[Top]

[Back]

[Next]

[Bottom]

3 Agents

"agent (a'jent) n. a person who acts on behalf of another person, business or government, etc. [C15th: From Latin agent-, noun use of the present participle of agere to do]" - The Collins English Dictionary

3.1 What is an Agent?

• Upon logging into your computer system, you are greeted by your personal digital assistant (PDA), who presents you with a summation of activities that have occurred since you last logged in. The PDA has sorted your email messages into order of importance and has automatically cross-referenced them to find any related email messages. One email in particular is from your supervisor requesting a meeting to discuss your work; the assistant has checked your calendar and already negotiated a suitable time and date with the PDA of your supervisor. In addition to this, your PDA has also searched the network news groups and presents you with a list of articles that it knows you are interested in; the assistant draws your attention to a thread which describes work that is very closely related to your own. In anticipation that this will be of interest to you, the PDA has already obtained a technical report from an FTP site and compiled a list of other references which are available and detail this work further.
• You are editing a file when your PDA requests your attention; an email message has arrived that contains notification of acceptance of a paper that you submitted to a conference. Without prompting, your PDA has looked into travel arrangements and has determined all of the planning requirements, from taxis to aeroplanes. In a short while, you are presented with a summary of the most convenient travel options. Once you have confirmed a suitable option, your PDA makes reservations on your behalf, marks your calendar accordingly and negotiates with the conference site about locating your data locally for the duration of the conference. While you are in transit to the conference, the assistant transfers your data (including the files necessary to your presentation) to the conference site, redirects your information sources (email, for example) and informs the PDA of the conference supervisor of your impending arrival.

At an elementary and conceptual level, agents can be regarded as entities that perform actions on a person's behalf. For example, we visit a travel agent and book our holidays; the travel agent is given the task of organising the flight details, the hotel bookings and the insurance arrangements on our (the holiday maker's) behalf. We do this because it would be tiresome to have to organise all of the details ourselves; we delegate responsibility to an agency we know can perform the task.

An analogous description in the software world could be that an agent is `some software that performs tasks on a user's behalf'. However, in blithely making this parallel and believing that the analogy translates correctly, a number of fundamental assumptions are being taken for granted. In our real-world scenario, we trust a travel agent to book the flight and we believe that they will book the hotel for the correct dates. In other words, we have delegated not only our trust in the ability of the travel agent to complete the task that we have conferred to them, but we also believe that they will execute it correctly.

Therefore, the usefulness of an agent is directly related to the amount of trust and believability it generates and this is possibly more important than the actual task it can perform; agents that fulfil their tasks badly (either incorrectly or incompletely) are of little use to us, since we would employ the services of another agent or simply complete the task ourselves, rather than use it again.

These concepts of trust and believability need to be identified and embodied in software agents (hereafter referred to simply as an agent), before their tasks can be accorded the same amount of trust that is granted to real-life agents.

3.2 Notions of Agency

3.2.1 A Weak Notion for Describing Agents

• Autonomy. Once launched with the information describing the bounds and limitations of their tasks, an agent should be able to operate independently from their user, that is, autonomously in the background (Castelfranchi, 1995). To this end, an agent needs to have control over its actions so that it can determine what to do when an action succeeds or fails. Moreover, an agent must be able to augment its internal state so that it can make rational decisions based upon the information that it has gathered.
• Social ability. To effect changes or interrogate their environment, an agent must possess the ability to communicate with the outside world (Genesereth et al., 1994; Mayfield et al., 1995). This interaction can exist at a number of levels depending upon the remit of the agent, but typically an agent would need to communicate with other agents and the local environment (to maintain/discover information) and users (to appraise them of their progress).
• Reactivity. Agents need to be able to perceive their environment and respond to changes to it in a timely fashion, depending upon their remit. For example, an agent's task could be to monitor a local file system, informing the user when changes occur to a particular file set. This implies that the agent has an awareness of the appropriate filing system and how to interrogate it; agents need not only to be aware of their environment, but they need to be aware of what the state and changes in that environment mean and how to react to them.
• Pro-activeness. To help differentiate an agent from another piece of software, agents need to be able to exhibit pro-activeness, that is, the ability to effect actions to achieve their goals by taking the initiative. This means that an agent needs to appreciate the state of their environment and to decide how best to fulfil their mission target.

