Towards an Architecture for A-life Agents



D.N.Davis & B. Berbank Green



Computational Intelligence for Modelling, Control and Automation.
Vienna, Austria, 1999


Abstract

This paper looks at the combination of artificial life and more orthodox AI techniques. We report on the application of hybrid agents, combining Cellular Automata and goal-oriented processing, to the game of GO. GO currently defeats orthodox AI techniques such as search for very obvious reasons. GO has a high branching factor with approximately 10768 sequences of moves in complex games; chess has a smaller state space with only 10120 possible board positions. Apart from providing a novel approach to an interesting application, this work feeds into a developing theory of agency and may be useful for future (more mainstream) agent applications such as data mining and loosely coupled intelligent decision support systems.