggest.
HOW TO DO THE TRICK
(Reply to Tom Clark)
T.CLARK: “I’m wondering if you have even the glimmer of an idea of what *would* do the trick for you if you rule out as non-starters the sorts of non-causal entailments I suggest.”
I have the same powerful, convincing causal idea everyone has — psychokinesis — but alas it’s wrong.
Discovering a fifth force — feeling — would of course solve the mind/body problem. Trying to ask “why does the fundamental feeling force feel?” would be like asking “why does the fundamental gravitational force pull?” It just does. (But unfortunately there is no fundamental feeling force.)
(Please don’t reply with the details of space-time curvature. The feeling force could have some of that too. Substitute “Why is space curved?” if you most, to illustrate that the explanatory buck must stop somewhere, with the fundamental forces…)
Stuart W. Mirsky
Apropos the question of the relation of behavior to “feeling”, and just to add a little reality to this discussion, here is a link to a site which purports to decipher canine behaviors in terms of their meanings. Note that all the descriptions imply (and depend on the notion) that the dog’s behavior indicates its feelings:
I submit that this supports my earlier point that behaviors ARE the criteria upon which our references to feelings in others stand and that it’s inconceivable that any entity passing the lifetime Turing Test for having feeling would be thought to lack it. If this is so, then the idea that we don’t “see” (as in have direct access to) the feelings others have cannot be an obstacle to supposing feelings are present — and if that’s the case, then any explanation that accounts for behavioral outcomes in terms of why the entity, say the dog baring its teeth, is doing what it does is enough to account for the presence of the feeling(s). The Other Minds problem has no role in scientific questions about what brains do and how they do it, and is misleading philosophically as well.
NOTHING UP MY SLEEVE (reply to Stevan Harnad) “My questions are simple; if the answers are complicated and book-length, the questions have not been understood and the answers are answers to something else.” Surely you can’t be serious! These are the words of someone who has already decided that the question they are asking can’t answered. And what reasons do you give? An intuition that even once we are able to build a T4 robot we will not be able to explain why and how it feels. But what reason do we have to take your intuition seriously? The kind of in principle claim that you are making is easy to rebut because all you need is one possible answer and that is where the higher-order stuff comes in (though there are probably others, I just happen to understand this one). It is not enough to simply declare it a non-answer because that already assumes that you are correct and so begs the question. This insistence on simplicity looks like a smokescreen for lack of an argument especially since I and others have the contrary intuition. So given the intuition standoff we need to actually look at the proposed explanations themselves and see if they hold up. “awareness of toothache, or awareness of awareness of toothache are merely felt states, like toothache itself” But this actually shows that you do not understand “enough of higher-order approaches to see that they beg the question”. According to the theory toothaches can occur unconsciously and when they do they are not felt states. They are unfelt states that none the less have (mostly) all of their causal powers. They are functed but not felt. But in order to understand that you need to make some distinctions and get clear on what you are talking about and these things sometimes require *gasp* extended treatment to get clear on. But look, I agree that consciousness certainly seems special and mysterious and that one can get oneself all worked up about it if one tries but that is absolutely no reason to think that once we are in a position to build something like T3 or, more likely, T4 we will feel the same. We don’t need a priori convictions or predictions, we need close examinations of empirical evidence and proposed explanations.
CALL FOR COUNTEREXAMPLES!
(Reply to Richard Brown)
R.BROWN: “[you have] already decided that the question [you] are asking can’t answered…”
No, not decided: concluded (from the falsity of psychokinesis and the superfluity of feeling for doing, or causing doing).
R.BROWN: “And what reasons do you give? An intuition that even once we are able to build a T4 robot we will not be able to explain why and how it feels. But what reason do we have to take your intuition seriously?”
Not intuition, reasons: the falsity of psychokinesis and the superfluity of feeling for doing, or causing doing.
R.BROWN: “The kind of in principle claim that you are making is easy to rebut because all you need is one possible answer…”
(No principle either, just innocent, prima facie requests for a causal explanation.)
But I do agree *completely* that just one potentially viable explanation (aside from psychokinesis, which we already know is false) would be enough to refute me. That’s why I invited candidate causal explanations. But all I’ve heard so far — it’s hard to resist the obvious naughty pun here — has been thin air!
I have a skill that has not even been tested yet in this symposium: I am very good at showing how and why candidate causal explanations for feeling fail. But to air my skill (and to earn my comeuppance) I need candidate examples. Here’s an example (alas easily shot down):
“Pain needs to be felt, because if it was just functed, you might not notice the tissue injury, or take it seriously enough to act on it.”
Ghost-Buster: Why doesn’t whatever brings the tissue injury to your attention so you act on it just act on it?
R.BROWN: “and that is where the higher-order stuff comes in… we need to actually look at the proposed explanations themselves and see if they hold up…”
Agreed.
R.BROWN: “According to the theory toothaches can occur unconsciously and when they do they are not felt states.”
And they are not toothaches. At most, they are the physiological consequences of tooth injury.
R.BROWN: “They are unfelt states that none the less have (mostly) all of their causal powers. They are functed but not felt.”
Indeed they are: Assuming they’re enough to make the organism stop chewing on that side, so as not to injure the tooth further, and enough to make it learn to avoid whatever it was that caused its tooth to become injured, etc., all that functional, adaptive stuff that you’ve kindly agreed with me to call “functing.”
So my question is: why and how is all of our adaptive capacity *not* just that — unfelt functing, getting all the requisite doing done, T3-scale, so we can survive and reproduce, etc. Why and how is some of it felt?
(Maybe the clue is in those residual causal powers you are suggesting are left out: Do you mean that not all of T3 can be passed without feeling? But then I’m all ears as to which parts of our T3 capacity require feeling in order to be doable at all — and how and why they duly become felt.
R.BROWN: “in order to understand that you need to make some distinctions and get clear on what you are talking about and these things sometimes require *gasp* extended treatment to get clear on.”
Again, I’m all ears. Maybe we could do a little bit of the extended treatment, in order to be convinced that there really is causal explanation at the end of the tunnel? Because all I’ve heard from HOT partisans so far is that there’s something inside that takes something else inside as its object and “bingo” the feeling lights go on…
R.BROWN: “once we are in a position to build something like T3 or, more likely, T4 [things] will [not] feel [so] special and mysterious. We don’t need a priori convictions or predictions, we need close examinations of empirical evidence and proposed explanations.”
Reverse-engineering a causal mechanism that can pass T3 (or perhaps T4) is the right empirical program for cognitive science, just as Turing suggested. I can hardly disagree with that.
Harnad, S. and Scherzer, P. (2008) First, Scale Up to the Robotic Turing Test, Then Worry About Feeling. Artificial Intelligence in Medicine 44(2): 83-89
But it doesn’t — and won’t — touch feeling (consciousness).
(Isn’t it odd, though, that those who are — rightly — saying we need to wait to conquer T3 or T4 already think they have a potential explanation of feeling now, with HOT? Makes you wonder whether HOT has much to do with causal explanation at all.)
Joshua Stern
Tom Clark (@3/2/2011 12:03): For me it at least gets us in the vicinity of qualia by non-causal means, but I’m not surprised you’re not impressed. What we all instinctively want, perhaps, is some sort of causal, mechanistic or even emergentist account of feelings (e.g., HOT-type internal observation) involving public objects, since that’s how 3rd person explanations go. Such explanations likely won’t be forthcoming since feelings (qualia) are categorically private, system-bound realities, intersubjectively unobservable (stronger: they don’t exist intersubjectively). I’m one of we all who want that HOT-type internal observation, so let’s see how that impacts what we can say about qualia. Certainly we know about the various “physical correlates” as we like to say, and those are in principle observable intersubjectively. The story goes that there is still some phenomenological residue, the subjective aspect, that is private – but wait a moment, let’s not mistake what that means. The HOT idea is that we can get very close, at least, to that phenomenological occurrence, that there is a physical correlate to the occurrence of the subjective. So if the C-fiber stimulation is blocked, the brain never gets it, no pain. If the stimulation reaches the brain and fires all the triggers, but some painkilling drug prevents that last correlate from firing, no pain. You might even ask the subject, “do you feel that?” and get an answer like, “now that you ask, yes.” The point is to remind ourselves that the privacy of qualia is only regarding the subjective aspects. And that’s the pessimistic view. The optimistic view is that, at the last, it will turn out that the HOT – or some other simple, causal, explanation – will after all completely comprise subjective phenomena. We will simply have to retune our intuitions. Consider the case where we learn how to do this with mere programming of even common computers. Even then, will we “feel” the red, when we read about the program “feeling” red? Probably not. But the problem then is no longer with the mystery of subjectivity, but with our limited empathic skills.
Bernie Ranson
B.RANSON: “‘Money’ only exists when somebody says it does…”
S. HARNAD: “And feeling only exists when somebody feels it… But it doesn’t only exist when somebody says it does. Knowing that feeling only exists when somebody feels doesn’t make any contribution to our knowledge, it’s a tautology. But knowing that consciousness exists whatever anybody says about it, and that the same is not true of computers, does add to our knowledge.
S. HARNAD: “Intentionality” is another of the weasel-words. If “intrinisic” intentionality’s not just T3 grounding, then it’s “derived” if it is unfelt, and “intrinsic” if it is felt. A book’s and a computer’s “intentionality” is not only unfelt (hence “derived”), No, I’m afraid that is wrong. The book and computer’s intentionality is felt, by the observer when they have those ideas. Intentionality is a mental phenomenon, it is always “felt” in your terminology. The book and the computer’s intentionality is derived because it is felt not by the book but by the observer.
S. HARNAD cont: but it is also ungrounded (no connection between internal symbols and the external things they are about). Sensorimotor grounding remedies that in a T3 robot — The only intentionality a computer has is derived intentionality. If “grounding” means a connection between the symbols and external things, then that derived intentionality is grounded, those connections are made in the mind of the observer. The computer
doesn’t have any intentionality of its own towards the symbols. So adding sensorimotor capacities can’t ground the computer’s intrinsic intentionality, because it hasn’t got any. B.RANSON: “robots aren’t autonomous systems… [they are] observer-dependent… Something is only part of a robot when somebody says it is.”
S. HARNAD: I think you’re wrong on this one, on anyone’s sense of “autonomy.” What they have is “derived autonomy”, which isn’t really autonomy at all. Genuine autonomy on the other hand is what is crucially missing from computers and robots, it’s another way of describing the thing that you challenged us to identify.
S. HARNAD: (And, as I said, Searle went too far when he thought everything was a computer: what he meant, or should have meant, was just about anything can be simulated by a computer… No, he meant what he said, he didn’t mean what you say at all. He didn’t say everything is a computer, he said that anything of sufficient complexity could be used as a computer. Anything with the features of a Universal Turing Machine, which as you know can be specified in a few sentences.
S. HARNAD: “I also think that your conception of a robot as having to be just a computer plus peripherals is unduly narrow. A robot (whether man-made or growing on trees) is any autonomous organism-like dynamical system, …” If you widen the scope of the term “robot” until it covers us, then questions like “can a robot feel” become entirely uninteresting, and your questions about what robots lack fail to make sense. I know the kinds of things that exist today that are called robots, the amazing welding and assembly machines in car factories, the humanoid figures that can run around the room and open doors and appear to be looking at you and talking plausibly about things they can see or hear. They are just a computer plus peripherals, or just peripherals. You say we could add something else, these “dynamics”, but how are they going to be any different to the existing computer and/or peripherals?
S. HARNAD: and there may be a lot of dynamics going on inside it that are other than computation. How much of that could in principle be replaced by internal computer simulations is an empirical question, but I would be skeptical about, say, simulating pharmacological effects… None of the relevant dynamics could be replaced by computation. You can simulate anything you like, what is going on in a simulation is observer-dependent, it doesn’t have the causal properties of the thing it is simulating, you don’t get wet in a simulated thunderstorm and you don’t get feeling from simulating a feeling organism.