3.2.2 A Stronger Notion for Describing Agents

An intentional system is a system that can best be described by the intentional stance; the ascription of abstraction notions to systems for the purpose of describing how they work. For example, although the technical description of a computer system may be available, it is too complex to use when describing, say, why a menu appears when a mouse button is clicked over a certain area of the display. The intentional notions as described by Shoham are useful for providing convenient and familiar ways of describing, explaining and predicting the behaviour of complex systems: Dennett suggests that strong agents are best described by the intentional stance.

Bates uses this the concept of strong agents and takes it into more anthropomorphic areas by considering the implications of believable agents, that is, agents that try to model a human approach to their interaction with the user by displaying emotions (Bates, 1994). Additionally, Maes talks about representing agents visually by attaching an icon or a face to associate them with cartoon or computer characters (Maes, 1994). These types of agents are being used in both Human Computer Interaction (HCI) scenarios to help the social interaction between a user and their agents, and also in the computer gaming community to produce virtual characters that react in believable and human ways to given situations.

3.2.3 Intelligence and Learning

However, as Brooks stated (Brooks, 1990; Brooks, 1991; Brooks, 1991b), intelligence is `in the eye of the beholder', since the question `what is intelligence?' is as difficult to define as `what is an agent?'! He goes on to make two key points about intelligence:

• Real intelligence is situated in the world, not in theorem provers and expert systems, and,
• Intelligent behaviour arises as a result of an agent's interaction with its environment.

Edwards (Edwards, 1995) argues that it is not enough for agents to react intelligently to their environment, but they must also be able to adapt and alter to changes by learning. An agent that can learn through exposure to given situations and examples could be more useful to a user than an agent whose intelligence is fixed. However, it is far more difficult to predict the behaviour of an agent that can learn, since it is not possible to determine exactly what it will learn and how it will apply that information.

3.2.4 Other Attributes of Agents

• Mobility. The ability for an agent to move across networks to fulfil its goals (Gray, 1995b). Mobility will be explored in greater depth in the next chapter, but it is enough to say here that it introduces some interesting concepts of persistence and `lifetime'.
• Rationality. The assumption that an agent will not act in a manner that prevents it from achieving its goals and will always attempt to fulfil those goals (Galliers, 1988).
• Veracity. The concept that an agent will not knowingly communicate false information (Galliers, 1988).
• Benevolence. An agent cannot have conflicting goals that either force it to transmit false information or to effect actions that cause its goals to be unfulfilled or impeded (Rosenschien et al., 1985).

3.2.5 Summary

Intelligence and learning appear to offer methods for building trust and believability into agents; intelligence allows agents to fulfil tasks competently and learning allows agent to react to situations in an intelligent manner. To ensure that an intelligent and learning agent does not grow outside of its remit, other agent attributes, such as benevolence, veracity and rationality may need to be incorporated.

3.3 The Differing Views on Agents

3.3.1 The Traditional Agent

The ultimate goal of AI agents is to provide intelligence and reasoning capabilities that are comparable to that of human beings. As McCarthy (McCarthy, 1978) puts it:

"[Artificial Intelligence is] the science of making computers do things which if done by humans would require intelligence."

Traditional AI architectures are generally based around three core philosophies:

