A reactively learning neural network that decides behaviours for an artificial life system with homogeneous agents
Samuel Jonathan Slade · 2016
Artificial life has been a subject of much debate over the years, the subject tackles defining life and simulating aspects of it based on those definitions. This work hopes to explore the implementation of self-organising systems, evolution and machine learning approaches for reactive decision making. The project implements agents with predetermined behaviours which are guided by a simple blackboard-like entity that chooses the main tasks for the agent cluster via a neural network. The results show that a reactive learning neural network is marginally better at guiding the agent cluster during points of stagnation than a random selection of decisions at a point of stagnation. The paper highlights issues with the neural network approach and shows that stagnation detection has a reasonable impact on self-organization and autopoiesis of the system.