S. HARNAD: There’s a distinction that has to be made between (1) “multiple realizability” in general (there are plenty of examples, both in engineering and in convergent evolution), in which the same I/O is generated with different mechanisms and the special case of multiple realizability, which is the (2) “implementation-independence” of computation (the hardware/software distinction). You are conflating them. I don’t think so, I think from my perspective 2. is as you say merely a special case of 1. I would suggest that you are applying the concept of multiple realisability over-optimistically: there’s an implication that because the same I/O can be generated by different mechanisms, then the desired I/O will be generated by different mechanisms. We can see the kinds of things that are going on in the brain that are related to feeling, we can manipulate them in many different ways with predictable results, we know that feeling is caused by our physical brains. Yes, it’s possible that this could be done using some other mechanism: but we would need some reason to think that any given specific mechanism might be able to do it. And there isn’t any reason to think that something like a computer would be able to do something like that, any more than a toaster would. If you think a computer/robot has specific features that allow it to generate feeling, please say which those are and how they generate feeling.
S. HARNAD If you think feeling can only be generated by certain specific features of T4, please say which those are, and how and why they generate feeling… We don’t yet know what specific features of the body cause feeling. I have somebody working on this (my son the Neuropsychologist). He has enormous microscopes and all different kinds of rays and imaging devices and I’ll be sure to let you know if he comes up with anything. So we don’t know the how yet, but in the meantime: the how is the why. Evolution found a way to make feeling, it was useful, so it survived.
ASK NOT WHETHER WE ARE ROBOTS; ASK WHAT KIND OF ROBOTS WE ARE
(Reply to Bernie Ranson)
B.RANSON: “If you widen the scope of the term “robot” until it covers us, then questions like “can a robot feel” become entirely uninteresting, and your questions about what robots lack fail to make sense.”
Not if you widen the scope of what the robot must be able to do (T3). We can ask for the causal mechanism of T3, and the causal explanation of how/why a system with T3 power feels (if it does).
This does not widen “robot” so much as too much to make it uninteresting: It widens what a robot can do just wide enough to make it interesting to explain how — and then ask how and why it feels, if it does.
B.RANSON: “You say we could add something else, these “dynamics”, but how are they going to be any different to the existing computer and/or peripherals?”
For example, if the only way get it to do human-scale Shepard rotation tasks were to build in an analog rotator, rather than a computation in polar coordinates, that would be deep dynamics. Or if (as in the brain) 16 internal isomorphic projections of the retinal projection proved more powerful and useful than a bitmap and computations…
B.RANSON: “If you think a computer/robot has specific features that allow it to generate feeling, please say which those are and how they generate feeling.”
But you see, I haven’t the faintest idea, and that’s precisely why I’m asking how *any* system could generate feeling. On the other hand, I have more hope for explaining how it could generate T3 or T4 performance power…
Philip Goff
Steve I am in agreement with your conclusion, but I really think we’re obliged to take on the best cutting edge work of our opponents, i.e. a posteriori physicalists who employ the so called ‘phenomenal concepts strategy’, e.g. David Papineau, Brian Loar, Joseph Levine. Nothing you say in your paper really touches them. They agree that there is a conceptual gap between physical and phenomenal qualities, but they try to explain this in terms of features of phenomenal concepts. Papineau’s explanation of why pain and c-fibres firing are correlated is that the concept of pain refers to the property of c-fibres firing; it’s just that we can’t know a priori that that’s the case (hence the conceptual gap).
FUNCTING PAIN
(Reply to Philip Goff)
P.GOFF: “a posteriori physicalists who employ the so called ‘phenomenal concepts strategy’, e.g. David Papineau, Brian Loar, Joseph Levine. Nothing you say in your paper really touches them. They agree that there is a conceptual gap between physical and phenomenal qualities, but they try to explain this in terms of features of phenomenal concepts. Papineau’s explanation of why pain and c-fibres firing are correlated is that the concept of pain refers to the property of c-fibres firing; it’s just that we can’t know a priori that that’s the case (hence the conceptual gap).”
It may be that nothing I say touches those who believe that pain refers to the property of c-fibres firing. But I was asking how and why pain is felt rather than just functed. That would apply to c-fibres firing, too…
Bernie Ranson
S. MIRSKY: You say that something is only a part of a robot when somebody says it is. Well, yes and no. Certainly whatever we encounter in the world is what we believe about it in a certain sense. Hi Stuart, The important point is that some things are what they are whatever anybody says. Gravity, mass, consciousness, digestion. Robots and computers aren’t in that category
S. MIRSKY: If we built a robot and endowed it with a million year energy supply and it then managed to remain operational for a million years, but mankind didn’t, would it cease to be what it is (an operating robot) once all the people who used to know it as a “robot”, capable of doing this and this and this, were no longer around to think of it in that way? It wouldn’t cease to be anything, it would cease to be described as something.
B. RANSON: “Consciousness is caused by very specific physical processes. It isn’t reasonable to believe that it is also going to be caused by innumerable other quite different processes.”
S. MIRSKY: Why not? Because there isn’t any reason. There isn’t anything about what computers do that would make you think they are conscious, any more than a toaster.
Stuart W. Mirsky
Reply to Bernie Ranson’s Response to Stevan Harnad @ 3/2/11 14:21 Bernie Ranson writes: “The only intentionality a computer has is derived intentionality.” This is just an expression of a belief. It’s not an argument, it’s dogma. BR: “(Searle) didn’t say everything is a computer, he said that anything of sufficient complexity could be used as a computer.” SWM: No, he said it could be called a computer. (Of course such a broad claim makes the idea of “computers” pointless since if anything can be called, and thus treated, as one, then there’s no point in doing so because you cannot distinguish between computers and what aren’t. If you cannot, why bother calling something a computer at all?) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ BR: “I know the kinds of things that exist today that are called robots, the amazing welding and assembly machines in car factories, the humanoid figures that can run around the room and open doors and appear to be looking at you and talking plausibly about things they can see or hear. They are just a computer plus peripherals, or just peripherals. You say we could add something else, these ‘dynamics’, but how are they going to be any different to the existing computer and/or peripherals?” SWM: The point is not what current computers and robots do but what computers and robots CAN BE MADE to do. You can’t deny the possibility of something just by denying its actuality. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ BR: “None of the relevant dynamics could be replaced by computation.” SWM: Again, this is argument by asserting belief. If the point is to ask whether we can do what this statement denies (which it is), then you have to have a reason that goes beyond the mere recitation of the denial. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ BR: “We can see the kinds of things that are going on in the brain that are related to feeling, we can manipulate them in many different ways with predictable results, we know that feeling is caused by our physical brains. Yes, it’s possible that this could be done using some other mechanism: but we would need some reason to think that any given specific mechanism might be able to do it. . . there isn’t any reason to think that something like a computer would be able to do something like that, any more than a toaster would.” SWM: One reason: Similarity in the mechanics. Brains appear to operate by passing electrical signals along particular pathways formed by neurons, neuronal clusters and synaptic spaces between while computers operate by passing electrical signals along particular pathways through various logic gates embedded in chips and chip arrays. Another reason: Similarity of outcomes. A lot of the things brains do are also doable using today’s computers (as in playing chess, answering questions, calculating equations and so forth). If so, then it is possible computers can do more of the things brains do. These are strong reasons to think that computers could do what brains do while chairs and tables would not be able to. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ BR: “Evolution found a way to make feeling, it was useful, so it survived.” SWM: This is certainly true on its face but so what? It says nothing about the reverse engineering called for by Stevan. The issue isn’t what factors in biological history brought about the development of brains which happen to produce consciousness or even what purpose consciousness in brains serves for the organism having it. Rather it’s what do brains do that produces it and could it be built synthetically, either using computers or some other inorganic platform?
Stuart W. Mirsky
Reply to Bernie Ranson’s response to me @ 3/2/11 14:50 Bernie Ranson wrote: “The important point is that some things are what they are whatever anybody says. Gravity, mass, consciousness, digestion. “Robots and computers aren’t in that category” SWM: Neither are gravity and mass in the same category as digestion and, so far, no one really knows what category consciousness is in. What about robots and computers? I go with Stevan Harnad’s view, at least with regard to robots. They are any autonomous operating system “grounded” in real world interactions. Of course, in ordinary usage we mean by “robots” only those certain mechanical devices made of inorganic material that are capable of getting about on their own and performing operations in the world with some degree of autonomy (but not necessarily with complete autonomy). But the issue before us is what could a robot be built to do, not what current robots can now do. You can’t argue that robots can’t be built to be conscious just because they’re, well, robots and no robots do what is at issue! That begs the question in a huge way. The same applies to computers. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
S. MIRSKY: If we built a robot and endowed it with a million year energy supply and it then managed to remain operational for a million years, but mankind didn’t, would it cease to be what it is (an operating robot) once all the people who used to know it as a “robot”, capable of doing this and this and this, were no longer around to think of it in that way? BR: “It wouldn’t cease to be anything, it would cease to be described as something.” SWM: Right. But this isn’t about describing. It doesn’t matter if we call a human a human or a featherless biped or a big-brained hairless primate, etc. And it doesn’t matter if we call a robot a robot. What matters is what the entity in question can DO. A robot in a world without humans (or any other intelligent species capable of naming and describing) would no longer be known as a robot. Indeed it would no longer be known at all — unless it had its own consciousness and knew itself (or unless we want to assume the presence of less intelligent species than humans which “know” what they know in a different way than we do. My late cat which would never have mistaken a robot for me but might have known it for a bothersome obstacle across its path or as a nice place to curl up on, once it stopped moving). ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
B. RANSON: “Consciousness is caused by very specific physical processes. It isn’t reasonable to believe that it is also going to be caused by innumerable other quite different processes.”
S. MIRSKY: Why not? BR:”Because there isn’t any reason. There isn’t anything about what computers do that would make you think they are conscious, any more than a toaster.” SWM: This just sounds like dogma. In fact there are certainly reasons and quite good ones. Here are a couple: 1) Similarity in the mechanics: Brains appear to operate by passing electrical signals along particular pathways formed by neurons, neuronal clusters and synaptic spaces between while computers operate by passing electrical signals along particular pathways through various logic gates embedded in chips and chip arrays. 2) Similarity of outcomes: A lot of the things brains do are also doable using today’s computers (as in playing chess, answering questions, calculating equations and so forth). If so, then it is possible computers can do more of the things brains do. These are strong reasons to think that computers could do what brains do while chairs and tables would not be able to.
Stuart W. Mirsky
In replying to Richard Brown, Stevan Harnad wrote (among other things) that his position was argued for, contra Richard’s point that he was invoking intuition. Stevan said: “Not intuition, reasons: the falsity of psychokinesis and the superfluity of feeling for doing, or causing doing.” And yet Stevan’s denial of my argument, that it’s at least theoretically possible to explain “feeling” in terms of a process based system, hinges on his invocation of the Other Minds Problem, a notoriously insoluble philosophical conundrum (unless one is prepared to deconstruct the problem and deny it meaning a la Wittgenstein). The Other Minds Problem cannot be argued against because it hangs on a certain way of understanding things, i.e., of acceptance of the idea that knowledge requires indubitablity which is only achievable through direct access or reasoning from information acquired via direct access. I suggest that this IS intuition-driven and, therefore, Stevan’s denial of my proposed answer to his challenge looks to be founded not on reasons, contra his claim, but on intuition too. Stevan further wrote: “I invited candidate causal explanations. But all I’ve heard so far — it’s hard to resist the obvious naughty pun here — has been thin air!” But, frankly, how is a denial of a thesis based on the Other Minds problem be relevant to a scientific question or supported by reasons, as he asserts, when the problem is not, itself, grounded in reasons but in a certain way of understanding what we mean by knowledge in cases like this? If we assume that the only way we can know, with any certainty, about others’ feelings is via direct access, but are denied that access, then we are obliged to accept the claim that we cannot know about others’ feelings. And yet, we DO know about others’ feelings. Even Stevan admits this when he says he would not deny the feeling of any entity that passed the lifetime Turing Test at, at least, the T3 level. So why demand more of the T3 passing entity scientifically? And has an answer to the proposal offered, that hinges on an intuition that absence of access amounts to absence of knowledge, saatisfied Stevan’s assertion that his is a reasoned stance rather than an intuitive one?
