Applying Artificial Life Techniques to a Symbolic Learning Task Applying Artificial Life Techniques to a Symbolic Learning Task

According to the popular folklore of the two fields, the classical, symbolic paradigm of Artificial Intelligence research is incompatible with new ideas from Artificial Life, such as behaviour-based architecture, collective intelligence, and evolutionary computation. This paper seeks to debunk this myth through three experiments which employ techniques from both paradigms. These experiments all concern the acquisition of a declarative representation of the laws of physics of a simple microworld. The first uses a single robot which combines a behaviour-based architecture with the classical ID3 induction algorithm. The second uses a population of such robots. And the third uses a technique resembling a genetic algorithm to evolve a population of robots with appropriate declarative representations.