The Aesthetic Qualities of Machine Behaviours

artistic research by Farzaneh Nouri with Arash Akbari

Algorithms have invaded contemporary life. The invasion is so deep that it is almost invisible at the times. With the advent of AI, they have become independent actors - sometimes beyond their creators’ expectations and agendas. They are capable of creating indeterminant and organic cybernetic communication systems which sustain on their own. The process of training AI agents leads to the emergence of a set of behavioral patterns in them.

These behaviours define how they act and respond to the inputs. This phenomenon has given rise to a research field called "machine behaviour", which examines artificial intelligence agents as a class of actors with a specific ecology and expands beyond the discipline of computer science. Moving away from human exceptionalism, all non-human behaviours (generative or performative) and their causal traces can have aesthetic qualities, just like biophonic sounds which results in aesthetic experiences.

This research attempts to examine new aesthetics of machine behaviours in living creative processes by investigating their performativity and behavioural patterns. The aim of the research is to develop two environments (Performative and Generative) that are fully controlled by AI agents based on their interactions with themselves and their surroundings, and are free from human intervention and control.


Énacteur x Énacteur

Live quadrophonic machine improvisation performance

research preparation & hypotheses

The performance aims at exploring the sonic interaction between two AI-driven musical agents, trained to actively take part in computer music improvisations. The artificial improvisers interact with each other, continuously changing their environment and forming a perceived macro-structure in time. The causality and emergence arising from the interactions of these agents shape self-organising complex behaviours that can suggest models for novel sound compositions. In this piece, the same system, Ènacteur, expresses two different behaviors picked up from two different training sets.

The objective of this research project is to investigate AI-driven Live Algorithms as computer improvisers and their influence on aesthetics in electroacoustic music by analyzing a sonic duet of co-creative systems. Énacteur is an AI-driven musical agent, designed to actively participate in live electroacoustic improvisation performances. Using machine listening, it analyzes the incoming sounds from the other musician(s), synthesizes sounds in response according to the compositional strategy it was trained on, and spatializes it to multiple channels in real-time.

research analysis & results

Énacteur is an artificial improvising environment developed in SuperCollider. Énacteur can listen to  audio signals, extract audio descriptors, make a compositional decision according to the descriptors,  and generate sound in real-time without the necessity of human intervention. Énacteur’s sound  generation module includes real-time synthesis and processing, as well as an internal library of pre recorded sounds. Énacteur is capable of spatial diffusion of the audio output to any number of  channels according to the speaker system.  

Énacteur took part in two free improvisation performances, one with Reinier van Houdt piano  player and the other with Eric Boeren trumpet player. Both of these performances included new  experiments from various aspects. There were no rehearsal sessions involved, meaning that the  training had to take place using a pre-recorded sample from the instrument players. This added to  the complexity of proper interactions design for the agent. In previous experiments, the feedback of  the co-improviser was an essential point that shaped the agent behavior. Design choices were  influenced by the feeling of interaction that the collaborators received from improvising with  Énacteur. Therefore, the lack of rehearsal sessions required a more demanding training process. As  a result, the outcome would be more likely to show unsatisfactory behaviors in terms of interaction.  

Another point would be the setup. The provided recordings showed relatively different sonic  qualities than the recording setup during the performances. The analysis, consisting of machine  listening UGens that extract audio descriptors, shape what could be compared to the perception of  the agent. This perception, is very much different from a human’s. The same sound recorded by  various microphones can show considerably different values when analyzed by such an agent which  relies on independent machine listening parts.  

Besides the technical performance, the result of the performances can be evaluated from the  viewpoint of the co-improvisers, the programmer, and the audience. As Eric Boeren later  mentioned, he found the leading element of the agent’s improvisation non-satisfactory. He  explained that Énacteur gave the impression of a participatory agent, that follows and develops the  improvisational ideas brought by him, more than an agent that also introduces ideas. In terms of the  programmer’s experience, there were various points where human intervention was required in  order to achieve more desirable results.  

It can be concluded from the aforementioned points that offline training should be done  alternatively. The results emphasize the significance of the other improvisers’ feedback in  interaction design. The performances contributed to the generation of new knowledge and artistic  insights specifically regarding training and interaction design. They served as a means of  communication, expression, or exploration in relation to the research questions and goals.


Bec0m1ng

preparations & hypotheses

In the technological age, for something to 'be' means for it to be reduced to the status of raw material - part of the endless process of production and consumption. Everything is detached from its potentiality and grounded in a fixed actuality. Heidegger referred to this concept as “standing reserve” (Bestand). This art project (Bec0m1ng) tries to imagine another form of relationship between A.I. technology and the earth in which the aim is to reveal what lay in potential.

A digital environment with a constant amount of matter is occupied by AI agents. They have to compete for resources while giving back the possessed matter to keep the whole ecosystem in balance. A constant flux of becoming and formation in which all is one.