Joshua Stern
SH (@ 3/2/2011 16:21) (2) But it’s the best we can do (Turing Indistinguishable). (That’s why I wouldn’t eat or beat a T3 robot.) Seems like a rash commitment, not that either eating or beating a piece of metal seems like a good idea in the first place. Would you turn it off for the night? Would you give it the vote? There’s a fair amount of (science) fiction literature on all of this, of course, much of it more along the lines of whether it would eat or beat *you*! But seriously, folks, isn’t an appeal to Turing Indistinguishability moving in the wrong direction? If that was enough, then Searle’s posit of the CR would not, could not, suffer from its (alleged) shortcomings. And I thought we were all agreeing that we want a constructive answer – we don’t have an answer until we have a constructive answer. Something something about hermeneutics otherwise. SH: (3) Even if T3 (somehow) guaranteed feeling, we would not know how, and T3 would not explain it. Again, this seems to give up the argument. What exactly might it be, for us to know how the feelings are guaranteed? On the one hand we have literature, as in Wittgenstein, on the role and surveyability of proofs. On the other hand we have the pragmatic experiences of writing and using computer programs. If I write a program to compute prime numbers, do I know “how” – “how” what, exactly? How they are computed in the abstract, or in the particular case of 2^n-1, n=123456789? If I write a program (or build a robot) that does X, is there any more to know about generating and guaranteeing X, than what was done? The Wittgensteinian answer is, to make a long story short, no, that’s it. The compsci answer, that pesky Systems Reply again, is also – this is it, all there is. Do we know how a program puts a picture of Obama on the screen, even though all it does is move around ones and zeroes? Does that guarantee the image I? Could there be more to guaranteeing a feeling F? I don’t see how, unless a priori we distinguish F from all other things that are not F, but if we do that we’ve prejudged the issue and are just moving around the pieces. If it comes to prejudgement, mine is that T2 is as good at Turing Indistinguishability as any other possible TX, and has the historical advantage that Turing wrote it that way. The only reason to move away from T2 is if it gives us a more constructive answer, such that we would have exactly the knowledge, to the degree that one can ever have that kind of knowledge, of what those feelings are and how they are generated. That does seem to be a call to move beyond Turing Indistinguishability, that follows from your argument, and in some way I also credit Searle for motivating us in that direction.
CONSTRUCTION AND EXPLANATION
(Reply to Joshua Stern)
J.STERN: “What exactly might it be, for us to know how the feelings are guaranteed?”
We don’t want guarantees. (This is not about the other-minds problem.) We want causal explanation for the feeling, just as we would get it for the (T3, T4) doing.
J.STERN: “I write a program to compute prime numbers, do I know ‘how’”
Yes.
J.STERN: “If I write a program (or build a robot) that does X, is there any more to know about generating and guaranteeing X, than what was done?”
No. And what was done with the prime numbers, was the computation of primes. And with T3, the generation of the T3 performance capacity. And if that also generated feeling, we want to know how that was done. Without further notice, if it generates T3 all we have explained is how it generated T3.
J.STERN: “Do we know how a program puts a picture of Obama on the screen, even though all it does is move around ones and zeroes? Does that guarantee the image I?”
Yes.
J.STERN: “Could there be more to guaranteeing a feeling F?”
Yes indeed.
J.STERN: “I don’t see how, unless a priori we distinguish F from all other things that are not F”
Here’s a way they distinguish themselves: With all those other things, you can explain how and why your model generates them; with F you can’t.
J.STERN: “but if we do that we’ve prejudged the issue and are just moving around the pieces”
No, we’ve just called a spade a spade, and admitted that we have a gaping explanatory gap.
J.STERN: “If it comes to prejudgement, mine is that T2 is as good at Turing Indistinguishability as any other possible TX, and has the historical advantage that Turing wrote it that way.”
Yes, but it might be better to keep thinking…
J.STERN: “The only reason to move away from T2 is if it gives us a more constructive answer”
Your point has been made, multiple times, Joshua. You want a working model. Fine. Point taken. But that is not an argument. It is changing the subject.
Stuart W. Mirsky
Intuition and Argument: Earlier on I posted something suggesting that Stevan’s argument was intuition-based, contra his response to Richard Brown that he was relying entirely on reasons which he believed were supportable to make his case. The case he makes is that what is central to consciousness, the condition of feeling anything at all (which Stevan holds is the most basic and least misleading way of speaking about consciousness), can never be accessed in others and therefore can never be explained scientifically. We can, Stevan maintains, scientifically account for the performance of an entity through reverse engineering and, if that entity is capable of acting with apparent autonomy, which looks like what our autonomous behavior looks like, then we can at least conclude that we have captured the behavior mechanisms that match our own. But, he has argued, those mechanisms need not involve feeling like we have and so we can never have reason to think that such a reverse-engineered entity feels anything at all. Stevan denies that his argument has to do with the philosophical zombie question (even though it seems to imply the possibility of such things) and, while periodically invoking the historically troublesome philosophical problem of Other Minds, he has also denied that this is relevant to his position as well. His position appears to boil down to the claim that cognitive science must forever limit itself to questions of behaviors, and their mechanisms, while giving up on questions of “feeling” entirely. In his paper, he invited others to try to counter his position (that no explanation of the phenomenon of “feeling” is possible) by offering arguments which showed that some explanation was possible. Several have since been offered here but Stevan has rejected them all. His grounds for rejecting them seem to come down to these: 1) Psychokinesis (mind over matter?) would need to be possible for feeling and physical entities to be linked and there is no evidence that it is; and 2) That there is a “superfluity of feeling for doing or causing doing”. (Presumably he means that feeling and doing are conceptually disconnected.) Richard Brown suggested that at least the second basis for rejection seems to be consistent with epiphenomenalism and a commitment to the conceivability and, therefore, possibility, of philosophical zombies. But Stevan denies this while continuing to maintain that his claim about feeling and physicality is reason based, open to argument and, therefore, refutation. My last response indicated my agreement with Richard’s position, that Stevan’s approach looked like an intuition-based one to me. (Of course, I could be said to have an ax to grind as my response to Stevan’s challenge was one of those he rejected out of hand.) On reconsideration, though, I think my explanation of why Stevan’s response is intuitionistic was inadequately explicated. I put it all down to the role of the Other Minds Problem in Stevan’s argument (which, of course, Stevan has denied is relevant to his claims) and traced it to the problem of certainty as it relates to what we can know. While not withdrawing my point about the role of the Other Minds Problem in Stevan’s argument, I think I could have been a bit clearer. The Other Minds Problem (which holds that, because we can’t access the minds of others in the way we have access to our own, we can’t be sure anyone else has a mind but ourselves) is, it seems to me, driven by two factors: 1) The supposition that minds (“feelings”) have an ontic status of the same type as other objects of our knowledge (i.e., that they have some kind of existence apart from their physical base — that they aren’t, finally, physical, whatever their provenance or genesis); and 2) The supposition that, to know anything, we must be certain of it and that certainty amounts to a claim’s being seen to be true in an undeniable way (with no room for doubt). I focused in my response on the second element of the Other Minds Problem. But that, I now think, is not a matter of intuition but of linguistic misapplication, i.e., it’s a case of using our terms wrongly (expecting “certainty”, which is important to claims of knowledge, to always be a matter of indubitability — in the way it is for math equations, deductive syllogisms and so forth). I don’t think we do this intuitively but only when we aren’t paying close enough attention to what we actually mean when we use terms having to do with knowing things, etc. In fact, I should have focused on the first part of the problem which is where the real issue of intuition resides, I think. Given that our language is publicly grounded, we expect mental referents to be like referents in the public realm of our experience. Unlike the issue for #2 above, this is NOT a misuse of our language but the result of thinking in certain categories. That is, it’s a failure to see how our language can and often does lead us astray in certain applications. Unlike the question of what it means to know or be certain (these uses vary by context and cannot simply be substituted one for another), we DO routinely refer to our mental lives and this cannot be an error of use because it’s an important part of our discourse. Rather it looks like an error of conceptualization in that we may allow the pictures from one area of our operation to mpose themselves on another. Once we recognize the distinction between public referents and the private (mental) kind, and how language is used differently in the two contexts, the expectation that mental phenomena must be entity-like, and therefore exist in the way public objects exist, evaporates. But this cannot be argued for. It involves seeing the way our language works in a new way and that’s why it’s finally a matter of intuition — and why, I think, Stevan’s second reason for denying the possibility of an explanation of how physical phenomena result in feelings is intuition-based.
SIMPLE WHY/HOW QUESTIONS
(Reply to Stuart W. Mirsky)
S.MIRSKY: “The case he makes is that… [feeling] can never be accessed in others and therefore can never be explained scientifically”
No. Feeling cannot be explained causally because it is causally superfluous to explain doing and we can only explain doing.
The constant importation of the other-minds problem (“Would a T3 robot feel or not?”) has next to nothing to do with it. Whether or not the T3 robot feels, we cannot explain how or why it does.
S.MIRSKY: “if [T3's] behavior [is indistinguishable from ours] then we can at least conclude that we have captured the behavior mechanisms that match our own. But, [Stevan] has argued, those mechanisms need not involve feeling like we have and so we can never have reason to think that such a reverse-engineered entity feels anything at all.”
The constant importation of the other-minds problem (“Would a T3 robot feel or not?”) has next to nothing to do with it. Whether or not the T3 robot feels, we cannot explain how or why it does.
S.MIRSKY: “Stevan denies that his argument has to do with the philosophical zombie question… and…[he has also denied that] the historically troublesome philosophical problem of Other Minds… is relevant to his position.”
Correct.
S.MIRSKY: “His position appears to boil down to the claim that cognitive science must forever limit itself to questions of behaviors, and their mechanisms, while giving up on questions of “feeling” entirely.”
Correct.
S.MIRSKY: “Stevan has rejected [attempted explanations on the grounds of] “superfluity of feeling for doing or causing doing”. (Presumably he means that feeling and doing are conceptually disconnected.)”
I’m not sure what “conceptually disconnected” means, but I mean that feeling is causally superfluous in explaining doing, or the causes of doing.
S.MIRSKY: “Richard Brown suggested that at least the second basis for rejection seems to be consistent with epiphenomenalism and a commitment to the conceivability and, therefore, possibility, of philosophical zombies. But Stevan denies this.”
That’s right. I have no idea what “epiphenomenalism” means. I’m just asking for (and not hearing) a viable causal explanation of why and how we feel. If people find it informative to dub the absence of that explanation “epiphenomenalism,” they’re welcome to call it what they like. I don’t find that the term conveys any information at all.
And as for zombies: what more can I say than that I don’t believe in them; I don’t use them; I don’t regard “conceivability” as any sort of rigorous criterion for anything; and if (as I believe) zombies are not possible, I certainly cannot explain why or how they are impossible, because that would be the same as explaining why and how we feel.
S.MIRSKY: “[Stevan] maintain[s] that his claim about feeling and physicality is reason based, open to argument and, therefore, refutation.”
In principle, yes, it is, since I can no more prove, a priori, that there cannot be a causal explanation of how and why we feel than I can prove that there cannot be zombies. All I can do is give the reasons I think such attempted explanations are bound to fail, and then try to show, one by one, with actual attempted explanations, how they do indeed fail.
S.MIRSKY: “Stevan’s response is intuitionistic”
I don’t think so: It no more appeals to intuitions than it appeals to zombies, the other-minds problem, or epiphenomenalism. It is not an appeal *to* but an appeal *for* — causal explanation.
S.MIRSKY: “The supposition that minds (“feelings”) have an ontic status of the same type as other objects of our knowledge…”
I don’t understand “ontic status” or “ontic status type”: There are (roughly) objects, events, actions, properties and states. I suppose a feeling is a kind of state. But that doesn’t help if what we want is a causal explanation.
S.MIRSKY: “The supposition that, to know anything, we must be certain of it and that certainty amounts to a claim’s being seen to be true in an undeniable way (with no room for doubt).”
Who asked for certainty? I can’t even be certain about the T3 explanation of doing (or about the scientific explanation of any empirical phenomenon). I’m with Descartes and the soft skeptics on certainty: it’s only possible with (1) mathematically provable necessity and (2) the Cogito (“I feel, therefore there is feeling”). The rest is just probability and inference to the best explanation.
