The work
Multiple Realities is a 5-10 minutes audio-visual performance generated by an interactive multi-agent system. The system supports musical expression by dynamically generating patterns and progression through an emergence process. Architecturally, it consists of three components: a generative director, a visual generator, and an audio generator. The system can play autonomously from an initial configuration or used online, with a performing guiding and adjusting the system evolution. Unlike most multi-agent systems, the agents here are inspired by the dynamical hypothesis of cognition. Self-organization appears as the result of coupling of many simple dynamical systems by means of short-range interactions.
Inspiration |
This work was inspired by the cosmological theory of inflation. It posits the the space expanded exponentially after the Big Bang and was first developed by Alan Guth at Cornell University in 1979. It was later shown, that most inflation models lead to multiverses, in which the space itself is broken into bubble universes with their own physical properties. So far, experiments seem to confirm the predictions of the inflation theory, thus making it more likely that we live in a multiverse!
|
Approach
The dynamical hypothesis of cognition regards it as the result of the evolution of a system in the space-state. Interestingly, connectionist systems often mentioned in the context of intelligent agents, can also be viewed as a realization of a dynamical system. In this case, we view the trained network as one specific configuration of the “activation space” that leads to the problem solution. Self-organization can also appear when ensembles of dynamical systems are coupled through short range interactions. This configuration has been used to investigate emergence from primordial soups of large macromolecules and has been linked to the origin of inference capabilities.
The agents exist and move in 3D space. In every time step, the agents move close to another agent within the visibility radius. If they get too close, they slowly move apart until a predefined separation distance is reached. Agents also grow and pulsate following the internal evolution of the dynamical system. When an agent gets in contact with another, there is a 50% chance that a cross-over will happen, effectively switching the parameters of involved dynamical systems. This causes each agent to move to a different part of the state-space and potentially initiate other exchanges.
The evolution and state of the agents can be used to generate music and video.
The evolution and state of the agents can be used to generate music and video.
The system developed for the Multiple Realities performance consists of three components: a generative director, a visual generator, and an audio generator. The system can be configured in real-time or in the beginning of each performance (if desired) to yield unique experiences.
Dynamic parameters for the agent behavior are:
01 - 3D position
02 - Logistic constant
Fixed parameters are:
01 - Logistic constant
02 - Attraction force
03 - Repulsion force
04 - Time scale
05 - Initial 3D position
01 - 3D position
02 - Logistic constant
Fixed parameters are:
01 - Logistic constant
02 - Attraction force
03 - Repulsion force
04 - Time scale
05 - Initial 3D position
Audio |
The output from the Director is fed directly to the audio synthesis software SuperCollider using an Open Sound Control interface. The population value is used to choose a tone from the major scale. The variable parameters are:
01 - Degree in the scale 02 -Duration 03 - Amplitude The fixed parameters (set for a specific work) are: 01 - Channel 02 - Degree offset Multiple Realities use three agents. A script generates tones resembling a string synthesizer with chorus effects. Agents operating below the first bifurcation (r»3.0) are muted to prevent distracting continuous sounds. |
Visuals |
The outputs from the Director are used to control three groups elements, each containing a few hundred copies of a geometric primitive. The groups are controlled by adjusting the distance of each element from the original position, scale of each element, and the distance among elements of the same group.
Dynamic parameters are: 01 - Distribution perturbation 02 - Distribution scale 03 - Object scale Random parameters are: 01 - Color |
Performance Photos
Natural History Museum, Verona - December 2018
|
|