• Symbolic. The reduction of the world to a representation of realisable symbols that can be combined to form structures upon which processes can be executed to operate upon the symbols according to symbolically coded sets of instructions (Newell et al., 1976). Decisions regarding actions to perform are made through logical reasoning, based upon pattern matching and symbolic manipulation. Example systems include Homer (Vere et al., 1990), a robot submarine which explores a two-dimensional Seaworld, can respond to an 800-word spoken vocabulary and has a limited episodic memory, and Grate* (Jennings, 1992), a simulation of electricity transportation management where agents act in an individual, selfish and cooperative manner to achieve their intentions.
• Reactive. A major problem with symbolic AI is the processing power required to analyse the information about the real world, plan a suitable solution and then implement a chosen action. Critics of symbolic AI have advocated the use of reactive architectures; architectures where there is no complex representation of the real world in symbolic terms and where no symbolic reasoning is performed. The most vociferous critic, Rodney Brooks, has developed an approach based around a reactive model, called the subsumption architecture (Brooks, 1986). A subsumption architecture is a hierarchy of behaviours designed to accomplish tasks. Each behaviour competes with other behaviours in order to influence the actions of the agent; lower layers in the hierarchy represent primitive behavioural styles (for example, avoiding obstacles in the Homer Seaworld) and higher layers represent more abstract behaviours (for example, collecting an object and returning it to a given location). The amount of computation required for these systems is very small when compared with symbolic AI systems. Another reactive architecture, based along the same lines as the subsumption architecture, is Pengi (Agre et al., 1987), a computer simulated game where routine tasks are encoded in low-level structures and are only updated to handle new problems that develop. The idea is that most decisions are routine and can therefore be performed quickly and efficiently.
• Hybrid. These types of architectures attempt to marry the best qualities of both symbolic and reactive approaches to AI. The method consists of building an agent system out of two (or more) subsystems; a symbolic world model which develops plans and makes decisions, and a reactive subsystem which is capable of reacting quickly to events that happen without having to resort to complex symbolic manipulation. Examples of such hybrid systems are Touring Machines (Ferguson, 1992) and InteRRaP (Müller, 1995). In these architectures a layered model is employed; a symbolic AI engine sits at the top of the model handling long-term goals and a reactive AI engine sits at the bottom handling low-level reactions. The problem with such systems is how to manage the interactions between the layers.

"...a personal assistant who is collaborating with the user in the same work environment."

To support this line of reasoning, the AI laboratory within MIT have developed a prototype interface agent called News Tailor, or NewT (Sheth, 1994; Maes, 1994). A NewT agent is a USENET news filter that can be `trained' by giving it a series of examples that show in which kind of articles the user is interested. From this, the NewT agent can search all news articles to try and find other articles which are similar to the one's initially indicated by the user. When the agent presents the other articles that it has found, the user gives feedback according to their applicability; thus, the NewT agent can widen or restrict its searching next time.

Other interface agent systems include NewsWeeder (Lang, 1995), UNA and LAW (Green et al., 1995), WebWatcher (Armstrong et al., 1995) and LIRA (Balabanovic et al., 1995).

3.3.3 The Information Agent

The key qualities of information agents lie in their ability to communicate with a wide range of information resources to ensure that the widest amount of information is processed to provide the user with the best results to their original request.

Theoretical studies on how agents can utilise the information that they receive from different resources are presented by Levy (Levy et al., 1994) and Gruber (Gruber, 1991). A more practical application has been presented by Voorhees (Voorhees, 1994), who describes a prototype system called the Information Retrieval Agent (IRA) which can search for loosely specified articles from differing document repositories.

3.3.4 The Distributed Agent

DAI agent technology is being employed in many real-world situations, for example, air traffic control (Steeb et al., 1988), particle accelerator control (Jennings et al., 1993) and telecommunications network management (Weihmayer et al., 1994). However, a key problem with DAI is ensuring that problem decomposition and subsequent communication and discussion between communities of agents can take place timely enough to produce useful and achievable results.

3.4 Summary

Most technologies that use agents realise that the more attributes that an agent possesses, the more complex the task becomes of specifying, designing and implementing that agent. This helps to explain why there has been a general trend over the past 10 years away from AI dreams (the HAL computer, for example, from 2001) to more realistic areas of actual applicability.

Agents that are useful to the user in everyday activities seem to be the way that agent technology as a whole is moving. It is hoped that by starting in the small with relatively easily specified agents which have limited capabilities and limited intelligence/learning, then the experience gained will begin to show the way of progressing up to computing with agents in the large.

General progress has been made over recent years, especially in the area of information agents, and this appears to be where future research is heading. The following chapter discusses a particular aspect of agent technology, mobility, and describes how applicable current mobile agent systems are for distributed information management.

[Top]

[Back]

[Next]

[Bottom]