But with feeling, there is no explanation at all.
S.MIRSKY: “language publicly grounded… mental referents… public experience realm… NOT language misuse… thinking in certain categories… language leads astray in certain applications… what it means to know or be certain… part of our discourse… error of conceptualization… pictures from one area impose… public referents… private (mental) referents… language use… different contexts… entity-like expectation… public object existence… language works in a new way… matter of intuition — and why, I think, Stevan’s second reason for denying the possibility of an explanation of how physical phenomena result in feelings is intuition-based…”
Such a complicated way to say: We do and we feel. We can explain how and why we do, but not how and why we feel. Amen. (And may Wittgenstein rest in peace.)
Miguel Sebastian
Great discussion! Thanks for your cross-post. I take it that what we all are interested in is what you call feelings. If one is an a posteriori physicalist and endorses the phenomenal concept strategy that Philip Goff mentioned, she will explain why your question won’t receive the answer you are expecting. Pain might be identical to a functional state, but you won’t be able to deduce phenomenal properties (the way it feels or the fact that it feels) from its functional description. With regard to the discussion about HOT. Very roughly, for HOT theories a state is conscious iff it is the target of the right kind of higher order thought. Imagine the state you are in when you are in pain, M. According to HOT M is conscious (it feels “painfully” for me to be in this state) because it is targeted by another higher-order state of the right kind. I have understood that M has its “painfulness” independently on whether it is conscious or not, it is a qualitative state. I do not understand very well what is a non-conscious qualitative state, so maybe you can ask what distinguishes non-conscious qualitative states from non qualitative states (I don’t know the answer). This way you could understand why the HOT is required to get a feeling and how this could address your question or whether you think that this qualitative state are already ‘feeling states’.
COOL ON HOT: FELT AND UNFELT FEELING
(Reply to Miguel Sebastian)
M.SEBASTIAN: “an a posteriori physicalist [who] endorses the phenomenal concept strategy that Philip Goff mentioned… will explain why your question won’t receive [an] answer… Pain might be identical to a functional state, but you won’t be able to deduce phenomenal properties (the way it feels or the fact that it feels) from its functional description.”
A rather complicated explanation for why we cannot explain how and why we feel. But I would say it was already apparent that we could not…
M.SEBASTIAN: “for HOT theories a state is conscious iff it is the target of the right kind of higher order thought. Imagine the state you are in when you are in pain, M. According to HOT M is conscious (it feels “painfully” for me to be in this state) because it is targeted by another higher-order state of the right kind. I have understood that M has its “painfulness” independently on whether it is conscious or not, it is a qualitative state.”
Deflationary transcription:
“for HOT theories a state is felt iff it is the target of the right kind of higher order thought. Imagine the state you are in when you are feeling pain, M. According to HOT M is felt (it feels “painfully” for me to be in this state) because it is targeted by another higher-order state of the right kind. I have understood that M has its “painfulness” independently of whether it is felt or not, it is a felt state”
Is that supposed to be an answer to my question of why and how I feel? If so, it doesn’t quite do the trick. It sounds like process hermeneutics, begging the question while embedding it in an interpretative system.
M.SEBASTIAN: “I do not understand very well what is a non-conscious qualitative state”
Nor do I. In my more direct language it is an unfelt feeling, and I haven’t the faintest idea what that’s supposed to be!
M.SEBASTIAN: “so maybe you can ask what distinguishes non-conscious qualitative states from non qualitative states (I don’t know the answer).
Deflated: “maybe you can ask what distinguishes unfelt felt states from unfelt states”
I sure don’t know the answer either: I can’t even understand the question!
M.SEBASTIAN: “This way you could understand why the HOT is required to get a feeling and how this could address your question or whether you think that this qualitative state are already ‘feeling states’”
I’m afraid HOT only serves to cool my understanding. My suspicion is that it conceals the lack of understanding under layers of interpretation.
Harnad: “I invited candidate causal explanations. But all I’ve heard so far — it’s hard to resist the obvious naughty pun here — has been thin air!” And: “I have the same powerful, convincing causal idea everyone has — psychokinesis — but alas it’s wrong. “Discovering a fifth force — feeling — would of course solve the mind/body problem. Trying to ask “why does the fundamental feeling force feel?” would be like asking “why does the fundamental gravitational force pull?” It just does. (But unfortunately there is no fundamental feeling force.)” You seem to be stuck on the notion that to really explain feelings, the explanation has to be causal, involving mechanisms and/or undiscovered forces that somehow produce or generate them. But that claim is not obviously true, and indeed the idea that a causal explanation of consciousness could in principle exist seems to me unlikely, given that feelings are categorically private, unlike their neural correlates. Were feelings generated or caused by observable forces or mechanisms, feelings would be in the public domain as is every other product of causal processes, but they aren’t. This suggests they aren’t caused, produced or generated, but instead entailed (for the system alone) by *being* a system with certain sorts of representational capacities. There are philo-scientific considerations that count in favor of this possibility adduced at http://www.naturalism.org/appearance.htm I’m not saying my proposal is correct, only that the fact that it isn’t causal shouldn’t lead one to reject it out of hand as an answer to your innocent question: “Why and how do we feel (rather than just do)?”
Stuart W. Mirsky
Response to Tom Clark’s comments of 3/3/11 @ 13:08 You write that “the idea that a causal explanation of consciousness could in principle exist seems to me unlikely, given that feelings are categorically private, unlike their neural correlates.” But why should this preclude a causal explanation? What is “causal” but a claim that X brings about Y (as in makes it happen, results in it, is sufficient for its occurrence, etc.)? If an account is given of a set of operations which will produce in a given machine “feeling”-related behavior which is convincingly that (in every conceivable way we might need to be convinced), then why would that not be sufficiently “causal” to fit the bill? Stevan’s position seems to be that the inherent subjectivity of feelings, the fact that they are private to the entity having them, precludes our ability to ascertain that they are really present and therefore such an explanation can only explain the occurrence of the behaviors, not the feelings (which may even be self-reported by the machine but which still, being private to the machine, cannot be credited with being there). On these grounds he seems to conclude that a causal explanation isn’t possible. But if he’s wrong about what’s needed to assure ourselves of the occurrence of the feelings in the machine, then hasn’t a causal explanation (in all the usual senses of “causal”) been provided? What else would be needed and doesn’t your concession that “causal” isn’t relevant because it doesn’t apply in this kind of case amount to an affirmation of Stevan’s position that a “causal” explanation isn’t possible? After all, that’s all he really seems to be claiming here.
Joshua Stern
J.STERN: “I don’t see how, unless a priori we distinguish F from all other things that are not F” SH (@ 3/2/2011 20:07: Here’s a way they distinguish themselves: With all those other things, you can explain how and why your model generates them; with F you can’t. I have not accepted that point: I assert that all such things are both caused and causal. I assert there is nothing but the physical, causal world. Any phenomena that occur within deserve an explanation, but I am unconvinced there is a problem here of special concern. If you want to accept this a priori difference between F and non-F, that’s fine, I am happy to point out, repeatedly as required, that I don’t accept it, don’t recognize it in the first place to be accepted, accept as just another phenomenon like the rising of the sun or the wetness of water. J.STERN: “If it comes to prejudgement, mine is that T2 is as good at Turing Indistinguishability as any other possible TX, and has the historical advantage that Turing wrote it that way.” SH: Yes, but it might be better to keep thinking… We agree on that. J.STERN: “The only reason to move away from T2 is if it gives us a more constructive answer” SH: Your point has been made, multiple times, Joshua. You want a working model. Fine. Point taken. But that is not an argument. It is changing the subject It is the only subject, it is the key to the analysis of the problem, and must be reviewed at every point. In discussions with SWM offline, perhaps we’ve come up with a better way to say this. SWM asks, where is the content, in the mind, or the symbols? My response is this – look at the CR problem again. Either there is something missing (“explanatory gap”, “hard problem”, “feelings”) that must be accounted for, or there is something present (“cognition”, “intelligence”, “consciousness”)
– that must be accounted for! Searle presents the parable as if it were in the first category. I applaud the intuition pump, but place the story in the second category – given the ex hypothesi intelligent behavior, just what could possibly account for it? Unfortunately Turing never really touched that, unless you count the TM story as the answer – and even think that is too broad a jump, talk about your explanatory gaps. So, we keep thinking.
Stuart W. Mirsky
Reply to Stevan Harnad’s post of 3/3/11 @ 14:29 I’ll try to be brief (but it’s a challenge). SH: “The constant importation of the other-minds problem (‘Would a T3 robot feel or not?’) has next to nothing to do with it. Whether or not the T3 robot feels, we cannot explain how or why it does.” SWM: As I recall, you mentioned the “other minds” problem a number of times as part of your justification for saying we can’t explain the occurrence of feelings. I’m merely looking at the implications. SH: “I’m not sure what ‘conceptually disconnected’ means, but I mean that feeling is causally superfluous in explaining doing, or the causes of doing.” SWM: I probably didn’t put that well enough. I meant conceptually disconnected because the referents are radically different. Anyway, your argument that a causal explanation of the occurrence of feeling is impossible because “feeling is causally superfluous in explaining doing, or the causes of doing” certainly has the look of circularity, don’t you think? At the least it begs the question since it assumes “causal superfluity” to justify the claim that a causal explanation can’t be deployed. One may not need much more than that to read this as a claim based on intuition. SH: “I have no idea what ‘epiphenomenalism’ means.” SWM: I’m guessing this is hyperbole but, on the off chance it isn’t — it’s the notion that consciousness (what you call “feeling”) is a-causal with regard to the physical workings of brains, merely along for the ride as a corollary of some physical phenomena in brains (kind of like the froth on an ocean wave). Certainly a claim that “feeling” may not accompany what we otherwise recognize as “feeling” behavior in an entity implies that we could have waves with or without the froth! And this is in keeping with the idea of philosophical zombiehood, as Richard Brown has suggested. I am inclined to see your claim in that way, too, at this juncture. SH: “[My response] . . . no more appeals to intuitions than it appeals to zombies, the other-minds problem, or epiphenomenalism. It is not an appeal *to* but an appeal *for* — causal explanation.” SWM: By “appeals to” do you mean “depends on”? Note that it’s not my claim that you make any such appeals explicitly to justify your argument, only that your explicit claims imply these other claims to varying degrees; if so, they need to be considered in light of those implications and what can be concluded from arguments for those positions. SH: “I don’t understand ‘ontic status’ or ‘ontic status type’: There are (roughly) objects, events, actions, properties and states. I suppose a feeling is a kind of state. But that doesn’t help if what we want is a causal explanation.” SWM: My reference was to the tendency to think of (and so want to talk about) feelings in the same way as we do other referents. If we make that move then we want to look for the same kinds of things we expect to find in referents in the public domain (the home of physical things). And now we run aground because, of course, the referents are in no way alike. The only similarity is that both occur within our experience. But they play a different part in that experience so why should we expect to treat them in the same way, or be surprised when we find we really cannot? SH: “Who asked for certainty? . . . I’m with Descartes and the soft skeptics . . . it’s only possible with (1) mathematically provable necessity and (2) the Cogito (‘I feel, therefore there is feeling’). The rest is just probability and inference to the best explanation. “But with feeling, there is no explanation at all.” SWM: There is if you don’t demand a certainty beyond probability and inference (though I go further in accepting the Wittgensteinian analysis — but given your acceptance of probability and inference there’s already enough to achieve an explanation of “feeling”). Anyway, it seems to me that you miss the point of the Wittgensteinian analysis vis a vis how we use language about mental referents. But it’s not even necessary as long as you grant that the probability of the presence of “feeling” is as high as it gets or is ever needed for recognizing the occurrence of “feeling” in an entity. Once there, we can explain what brings it about in entities (and test for it by determining the level of autonomous thought which is dependent on having feeling, etc.). I take it, though, from the tenor of these discussions that you want an account that shows how this or that physical event in brains (or some equivalent platform) just is feeling? Of course, I have suggested that an account which relies on a description of layered processes (computational ones would do, here) which produce a multi-level mapping of the world and of the mapping entity itself (the latter becoming the sense of being a self which the entity possesses), with sufficient interactivity of processes to allow the occurrence of the associative events that continuously update the mappings and depict the elements of the world and relationships between the elements, could be enough. After all, what is feeling but the occurrence of various awarenesses of things at varying levels? Why shouldn’t such an explanation suffice? And yet I think you will still say it does not. The only question then will be whether you give as your reason for such denial something that goes beyond restating the initial claim, that “feeling is causally superfluous in explaining doing, or the causes of doing”.
