Creatively Evolving Cooperative Behaviour with the 'NeuroEvolution of Augmenting Topologies' Algorithm

Abstract

This thesis combines creativity and cooperative behaviour, and aims to investigate if a machine learning algorithm called NeuroEvolution of Augmenting Topologies (NEAT) can evolve cooperative behaviour in a creative process. Such a process may be used to generate cooperative behaviours in creative systems, such as video games or other kinds of simulation software.

This research was mainly a design, implementation and experiment research. Inspired by previous work performed in the research field of computational creativity, a definition and evaluation criteria for creativity and cooperative behaviours were formulated. A system was designed and implemented to simulate and evolve the behaviour of multiple interacting agents. Experiments were conducted using this system. The cooperation of the generated artefacts and the creativity of the system were evaluated.

Experiments were run on four different environments. In two of the environments no cooperation was found. In one environment cooperation was found, but the results were inconclusive whether the behaviour emerged through a creative process. In the last environment the algorithm evolved behaviour that satisfied our definition of cooperation and which was evolved through a process that satisfied our criteria for creativity.