We submitted two kinds of strategies to the iterated prisoner's dilemma (IPD)competitions organized by Graham Kendall, Paul Darwen and Xin Yao in 2004 and2005. Our strategies performed exceedingly well in both years. One type is anintelligent and optimistic enhanced version of the well known TitForTatstrategy which we named OmegaTitForTat. It recognizes common behaviour patternsand detects and recovers from repairable mutual defect deadlock situations,otherwise behaving much like TitForTat. The second type consists of a set ofstrategies working together as a team. These group strategies have onedistinguished individual Godfather strategy that plays OmegaTitForTat againstnon-members while heavily profiting from the behaviour of the other members ofhis group, the Hitman. The Hitman willingly let themselves being abused bytheir Godfather while themselves lowering the scores of all other players asmuch as possible, thus further maximizing the performance of their Godfather inrelation to other participants. The study of collusion in the simplifiedframework of the iterated prisoner's dilemma allows us to draw parallels tomany common aspects of reality both in Nature as well as Human Society, andtherefore further extends the scope of the iterated prisoner's dilemma as ametaphor for the study of cooperative behaviour in a new and natural direction.We further provide evidence that it will be unavoidable that such groupstrategies will dominate all future iterated prisoner's dilemma competitions asthey can be stealthy camouflaged as non-group strategies with arbitrarysubtlety. Moreover, we show that the general problem of recognizing stealthcolluding strategies is undecidable in the theoretical sense.
展开▼