HOW/WHY THE “HARD PROBLEM” IS HARD
(Reply to Stuart W. Mirsky)
S.MIRSKY: “you mentioned the “other minds” problem a number of times as part of your justification for saying we can’t explain the occurrence of feelings”
I mentioned the other-minds problem to show we can’t expect to do better than Turing-Testing, but not in connection with the fact that we cannot explain why or how we feel.
S.MIRSKY: “your argument that a causal explanation of the occurrence of feeling is impossible because “feeling is causally superfluous in explaining doing, or the causes of doing” certainly has the look of circularity, don’t you think?”
No, it’s open to refutation. Just show how and why feeling’s not causally superfluous in explaining doing, and instead show how and why feeling is needed in order to have a successful explanation (T3, or T4).
S.MIRSKY: “At the least it begs the question since it assumes “causal superfluity” to justify the claim that a causal explanation can’t be deployed. One may not need much more than that to read this as a claim based on intuition.”
It’s open to counterexamples, if anyone can come up with one; but it’s not an appeal to intuition.
S.MIRSKY: “‘epiphenomenalism’ means… feeling is a-causal with regard to the physical workings of brains, merely along for the ride”
I know. I think it’s a bit more ontically circumspect to say that feeling cannot be explained causally than to say it’s acausal…
S.MIRSKY: “you grant that the probability of the presence of “feeling” is as high as it gets or is ever needed for recognizing the occurrence of “feeling” in an entity”
Yes, the correlates are reliable enough (in people, if not quite in T3) — if all we’re interested in is predicting or mind-reading. But if we’re interested in explaining how and why we feel rather than just do, that’s no help.
S.MIRSKY: “I take it… that you want an account that shows how this or that physical event in brains (or some equivalent platform) just is feeling?”
Stuart, it’s a lot more demanding than that! It’s something more along the lines of wanting an explanation of how and why this mechanism *could not do what it can do without feeling*. (But please, please spare me the zombies at this point where you usually invoke our agreement that they are inconceivable/impossible as some sort of justification for not facing the real question!).
Think of feeling as being a “widget” in your causal mechanism. I am simply asking, what would fail, and why, if this widget were not there?
S.MIRSKY: “I have suggested… an account which relies on a description of layered processes… multi-level mapping of the world… the mapping entity itself… the sense of being a self … sufficient interactivity of processes… associative events that continuously update the mappings… depict the elements of the world and relationships between the elements…”
That’s the explanation of how and why it feels rather than just functs? It all sounds like functing to me!
S.MIRSKY: “After all, what is feeling but the occurrence of various awarenesses of things at varying levels?”
There’s the weasel-word again:
Deflated: “After all, what is feeling but the occurrence of various feelings at varying levels?”
S.MIRSKY: “Why shouldn’t such an explanation suffice?”
Because it doesn’t explain. It just interprets (hypothetical) processes which (if there really turned out to be such processes) could be described (and causally explained) identically without supposing they were felt.
(Again, please don’t reply with our agreement that there are no zombies! And it has nothing to do with the other-minds problem either: Our supposition that the processes are felt could be right; that still does not amount to an explanation of how/why they are right.)
S.MIRSKY: “you will still say it does not [explain]. The only question then will be whether you give as your reason… something that goes beyond restating the initial claim, that ‘feeling is causally superfluous in explaining doing, or the causes of doing’”
There is nothing in your explanation of the “multi-level mapping” processes etc., that explains why and how they are felt rather than just functed. You simply assume it, without explanation. That’s probably because it’s all just hypothetical anyway. But even if you actually built a successful T3 and T4 (thereby making Josh Stern happy by providing a “construction”!), and you could actually point to the internal processes in question, it would make no difference. There’s still nothing in your account to explain how or why they are felt rather than functed (even if they truly are felt!).
In response to Stuart Mirsky 3/3 14:35: “But why should [the fact that feelings are categorically private, unlike their neural correlates] preclude a causal explanation?” It doesn’t necessarily. But in one of my responses to Stevan, I suggested that were feelings generated or caused by observable forces or mechanisms, they would be in the public domain as is every other product of causal processes, but they aren’t. This suggests they aren’t caused, produced or generated, but instead entailed (for the system alone) by *being* a system with certain sorts of representational capacities. There are philo-scientific considerations that count in favor of this possibility adduced at http:// www.naturalism.org/appearance.htm
“If an account is given of a set of operations which will produce in a given machine ‘feeling’-related behavior which is convincingly that (in every conceivable way we might need to be convinced), then why would that not be sufficiently “causal” to fit the bill?” The causal account is of the feeling-related *behavior*, which isn’t the same thing as feelings since feelings can exist without any behavior. Feelings (conscious phenomenal states) seem to closely co-vary with certain behavior controlling and cognitive processes going on in the head, see http://www.naturalism.org/kto.htm#Neuroscience But as far as I can tell, there’s no causal story about how feelings get produced or generated by those processes (one of the difficulties with epiphenomenalism, which has it that feelings *are* thus generated). “But if Stevan’s wrong about what’s needed to assure ourselves of the occurrence of the feelings in the machine, then hasn’t a causal explanation (in all the usual senses of “causal”) been provided?” Not necessarily, since as I’ve suggested there might be a non-causal explanation. “…doesn’t your concession that “causal” isn’t relevant because it doesn’t apply in this kind of case amount to an affirmation of Stevan’s position that a “causal” explanation isn’t possible?” Stevan is saying that he hasn’t seen a causal explanation that he finds convincing, *and* he seems convinced that only a causal explanation could be truly explanatory. There may be a causal explanation of consciousness but I’m skeptical for reasons I’ve sketched here and that are set out in more detail at http://www.naturalism.org/appearance.htm#part2 I’ve also suggested in part five of the same paper that there might be viable *non-causal* options for explaining consciousness.
ENTANGLEMENTS
(Reply to Tom Clark)
T.CLARK: “feelings… aren’t caused, produced or generated, but instead entailed (for the system alone) by *being* a system with certain sorts of representational capacities.”
Not caused but entailed? How entailed? Why entailed?
Stevan: “Not caused but entailed? How entailed? Why entailed?” My suggestion is that any complex behavior-controlling representational system (RS) will have a bottom level, not further decomposable, unmodifiable, epistemically impenetrable (unrepresentable) hence qualitative (non-decomposable, homogeneous) and ineffable set of representational elements. They therefore appear as irreducible and indubitable phenomenal realities for the RS alone. The “therefore” is of course completely contestable; I don’t claim to have proved the entailment, only suggested it. The supporting considerations, which you might find of interest and draw extensively from Metzinger’s work, are at
http://www.naturalism.org/appearance.htm#part5
Stuart W. Mirsky
Response to Tom Clark’s post of 3/3/11 @ 15:50 “But why should [the fact that feelings are categorically private, unlike their neural correlates] preclude a causal explanation?”
TC: It doesn’t necessarily. But in one of my responses to Stevan, I suggested that were feelings generated or caused by observable forces or mechanisms, they would be in the public domain as is every other product of causal processes, but they aren’t. This suggests they aren’t caused, produced or generated, but instead entailed (for the system alone) by *being* a system with certain sorts of representational capacities. There are philo-scientific considerations that count in favor of this possibility adduced at http:// www.naturalism.org/appearance.htm
SWM: In this case I would agree with what I take Stevan’s last response to be, to wit, that “entailed” sounds like just another way to speak of “caused”. The question then is just what is it that causes a feeling system (which, by virtue of being that, feels)! ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ “If an account is given of a set of operations which will produce in a given machine ‘feeling’-related behavior which is convincingly that (in every conceivable way we might need to be convinced), then why would that not be sufficiently “causal” to fit the bill?” TC: The causal account is of the feeling-related *behavior*, which isn’t the same thing as feelings since feelings can exist without any behavior. SWM: Yes, they can. But we aren’t looking for only non-behavior related feelings but for feelings period. Thus any feelings will do and behavior-related ones fit that bill. Stevan’s argument is that that isn’t enough to assure ourselves that our T3 entity isn’t a robot (that is, isn’t a zombie) and so, absent such assurance, all we have is an account of caused behavior. It is there I have mainly disagreed with him. I’m not sure where you are on this point though. Are feelings in others recognized through behavior and is that enough to determine that feelings are present? TC: . . . as far as I can tell, there’s no causal story about how feelings get produced or generated by those processes (one of the difficulties with epiphenomenalism, which has it that feelings *are* thus generated). SWM: Epiphenonmenalism aside, at least one issue worth exploring here is whether a process-based systems account is enough to explain the occurrence of feelings. Note that I am not arguing that any such account is true, only that, if it were true, it would be enough, contra Stevan. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ “But if Stevan’s wrong about what’s needed to assure ourselves of the occurrence of the feelings in the machine, then hasn’t a causal explanation (in all the usual senses of “causal”) been provided?” TC: Not necessarily, since as I’ve suggested there might be a non-causal explanation. SWM: My only point is that it would be enough IF IT WERE TRUE. This isn’t about what is true but about what would suffice if it were. Your non-causal explanation, on the other hand, only looks like a variation in the terms because it uses “entailment”, which is generally applied to questions of logic, to substitute for the “cause” word in this case. Of course, language does allow such flexibility but we need to be very careful that we don’t let it paper over real distinctions or suggest distinctions that aren’t there. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ “…doesn’t your concession that “causal” isn’t relevant because it doesn’t apply in this kind of case amount to an affirmation of Stevan’s position that a “causal” explanation isn’t possible?” TC: Stevan is saying that he hasn’t seen a causal explanation that he finds convincing, *and* he seems convinced that only a causal explanation could be truly explanatory. There may be a causal explanation of consciousness but I’m skeptical for reasons I’ve sketched here and that are set out in more detail at http://www.naturalism.org/appearance.htm#part2 I’ve also suggested in part five of the same paper that there might be viable *non-causal* options for explaining consciousness. SWM: I agree he is saying he hasn’t seen one that he finds convincing. Note that he has also said that he believes none is possible. When presented with possible explanations of this type, he denies their possibility. I take that to be a significant problem with his argument. But when you substitute “entail” for “cause”, it seems to me that, insofar as you aren’t simply substituting one term for another, you are already granting his claim that causal explanations of feeling aren’t possible.
