Human-like inclinations: composing a multimedia live installation with AI

Galaz, Pablo, Melen, Christopher, Lee, Tiarna, Hui Wai Nok, Angela, Kulina, Natálie, Danesi, Marco and Gorini, Paolo (2024) Human-like inclinations: composing a multimedia live installation with AI.

Abstract

Human-like is a 30-minute interdisciplinary work for performer, violin, bass clarinet, Seaboard, live electronics, and live video. Situated at the intersection of spatialised concert, live installation, and experimental musical theatre, the piece explores metaphors associated to machine learning’s training process through musical composition and performance.

The work investigates the central question: What would it mean for a machine to understand what it is like to be the kind of creature we are? It constructs a figurative sonic and visual representation of a training process, where the imaginary “dataset” consists of features associated with human nature. The compositional sound world integrates acoustic instruments, heartbeats (derived from actual medical datasets) and electronic glitches that evoke the electrical apparatus of a computer (many of which were captured from actual computers using electromagnetic microphones in analogy to medical auscultation).

The piece features four musicians spatially distributed in a cross formation around the audience. The solo performer stands before a smartphone with face-tracking capabilities, triggering and manipulating samples through facial gestures, such as blinking and smiling. The piece unfolds through repetition and variation, reflecting epochs in a metaphorical training process. As the work progresses, new sounds and gestures emerge. The performer evolves from only using her eyes to incorporating mouth and arm movements, culminating in a moment where the machine appears to have achieved human-like presence.

The reliance on repetition reflects both conceptual ideas and practical constraints: musicians perform in darkness, spaced apart, with the performer's eyes closed for long durations, requiring extensive memorisation and creating synchronisation difficulties, which repetition helps to address. Staging decisions and musical structure are intricately connected.

Human-like contributes to discourse on human-machine interaction, interrogating the extent to which AI can approximate human behaviour. It critically engages with the current debates on AI’s capabilities, inviting reflection on whether machines can truly capture human essence, and, more crucially, on the role such technologies should play in our lives.

This record has no associated files available for download.

Contributors

Download statistics