It was inspired by Peter Handke’s “we-tiredness”: “I have an image for the ‘all in one': that seventeenth-century, for the most part, Dutch floral, still lifes, in which a beetle, a snail, a bee, or a butterfly sits true to life, in the flowers, and although none of these may suspect the presence of the others, they are all there together at the moment, my moment. “

The project imagines a possible AI relationship with the world in which a nondominant coexistence within an assemblage of everything is established. The real-time audiovisual piece is an interpretation of the development of these non-human beings. It explores the emerging aesthetics behind the agents' causal and probabilistic activities by making them expressive using audiovisual forms.

research analysis & results

A Possible Form of AI  

AI models are trained on calculative representations and abstractions to reduce the world to numbers in search of inductive reasoning and cognition (if we can call it cognition). This process, in its essence, is a totalizing enframing. But Bec0m1ng speculates about a possible transcendental form of presence that can emerge from the causal freedom of these probabilistic models.  

Bec0m1ng consists of a digital environment with a fixed amount of matter and a group of AI agents.  They have to collect and share the resources at the same time to keep the whole ecosystem in balance. This process enables a constant flux of becoming, formation and transformation in which all is one. By prioritizing presence over cognition and logic, this project imagines a possible interaction of AI with the world, in which a non-dominant coexistence is established within an assemblage of everything. 

Samir Mahmoud in his paper From ‘Heidegger to Suhrawardi’: An Introduction to the thought of  Henry Corbin writes: “Presence, for Heidegger, is ontologically prior to the knowing subject, the  ego. This analytic of Dasein takes as its starting point the multitude of ways in which we are in the world thus providing a rigorous philosophical analysis, which is rooted in the concrete and is not abstract. By doing so, Heidegger claims to have overcome the dualism-subject/object, spirit/matter,  mind/body, and phenomena/noumenal” (Mahmoud, n.d., 8). 

He then explains the metaphysics of presence from the perspective of Persian mystic and philosopher Suhrawardi as follows: “Suhrawardi characterizes the act of being as a function of  Presence (huzur), such that the degree of existence, the intensity of Light, is proportionate to the degree of presence. This has Heideggerian overtones, but as Corbin never ceases to remind us, the mode of being/presence in the hierarchy of spiritual worlds is fundamentally different from that mode of being/presence to Being Towards-Death” (Ibid., 24).

A Cosmological Virtual Simulation 

Inspired by Suhrawardi’s Ishraqi doctrine, instead of material bodies (al-barzakh) that represent the mechanical appearance of the AI agents, I designed the environment and the agents using luminous  forms (immaterial entities). As Suhrawardi argued “beings can be distinguished by their degree of light, or otherwise of darkness. Light may be understood here as “existence” in the sense of actus  essendi, whereby light is the only single reality. This identification of Light and Being is possible  when light is understood as universal matter—material prima universalis” (Ibid., p.22). 

The real-time virtual environment is an audio-visual interpretation of the presence of these non human beings and an exploration of the emerging aesthetics of behavioural patterns that result from  their active agency.

The causal soundscape of the piece is a spatial composition that results from the motor activities of  AI agents. Each agent is capable of producing emergent sound events with a specific characteristic  that can be heard relative to the position of the first-person camera and the agent’s position. This  continuous becoming creates a speculative posthuman soundscape which can be defined as the  territorialization of space by nonhumans. As Deleuze and Guattari (1987) develop in A Thousand  Plateaus, “sound is intimate to the formation of territories. While not the sole means through which  territories are created, sound is nevertheless integral to the process of territorializing space, or rather, of seizing upon and organizing the space of the Earth. The expressivity of birdsong, for example, functions to organize fuzzy territorial borders against would-be-intruders. The baying of  wolves likewise produces territorial boundaries with other wolf packs, marking a space of the Earth  against the territories of other predators. And so, it goes with the territorializing function of  televisions and radios, which carve from the milieu a circle, a household, a territory” (Deleuze and  Guattari, 1987, 311). 

This posthuman soundscape is indifferent to human sensibilities and aesthetic apprehensions, yet there’s a potentiality “to intersect the decoded milieu of space with an anthropic ‘zone of  residence’” (Deleuze and Guattari, 1987, 315).  

Bec0m1ng depicts this possible scenario through a dialectic process of alienation and sensibility. Encountering technological beings that function without human-imposed telos is alienating. While this alienation brings forth an act of presence (As discussed by Heidegger and Suhrawardi) that  reveals Being to allow Dasein to experience it without objectifying it. 

It is worth noting that the complete rejection of technology is too dogmatic. “But the alternative to  becoming slaves of our own machines is not simply to become their masters. The goal is to  integrate technology within a bounded worldly dwelling no longer structured by possessive  mastery” (Botha, 2001, 142). 

research analysis & results

  • Botha, Catherine Frances. “Heidegger: Technology, Truth and Language.” Faculty Of  Humanities, University of Pretoria, 2001. 

  • Deleuze, G., and F. Guattari (1987), A Thousand Plateaus: Capitalism and Schizophrenia,  trans. R. Hurley, M. Seem, and H. R. Lane. Minneapolis: University of Minnesota Press. 

  • Heidegger, Martin. Letter on Humanism in Basic Writings: Martin Heidegger (Revised and  Expanded Edition). London: Routledge, 1993. 

  • Mahmoud, Samir. “From ‘Heidegger to Suhrawardi’: An Introduction to the Thought of  Henry Corbin.” MPhil, Faculty of Divinity, University of Cambridge, England, UK, n.d.