Stuart W. Mirsky
Reply to Stevan Harnad’s comments of March 3, 2011 at 16:54 S.MIRSKY: “your argument that a causal explanation of the occurrence of feeling is impossible because “feeling is causally superfluous in explaining doing, or the causes of doing” certainly has the look of circularity, don’t you think?” SH: No, it’s open to refutation. Just show how and why feeling’s not causally superfluous in explaining doing, and instead show how and why feeling is needed in order to have a successful explanation (T3, or T4). SWM: Given the “lifetime Turing Test” we’ve discussed, an entity could not keep up the simulation across the full range of behavioral requirements. That is, feeling in the sense the term is used here is an essential ingredient in the entity’s actions. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “At the least it begs the question since it assumes “causal superfluity” to justify the claim that a causal explanation can’t be deployed. One may not need much more than that to read this as a claim based on intuition.” SH: It’s open to counterexamples, if anyone can come up with one; but it’s not an appeal to intuition. SWM: An entity that met the Commander Data test would be the counterexample — unless you assume that absence of access equals implies at least the possibility of the lack of presence. If you don’t then any explanation that accounted for the Data entity would suffice. I think there’s a confusion in your position, Stevan. It’s as if you think that, because there’s a difference between having feelings and “reading” them in others, there’s no reason to grant their presence when everything else that’s relevant testifies to that presence. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “you grant that the probability of the presence of “feeling” is as high as it gets or is ever needed for recognizing the occurrence of “feeling” in an entity” SH: Yes, the correlates are reliable enough (in people, if not quite in T3) — if all we’re interested in is predicting or mind-reading. But if we’re interested in explaining how and why we feel rather than just do, that’s no help. SWM: If we feel because systems of a certain type have feeling, then any explanation for how such a system has feeling is enough. The kind of system example I have offered includes just such an explanation of what feeling is. Below I see you call that into question so I’ll delay amplifying my point until then. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “I take it… that you want an account that shows how this or that physical event in brains (or some equivalent platform) just is feeling?” SH: Stuart, it’s a lot more demanding than that! It’s something more along the lines of wanting an explanation of how and why this mechanism *could not do what it can do without feeling*.. . . Think of feeling as being a “widget” in your causal mechanism. I am simply asking, what would fail, and why, if this widget were not there? SWM: The entity would fail to pass the lifetime Turing Test at some point, however good a simulation it was. Why would it fail? It would lack the underpinnings that lead to the requisite behaviors. The theory proposes that the underpinnings are accomplished by installing certain functionalities in the system’s processes. If those functionalities were not added or didn’t work as specified or weren’t enough to generate the requisite mental life, the tested entity would fail. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “I have suggested… an account which relies on a description of layered processes… multi-level mapping of the world… the mapping entity itself… the sense of being a self … sufficient interactivity of processes… associative events that continuously update the mappings… depict the elements of the world and relationships between the elements…” SH: That’s the explanation of how and why it feels rather than just functs? It all sounds like functing to me! SWM: I can’t make sense of your neologism, “functing”. It just sounds like a way of avoiding acknowledging feeling to me. I agree that not all processes or process-based systems will feel. I am only arguing that feeling could be achievable via such an approach, insofar as a system can be gotten to the point where it replicates the things brains manage to do. I make no claim here that this is true or has been done, only that such a theory COULD explain the occurrence of feeling in an entity. But, as I’ve suggested before, there is an underlying conception of consciousness or feeling that may inform our different intuitions about this (why it seems to work for me but not for you). I have no problem thinking about feeling in terms of the various processes that it could break down to and, indeed, when I do that, I see nothing left out. It seems to me that you either do have a problem with this notion (that feeling is analyzable into non-feeling constituent processes performing particular tasks) or you think something is left out that I don’t see. If the latter, can you say what it is other than just the feeling itself? If not, then I must assume our difference lies in your discomfort with the idea that feeling is analyzable into constituents more basic than it is. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “After all, what is feeling but the occurrence of various awarenesses of things at varying levels?” SH: There’s the weasel-word again: Deflated: “After all, what is feeling but the occurrence of various feelings at varying levels?” SWM: A linguistic problem again. I don’t like the term “feeling” because it seems misleading to me. So I try to narrow it down by offering to describe it in terms of what I think is clearer. Since you think “feeling” is the clearest we can get, you don’t accept further analysis. But let’s stipulate for the moment that we mean exactly the same thing by “feeling” and “awareness”, and disregard all the other uses that both words lend themselves to. This gets to the question of reducibility of the phenomenon. Above I asked if you disagree with me about the possibility of analyzing feeling into constituent elements which aren’t feeling, or if you thought something is left out in my analysis that is not reducible. In either case though we come to a point where you reject my reducing instances of having feeling to instances of being aware. Isn’t this just a claim that feeling is irreducible? And doesn’t this then take us back to the core problem, that you see consciousness (feeling) as a bottom line basic, whereas I don’t? But if you do, then isn’t that fundamentally a dualist position? It would certainly explain why we cannot agree on common ground here and why you bridle at my suggestion that to describe a system that causes feeling behavior is to provide a causal explanation of feeling. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “Why shouldn’t such an explanation suffice?” SH: Because it doesn’t explain. It just interprets (hypothetical) processes which (if there really turned out to be such processes) could be described (and causally explained) identically without supposing they were felt. SWM: It would suppose they were felt if the measure is feeling behavior and the entity passes the lifetime Turing Test. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “you will still say it does not [explain]. The only question then will be whether you give as your reason… something that goes beyond restating the initial claim, that ‘feeling is causally superfluous in explaining doing, or the causes of doing’” SH: There is nothing in your explanation of the “multi-level mapping” processes etc., that explains why and how they are felt rather than just functed. You simply assume it, without explanation. . . . even if you actually built a successful T3 and T4 . . . There’s still nothing in your account to explain how or why they are felt rather than functed (even if they truly are felt!). SWM: The explanation lies in the notion that what we call feeling is just the conjoining of certain processes performing certain functions in a certain way. When the entity feels, it is having certain representations and doing something with them at a system level. We see the feeling in the feeling behavior in the context of what has been built into it in terms of process-based systems and we ascertain its presence to our satisfaction via appropriate testing of the behaviors. Nothing is left out unless you believe that feeling is irreducible and then you have the problem of defending a dualist position (which you have yet to take a position on or defend).
EXPLANATION VS WHETHER FORECASTING
(Reply to Stuart W. Mirsky)
S.MIRSKY: “An entity that met the Commander Data test would be the counterexample”
Commander Data is just T3. The counterexample is to provide a causal explanation of why and how we (or T3) feel (rather than just funct).
S.MIRSKY: “It’s as if you think that, because there’s a difference between having feelings and “reading” them in others, there’s no reason to grant their presence when everything else that’s relevant testifies to that presence.”
Stuart, there’s a point that you are systematically not taking on board: I am talking about explaining *how and why* we (or T3) feel (rather than just funct). You keep answering about *whether* we (or T3) feel.
We will not make any progress this way…
S.MIRSKY: “If we feel because systems of a certain type have feeling, then any explanation for how such a system has feeling is enough.”
Yes indeed; and I am waiting patiently for such an explanation…
S.MIRSKY: “[An unfeeling] entity would fail to pass the lifetime Turing Test at some point, however good a simulation it was. Why would it fail? It would lack the underpinnings that lead to the requisite behaviors. The theory proposes that the underpinnings are accomplished by installing certain functionalities in the system’s processes. If those functionalities were not added or didn’t work as specified or weren’t enough to generate the requisite mental life, the tested entity would fail.”
Well, Stuart, not to pull a Josh-Stern on you, but that’s rather a mouthful given that we have neither a working T3 “construction” based on your processes, nor a demonstration that without your processes it couldn’t pass T3!
But let’s suppose it’s so: You’ve built T3, and you’ve shown that if you don’t have processes in it of the type you describe, it fails T3. Now, how have you explained how and why it feels (rather than just how and why it passes T3)?
(It’s always the same thing: you conflate the ontic question of the causal status of feeling with the epistemic question of whether or not it feels.)
S.MIRSKY: “I can’t make sense of your neologism, “functing”. It just sounds like a way of avoiding acknowledging feeling to me.”
Actually, it’s a way of avoiding question-begging: Until further notice, whatever internal processes it takes to pass T3 or T4 are processes required to pass T3 or T4, i.e., processes required to generate the right doing. They do not explain why or how one has to feel to be able to do all that. (Reminder, even if the processes are perfectly correlated with feeling in people, and even if armed with them the robot can pass T3 or T4, and without them it can’t, it still does not explain how and why the processes are felt rather than just functed. This is not about whether they are felt, but about how and why.
S.MIRSKY: “I have no problem thinking about feeling in terms of the various processes that it could break down to and, indeed, when I do that, I see nothing left out.”
I’ve noticed, and I’m suggesting that it’s a problem that you have no problem with that…
S.MIRSKY: “It seems to me that you… think something is left out that I don’t see. If the latter, can you say what it is other than just the feeling itself?”
The feeling itself — in particular, how and why it’s there (agreeing for the sake of argument that it *is* there).
S.MIRSKY: “let’s stipulate for the moment that we mean exactly the same thing by ‘feeling’ and ‘awareness’”
Ok, but then that means that all the transcriptions I’ve done, swapping feeling for being aware, etc. have to be faced. And many of the transcriptions no longer make sense (e.g., “unfelt feeling”).
S.MIRSKY: “I asked if you disagree with me about the possibility of analyzing feeling into constituent elements which aren’t feeling”
I don’t even know what that means. Even if feeling were reducible to mean kinetic energy I still wouldn’t know how or why it was felt rather than functed. After all, heat, which really is mean kinetic energy, is just functed.
S.MIRSKY: “you reject my reducing instances of having feeling to instances of being aware”
Why would I reject reducing feeling to feeling (as we’ve agreed)? But at some point I’d like to move from reducing to explaining….
S.MIRSKY: “Isn’t this just a claim that feeling is irreducible?”
Give me a “reductive explanation” that’s explanatory and I’ll happily accept. But the reduction that “this feeling *just is* that functing” certainly does not do it for me.
[We're repeating ourselves dreadfully, and spectators must be really weary, but here's perhaps a tiny injection of something new: I've always hated fervent talk about "emergence" -- usually by way of invoking some analogy in physics or biology, in which a phenomenon -- like heat -- is explained by a "reductive" explanation -- like mean kinetic energy. Yes, there are unpredicted surprises in physics and biology. What looked like one kind of thing turns out to be another kind of thing, at bottom. But one of the invariant features of this sort of reduction and emergence is that it's all functing to functing. Heat is functing, and so is mean kinetic energy. Now I don't want to insist too much on this, but when we say that the emergent property was "unexpected," I think we are talking about appearances, which means we are talking about feelings: Heat did not *feel* like mean kinetic energy. Fine. But then we learn to think of it that way, as a kind of difference in scale, and the unexpected no longer looks so unexpected. What the emergent functing feels like is integrated with what the lower-level functing feels like. Well, I just want to suggest that this sort of thing may not work work quite as well when it is feeling itself that we are trying to cash in into functing. That just *might* be part of the reason why causal explanation is failing us here. But, if so, I don't think it's the whole reason.]
S.MIRSKY: “And doesn’t this then take us back to the core problem, that you see consciousness (feeling) as a bottom line basic, whereas I don’t? But if you do, then isn’t that fundamentally a dualist position?”
If it is, then it’s because I’m an epistemic (i.e., explanatory) dualist, not an ontic one. Just as I don’t believe in zombies, I fully believe that the brain causes feeling, with no remainder, causally. It’s just that we don’t seem to be able to explain — in the usual way we do with all other functing, whether or not “emergent,” — how and why the brain causes feeling. It’s duality in what can and can’t be explained causally, rather than a duality either in “substance” or in “properties” (unless a “property dualist” is defined as someone who holds that the difference between feeling and all other properties is that all other properties can be explained causally, whereas feeling cannot: but in that case I’d still rather write that out longhand than dub myself a “property dualist” without an explanation!) In closing, I will leave you with (what I now hope will be) the last word:
S.MIRSKY: “The explanation lies in the notion that what we call feeling is just the conjoining of certain processes performing certain functions in a certain way. When the entity feels, it is having certain representations and doing something with them at a system level. We see the feeling in the feeling behavior in the context of what has been built into it in terms of process-based systems and we ascertain its presence to our satisfaction via appropriate testing of the behaviors. Nothing is left out unless you believe that feeling is irreducible and then you have the problem of defending a dualist position (which you have yet to take a position on or defend)."
