Evolution of game-theory behaviour

Stemming from my earlier experimentation with the evolution of form as defined by a genetically controlled embryology, a concept inspired by Richard Darkins' fantastic 'The blind watchmaker', and one naturally compatible with my Nabokovian tendencies, the evolution of behaviour within game theory (specifically the famous and tenacious 'prisoner's dilemma problem; having programmed my imaginary creatures to play this, I now know who to whom the 'prisoner' refers) was a natural extension. Again, it was Darkins' books that introduced me into this perrenial mathematical problem, and I thought I'd make my own inimitable contribution by attempting to see if Darkins' conclusions were borne out if the organisms had effective free evolutionary movement. To accomplish this, I genrated several generic modes or styles of play, which in combination allow the organism to perfect simple strategy for the Prisoner's dilemma game. The organisms start with a little random variation, and play each other to avoid the spectre of 'relegation', better known as 'death'. Those that survive produce progeny that carry their tendecies for style of play, so perpectuating their behaviour. In this way, I was able to demonstrate that the good guys really can come first, and the purely nasty gene gets supressed pretty quickly. Currently, this program produces only a graph of the abundance of each gene, and so isn't that spectacular, but can easily be converted by someone interested. It's of interest either for people working on similar game theory programs, or those constructing dynamic and generic AI systems. The code is all in Pascal 6.0. The program is downloadable with full source.

Download behaviour evolution () | Back to News