Taming nature: tapping the creative potential of ecosystem models in the arts

Abstract

This paper describes new research into the use of evolutionary ecosystem models as tools for creative artistic discovery. We give a brief history of the use of artificial evolution in the arts, and discuss difficulties with interactive and optimisation-based approaches to creativity. We present an approach based on evolutionary ecosystemic models. We propose that evolutionary ecosystemic models are capable of (i) driving the evolution of interesting novel interactive behaviour in a population of agents, (ii) exhibiting stimulating global dynamics which make them a suitable medium for artistic exploration and experience, and (iii) offering radically new ways to interact with complex dynamical software systems. We show three studies using evolutionary ecosystemic models in creative domains, demonstrating the potential for ecosystem dynamics to drive the emergence of novel behaviour. The benefits and challenges of these models are discussed, leading to a proposed framework for developing more sophisticated creative ecosystems.

Keywords: evolution, multi-agent models, computer music, computational creativity, ecosystem, niche construction

Links to artworks discussed in the article

• Study 1: Niche constructing drawing agents
• Study 2: Niche construction along a spectrum:
  Launch applet.
  

  Stills from the same run of the model at different stages.

  

• Study 3: a sonic ecosystem
  

  Example of the sonic ecosytem with agents playing sound files drawn from a database of piano and household percussion samples.

  Video example of the sonic ecosystem with agents playing granulated sound files (from Apple's Garage Band library). In this example the visualisation indicates aspects of the agent population. The height of the line represents their energy, which is determined by a simple resource model based on a relation between the sound made by the agent and the collective sound of the installation environment. The line goes red when the agent’s energy is sufficient for it to reproduce. Sometimes agents’ energies drop below zero, at which point they will die with a certain probability (also meaning that they might get lucky and make it back from the brink). The width of the line and also the dots at the bottom of the screen show the energy of the sound they are producing. The numbers represent the ID of the sound file each agent is playing, which is genetically determined. This means that you can spot evolutionary lineages as groups of agents with the same ID, or similar IDs. All of these agents will also make similar sounds, so a populous species will produce many layers of the same overlapping sound. The rotating discs represent the activations of the neural net used by each agent, which maps the features of the input sound (of the environment) to controller value that modify the sound produced by the agents (sample position, granular size, rate and randomness). The behaviour of the nets is also genetically determined, and means that agents can develop adaptive behaviour over time.

Other links

• Oliver Bown publications
• Jon McCormack publications

Errata

• page 219, last sentence of the first paragraph in the second column should read: "The assumption that only by imposing or capturing human aesthetic preferences can we evolve things of aesthetic value may simply be mistaken, or even counter productive."