Stuart W. Mirsky
SH: Commander Data is just T3. The counterexample is to provide a causal explanation of why and how we (or T3) feel (rather than just funct). SWM: The causal explanation lies in Data’s specs. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “It’s as if you think that, because there’s a difference between having feelings and “reading” them in others, there’s no reason to grant their presence when everything else that’s relevant testifies to that presence.” SH: Stuart, there’s a point that you are systematically not taking on board: I am talking about explaining *how and why* we (or T3) feel (rather than just funct). You keep answering about *whether* we (or T3) feel. SWM: The “whether” question is relevant to the “why” question since your denial of my answer as to why always seems to hinge on the notion that we can’t know what’s really going on inside our Commander Data type entities (or others like ourselves). Of course, I’m arguing we can know. Therefore the specs of such entities, combined with what we can know about them from observing behaviors, gives us the full explanation: We know X causes Y when Y is seen to be present with the implementation of X. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ SH: We will not make any progress this way… SWM: That’s so if we can’t even agree on what counts as an explanation and what needs to be included to reach the conclusion that the explanation works. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “If we feel because systems of a certain type have feeling, then any explanation for how such a system has feeling is enough.” SH: Yes indeed; and I am waiting patiently for such an explanation… SWM: I don’t believe, at this point, that you would ever agree that an explanation involving the specs of a Data-type entity, combined with the evidence of feeling behavior in that entity (when the specs are implemented), would ever suffice. It has to do with what I take to be the conception of consciousness (feeling) you hold which differs fundamentally from my conception. I don’t see how we get past this fundamental difference. I will say this though. I once held the same view you apparently hold (or one very much like it). I don’t think I finally shifted until I was
confronted with Searle’s CRA which I initially found convincing. But the more I thought about it, the more mistaken it seemed to me to be. So Searle was my catalyst for adopting a different idea of consciousness from the one you seem to hold. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “[An unfeeling] entity would fail to pass the lifetime Turing Test at some point, however good a simulation it was. Why . . . ? It would lack the underpinnings that lead to the requisite behaviors. The theory proposes that the underpinnings are accomplished by installing certain functionalities in the system’s processes. If those functionalities were not added or didn’t work as specified or weren’t enough to generate the requisite mental life, the tested entity would fail.” SH: Well, Stuart, not to pull a Josh-Stern on you, but that’s rather a mouthful given that we have neither a working T3 “construction” based on your processes, nor a demonstration that without your processes it couldn’t pass T3! SWM: It’s not claim as to what’s true. It’s a basis for assessing the success or failure of the proposed approach, for measuring the predictive power of the system theory of mind. That’s how science works after all. We formulate hypotheses and test them. I don’t expect any more than that for the approach or claim more for it. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ SH: But let’s suppose it’s so: You’ve built T3, and you’ve shown that if you don’t have processes in it of the type you describe, it fails T3. Now, how have you explained how and why it feels (rather than just how and why it passes T3)? (It’s always the same thing: you conflate the ontic question of the causal status of feeling with the epistemic question of whether or not it feels.) SWM: The only way to scientifically arrive at an outcome is to test it. So what we can know about this, is relevant. Everything else is secondary. After all, you can’t arrive at a proper theory a priori! ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “I can’t make sense of your neologism, “functing”. It just sounds like a way of avoiding acknowledging feeling to me.” SH: Actually, it’s a way of avoiding question-begging: Until further notice, whatever internal processes it takes to pass T3 or T4 are processes required to pass T3 or T4, i.e., processes required to generate the right doing. They do not explain why or how one has to feel to be able to do all that. (Reminder, even if the processes are perfectly correlated with feeling in people, and even if armed with them the robot can pass T3 or T4, and without them it can’t, it still does not explain how and why the processes are felt rather than just functed. This is not about whether they are felt, but about how and why. SWM: There’s that neologism again. If feeling is reliably recognized in behavior and the right system yields the right behavior, then whatever theory underlies construction of the system explains the feeling manifested in the behavior. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “I have no problem thinking about feeling in terms of the various processes that it could break down to and, indeed, when I do that, I see nothing left out.” SH: I’ve noticed, and I’m suggesting that it’s a problem that you have no problem with that… SWM: Yes, I can see that’s your view. It’s why we’re apparently at opposite ends of this particular opinion spectrum vis a vis what consciousness is. But what you can’t say (or so far haven’t said) is why it’s a problem except to reiterate your view that T3 is about functing not feeling, and so forth. But that is just to take your stand on your bottom line conception of consciousness which differs from mine. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “It seems to me that you… think something is left out that I don’t see. If the latter, can you say what it is other than just the feeling itself?” SH: The feeling itself — in particular, how and why it’s there (agreeing for the sake of argument that it *is* there). SWM: I said “other than just the feeling itself”. As to the how and why, the type of theory I’ve offered gives us both the how and the why, i.e., it says feeling is just so many physical processes doing such and such things in the right way. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “let’s stipulate for the moment that we mean exactly the same thing by ‘feeling’ and ‘awareness’” SH: Ok, but then that means that all the transcriptions I’ve done, swapping feeling for being aware, etc. have to be faced. And many of the transcriptions no longer make sense (e.g., “unfelt feeling”). SWM: You won’t find me arguing for “unfelt feeling”. That’s not my game. I think feeling is explicable in terms of processes and functions, i.e., that the sense entities like us have of things, including of having senses of things, all reduce back to perfectly physical processes and that computational processes running on computers are as good a candidate for this as brain processes. But I don’t see how “unfelt feeling” can be intelligible except in some special sense (like a suppressed emotion that we don’t admit to having but which has an impact on other things we feel and do as in psychoanalytical claims). ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “I asked if you disagree with me about the possibility of analyzing feeling into constituent elements which aren’t feeling” SH: I don’t even know what that means. Even if feeling were reducible to mean kinetic energy I still wouldn’t know how or why it was felt rather than functed. After all, heat, which really is mean kinetic energy, is just functed. SWM: If feeling is a bottom line thing, irreducible to anything other than itself, then it is, to all intents and purposes non-physical and stands apart from the rest of the otherwise physical universe. If it’s reducible then it doesn’t. The former is an expression of dualism, the latter isn’t. If feeling is reducible in this way, then it’s conceivable that it’s just the result of so many non-feeling processes happening in a certain way, in which case we have an explanation of feeling in response to your challenge. If it’s bottom line, on the other hand, an ontologically basic phenomenon in the universe, then we would not expect to see it reduce to anything else and, of course, your thesis that it is not causally explainable would be sustained. I submit that the only reason you think your challenge cannot be met is because you think feeling IS bottom line in this way. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “you reject my reducing instances of having feeling to instances of being aware”
SH: Why would I reject reducing feeling to feeling (as we’ve agreed)? But at some point I’d like to move from reducing to explaining…. SWM: I was using “aware” as a way of more precisely zeroing in on your use of “feeling” in the statement of mine you are citing here. However, in what followed that statement, I shifted to the stipulative substitution we had formerly agreed on. (This is all a function of the imprecision of language in this arena. You think “feeling” is more precise, I think “awareness” is but the truth is it just depends on how we each agree to use the terms we like at any given point — and sometimes, because of the fuzziness of language, we are going to slip up, despite ongoing agreements.) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “Isn’t this just a claim that feeling is irreducible?” SH: Give me a “reductive explanation” that’s explanatory and I’ll happily accept. But the reduction that “this feeling *just is* that functing” certainly does not do it for me. SWM: ‘This feeling is that combination of processes doing such and such things within the larger system.’ Insofar as the theory implied by that statement bears out in terms of a successful implementation (Commander Data), we have a successful explanation. But you reject that it is successful because you cannot agree that feeling is reducible to elements that aren’t themselves feeling. SH: [We're repeating ourselves dreadfully, and spectators must be really weary, SWM: Yes but it's almost over so a rest is in the cards for all! SH: but here's perhaps a tiny injection of something new: I've always hated fervent talk about "emergence" -- usually by way of invoking some analogy in physics or biology, in which a phenomenon -- like heat -- is explained by a "reductive" explanation -- like mean kinetic energy. Yes, there are unpredicted surprises in physics and biology. What looked like one kind of thing turns out to be another kind of thing, at bottom. But one of the invariant features of this sort of reduction and emergence is that it's all functing to functing. Heat is functing, and so is mean kinetic energy. Now I don't want to insist too much on this, but when we say that the emergent property was "unexpected," I think we are talking about appearances, which means we are talking about feelings: Heat did not *feel* like mean kinetic energy. SWM: That's because heat and "mean kinetic energy" are different concepts. One refers to what we feel under certain conditions (what burns us, melts us, exhausts us, etc.), the other to an atomic level description of what underlies the phenomenon we feel. They can refer to the same thing without denoting precisely the same things. That in some contexts we want to say they do denote the same thing doesn't mean they do in all contexts in which we can use these terms. SH: Fine. But then we learn to think of it that way, as a kind of difference in scale, and the unexpected no longer looks so unexpected. What the emergent functing feels like is integrated with what the lower-level functing feels like. Well, I just want to suggest that this sort of thing may not work work quite as well when it is feeling itself that we are trying to cash in into functing. That just *might* be part of the reason why causal explanation is failing us here. But, if so, I don't think it's the whole reason.] SWM: I think it’s an article of faith on your part that “causal explanation is failing us here”. In fact I think it’s perfectly feasible — as long as one is willing to grant at least the possibility that feeling is analyzable into constituent elements that aren’t, themselves, feeling. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ S.MIRSKY: “And doesn’t this then take us back to the core problem, that you see consciousness (feeling) as a bottom line basic, whereas I don’t? But if you do, then isn’t that fundamentally a dualist position?” SH: If it is, then it’s because I’m an epistemic (i.e., explanatory) dualist, not an ontic one. Just as I don’t believe in zombies, SWM: But you do believe in at least the possibility of T3 zombies which is the same thing! SH: I fully believe that the brain causes feeling, with no remainder, causally. SWM: Here we find agreement. The only problem, then, is whether, in believing that, you are also prepared to grant that in causing consciousness, the brain does so physically and that it is at least possible that it does this via its known physical processes. If so, then there are significant similarities between those processes and computer processes — enough, in fact, to suggest that computers might be able to achieve the same thing as brains achieve. And if THAT is so, then it’s hardly unreasonable to suppose they do it via a concatenation of processes constituting a system as I’ve already suggested. Thus the kind of system I’ve proposed COULD provide an account for how feeling is caused. SH: It’s just that we don’t seem to be able to explain — in the usual way we do with all other functing, whether or not “emergent,”
— how and why the brain causes feeling. It’s duality in what can and can’t be explained causally, rather than a duality either in “substance” or in “properties” (unless a “property dualist” is defined as someone who holds that the difference between feeling and all other properties is that all other properties can be explained causally, whereas feeling cannot: but in that case I’d still rather write that out longhand than dub myself a “property dualist” without an explanation!) SWM: Understood. Searle rejects the designation too. As far as I can see, if “property dualism” is just about there being some properties of systems that are what you call feeling, then it’s a fairly innocuous position. But if you think there’s no possibility of giving a causal account of feeling in terms of systems and system-level properties, then the dualism, admitted or not, looks like it goes a lot deeper and is certainly more problematic. It all rides, I think, on whether we grant that feeling is reducible or not to constituents that aren’t, themselves, feeling. Is feeling a bottom line property that somehow springs into being or attaches itself to some physical phenomena/events in an ultimately inexplicable way or is it just the manifestation of some combination of certain non-feeling constituent processes/functions? SH: In closing, I will leave you with (what I now hope will be) the last word: S.MIRSKY: “The explanation lies in the notion that what we call feeling is just the conjoining of certain processes performing certain functions in a certain way. When the entity feels, it is having certain representations and doing something with them at a system level. We see the feeling in the feeling behavior in the context of what has been built into it in terms of process-based systems and we ascertain its presence to our satisfaction via appropriate testing of the behaviors. Nothing is left out unless you believe that feeling is irreducible and then you have the problem of defending a dualist position (which you have yet to take a position on or defend). SWM: If you accept this view then wer’e on the same page which we haven’t seemed to be throughout these discussions. I take it
that was not the purpose of ending your response here! Who knows though, eh? I have enjoyed our conversations, even if it turns out we have found no common ground, and I thank you for taking the time and making the no doubt considerable effort to consider and respond to my comments.
Philip Goff
On one version of the view, the a posteriori physicalists think *the very feeling of pain* is identical to the functional property. That very phenomenal concept of pain you form when you reflect on your pain and think about it in terms of what it feels like, that very concept refers to the functional property. So to you question: ‘Why do we feel pain as opposed to just funct it?’, Papineau would answer, ‘The property of feeling pain and the property of functing pain are one and the same property. We just have two concepts for picking out a single property’. Papineau thinks it’s in some sense psychologically impossible for us to believe this identity, just as it may be in some sense psychologically impossible to believe that time doesn’t flow.
FUNCTIONAL REDUNDANCY: FAILURE OF IMAGINATION OR FAILURE OF EXPLANATION?
(Reply to Philip Goff)
P.GOFF: “a posteriori physicalists think *the very feeling of pain* is identical to the functional property… Papineau would [say] ‘The property of feeling pain and the property of functing pain are one and the same property. We just have two concepts for picking out a single property’… it’s in some sense psychologically impossible for us to believe this identity, just as it may be in some sense psychologically impossible to believe that time doesn’t flow.”
I can believe that feeling and its functed correlates are somehow the same thing (or “aspects” of the same thing, whatever that means) — in fact, I *do* believe it: I just don’t know how or why it is so.
But I agree that any view whose punchline is that the explanatory gap is unbridgeable can be considered a notational variant of any other view with the same punchline.
I just think that the explanatory failure itself is the heart of the problem, rather than just a symptom of some specific cognitive deficit we happen to all have regarding feeling and functing. It’s not our imaginations or intuitions but *causal explanation* — usually such a reliable guide — that’s failing us, in the special case of feeling.
And, if anything, recognizing that explanatory shortfall goes *against* our intuitions: What we feel intuitively is that we are active causal forces, not just redundant decorations on functing. Yet, once we learn (and we do *learn* it: we certainly don’t feel it naturally of our own accord) that functing is in fact sufficient to do the whole job, and there’s no causal room left, it’s because we *can* in fact successfully conceive and believe that, that we are impelled to ask: Well then why on earth is all that functing felt?
I’d say the “why” question (which is not a teleological question but a functional and evolutionary — hence practical, mechanical — question) is even more puzzling than the “how” question (though it is, of course, just a special case of the “how” question):
As laymen, we’re always ready to finesse the technical details underlying the scientific answer to “how.” (How many people really understand the fine points of electromagnetism or phase-transitions or space-time?) We simply accept the science’s punchline. But in the special case of feeling, it is the very superfluity and redundancy of feeling in the face of the perfectly sufficient and complete explanation of the functional substrate (even with the technical details taken on faith) that makes ask: Well then why on earth is all that functing felt?
It’s rather difficult to put this without inventing a sci-fi scenario so ad-hoc and fanciful that it defies belief, but here’s an attempt to give the flavor of the sense in which one could duplicate the puzzle of the redundant causality underlying the explanatory gap in a domain other than feeling, namely, doing:
Suppose we were told that there is a planet in which billiard balls move according to the usual laws of mechanics — what looks macroscopically like direct contact, collisions, and local transfer of kinetic energy. But, in addition to the usual universal laws of mechanics underlying those local collisions, there were also empirical evidence of a second mechanical force on that planet, unknown here on earth or elsewhere in the universe, whereby the outcomes of collisions are determined not just by the local mechanics, but by an additional macroscopic action-at-a-distance force, rather like gravity, except a repulsive rather than an attractive force. So when a billiard ball is headed toward a collision, long before the local mechanics are engaged by contact, there is already a (detectable) build-up of the action-at-a-distance force, likewise sufficient to determine the outcome of the collision. The collision occurs just as a classical Newtonian 2-body collision would occur on earth, the outcome perfectly predicted and explained by Newtonian 2-body mechanics, but, in addition, the outcome is also perfectly (and identically) predicted and explained by the action-at-a-distance force.
I suggest that under those conditions it would be quite natural for us to ask why there are two forces when the outcome would be identical if there were just one.
It’s rather like that with feeling and functing — and that’s not because we have trouble believing that functing causes feeling, somehow. We can even finesse the technical details of the “how”. It’s the causal redundancy underlying the “why” that is the real puzzle.
Nor is that resolved by the overhasty invocation of “identity” (pain is identical to c-fibre firing). At best, pain’s another property of c-fibre firing, besides the property of firing. And even if thinking of properties of things as being “caused” is rejected as simplistic — and we think instead of properties as just being “had” rather than being caused — the redundancy is still there: Why does c-fibre firing “have” this extra property of *feeling* like something, when just *doing* the firing is already enough to get whatever functional/ adaptive job there is to be done, done?
(To advert to your own thread, Philip: I also think that one of the allures of “panpsychism” — which I actually think is incoherent — is the wish at least to spread this intractably and embarassingly redundant property of feeling all over the universe, for that would make it somehow more credible, respectable, as a universal redundant property, rather than just an inexplicable local terrestrial fluke. But I think that in that case the cure is even worse than the disease!)
Miguel Sebastian
Hi Harnad, One minor point: what do you mean by an extra-property? IF the property of being in pain WERE the property of having c-fibers firing, there wouldn’t be any extra property (there are well-know arguments against this kind of identification that the phenomenal concept strategy -PCS- tries to answer). If the former statement were true, every (metaphysically) possible situation in which one have C-fibers firing is a situation in which one in pain. It is not possible to have C-fibers firing without being in pain. On the plausible view that necessary coextensive properties are identical, there is no extra-property. According to the PCS there are no two aspect of one “thing” there is just one “thing” and two ways of thinking about this unique “thing” (as pain and as c-fibers firing).
MULTIPLE PROPERTIES
(Reply to Miguel Sebastian)
M.SEBASTIAN: “what do you mean by an extra-property?” IF the property of being in pain WERE the property of having c-fibers firing, there wouldn’t be any extra property (there are well-know arguments against this kind of identification that the phenomenal concept strategy -PCS- tries to answer). If the former statement were true, every (metaphysically) possible situation in which one have C-fibers firing is a situation in which one in pain. It is not possible to have C-fibers firing without being in pain. On the plausible view that necessary coextensive properties are identical, there is no extra-property. According to the PCS there are no two aspect of one “thing” there is just one “thing” and two ways of thinking about this unique “thing” (as pain and as c-fibers firing).”
I confess I can’t understand or follow the metaphysical technicalities. I just mean something like: A red circle has many properties, among them that it is red and that it is round. C-fibre activity, too, could have many properties, among them that they are firing and that they are felt.
Stuart W. Mirsky
THOSE WEASEL WORDS: I didn’t want to end without addressing Stevan Harnad’s persistent complaint about “weasel words”. Repeatedly in this discussion he has taken me to task for using words other than “feeling” for the consciousness this is all about (this being the On-line Consciousness Conference, after all, not the On-line Feeling Conference). I understand his concern here because there’s little doubt that the words words we fall back on when referring to elements of our mental lives are notoriously slippery. The only question is whether “feeling” is intrinsically better than any of the alternatives as Stevan insists (and often in these discussions a great deal has seemed to ride on that for him). In his opening video he presented a whole list of words often used synonymously for this mysterious thing we typically call “consciousness” including “consciousness” itself, which he lumped in with all the other allegedly “weasel words”. But Stevan thinks that “feeling” alone isn’t weasely. It is, on his view, the most direct, clear, appropriate term available and everything else we want to say about consciousness seems to him to come down as some variant of what could be better called “feeling”. In particular, he has often objected to my use of “awareness” on the grounds that it suggests other things, such as attending to, which detract from the feeling aspect that he believes lies at the core of the actual intended meaning. But “feeling”, too, has other meanings, doesn’t it? Surely these admit of the same problem in specificity and intent. For instance, off hand I can think of six different ways we use “feeling: 1) When we reach out and touch something we are said to be feeling it. 2) When we get the sensation that comes with touching it under normal conditions we are said to feel it.
3) When we have any sensations at all, we are said to feel it, too, (by analogy with the narrower tactile sensory experience). 4) When we are in a particular emotional state or condition we are said to feel a certain way. 5) When we want to do something or say something we may describe this as feeling like doing or saying. 6) When we think about anything (have it as an object of thought in our mental sights, as it were) we may be said to feel it (Stevan’s main use, I think) — as in understanding a given symbol’s semantic content (meaning) having the character of our being aware of that symbol’s meaning. By contrast, my own preferred term, “awareness”, seems to offer just two possibilities: 1) Attending to something that has entered our range of observation (either in a sensory sense or in a conceptual sense — as when some idea may come to our attention) 2) Having a sense of anything at all as when we experience sensory phenomena or ideas. In fact, it rather looks like there are better terms yet, such as “experience” though this, too, has its alternative uses: 1) Whenever we go through any series of events of which we are aware (as in paying at least some level of attention to) we can be said to be having experience. 2) The elements of our mental lives characterized by daydreams, streams of consciousness, etc. 3) The phenomenon of being a subject, which seems to contain all other phenomena we encounter, whether objectively observable physical objects or events, or the mental imagery and emotions that characterize our interior lives including imaginary and remembered imagery. One common thread here is “subjective” (and its cognates) and another is “mental” while, of course, “consciousness” keeps recurring. On Stevan’s view such persistent shifting among terms and the dependence for meaning on others in this group is indicative of weaseliness. And yet, what comes clear I think from all this is that “feeling” is no better in that it is no more basic a term than the others. We may agree for argument’s sake to stick to a common term such as we have often tried to do here in these discussions, in deference to Stevan’s preference, and yet even then this has proved hard. Stevan often took me to be straying from his “feeling” when I lapsed into my preferred term of “awareness” and I often took Stevan’s “feeling” to be too broad to be helpful. Sometimes Stevan himself erred as when he mixed “feeling” as in migraine headaches (a sensory phenomenon) with “feeling” as in what Searle says we must have if we are to be said to, in fact, understand the Chinese ideograms fed to us in the Chinese Room. So I guess I want to say that stipulating to a term’s meaning for the purpose of a particular discussion or discussions may be a useful strategem, but it isn’t a cure-all for the deeper problem of language growing rather loosey-goosey at the margins of what is its natural habitat in the public realm. And mind words are clearly outside that habitat.
THE EXPLANANDUM
(Reply to Stuart W. Mirsky)
S.MIRSKY: “But “feeling”, too, has other meanings, doesn’t it?”
Many feelings, one meaning: We can feel lots of different kinds of things. What needs to be explained is how and why any of them feel like anything at all.
S.MIRSKY: “By contrast, my own preferred term, “awareness”, seems to offer just two possibilities… Attending to something that has entered our range of observation (either in a sensory sense or in a conceptual sense — as when some idea may come to our attention)”
Demustelation: “attending” (felt or unfelt?); “observing” (felt or unfelt?); “sensing” (felt or unfelt?); “idea” (felt or unfelt?); “attention” (ditto)
S.MIRSKY: “Having a sense of anything at all as when we experience sensory phenomena or ideas.”
Demustelation: “sense” (felt or unfelt?); “experience” (felt or unfelt?); “sensory phenomena” (ditto); “ideas” (ditto).
S.MIRSKY: “there are better terms yet, such as “experience”
Demustelation: “experience” (vide supra)
S.MIRSKY: “Whenever we go through any series of events of which we are aware (as in paying at least some level of attention to) we can be said to be having experience.”
Demustelation: “aware” (ditto); “attention” (ditto); “experience” (ditto)
S.MIRSKY: “The elements of our mental lives characterized by daydreams, streams of consciousness, etc."
Demustelation: “mental” (felt or unfelt?); daydreams (felt or unfelt?); “consciousness streams: (felt or unfelt?)
S.MIRSKY: “The phenomenon of being a subject, which seems to contain all other phenomena we encounter, whether objectively observable physical objects or events, or the mental imagery and emotions that characterize our interior lives including imaginary and remembered imagery.”
Demustelation: “subject” (feeling or unfeeling?); “phenomena” (felt or unfelt?); “observable” (felt or unfelt?); “events” (felt or unfelt?); imagery (felt or unfelt?); emotions (felt or unfelt?); “interior lives” (felt or unfelt?), etc. All equivocal. All just functing if unfelt; and, if felt, all of a muchness, and all demanding to know:
how felt rather than just functed?
why felt rather than just functed?
S.MIRSKY: “‘feeling’ is no better”
No better for what? It is just singling out that which needs to be explained.
S.MIRSKY: “Stevan himself erred… when he mixed “feeling” as in migraine headaches (a sensory phenomenon) with “feeling” as in what Searle says we must have if we are to be said to, in fact, understand the Chinese ideograms fed to us in the Chinese Room.”
No error: They are both
felt states, and *that* is the fact that needs explaining — not that
they feel like THIS or like THAT.