Tit for tat is the optimal solution for the Prisoner's Dilemma under normal conditions, meaning it would indeed be a waste to hold a programming competition for it. However, Tit for tat did once lose a competition to a "master/slave" combination.
One competition allowed multiple entries by the one team, so one group came up with the idea of having a "master" program and a horde of "slave" programs. Whenever either the master or slave started a match, they'd perform a more or less random pre-arranged "signature" of defections (eg defect-defect-coop-defect-coop or something like that), in order to identify each other. If the slaves detect that they are playing against the "master" program, they start always cooperating, while the "master" always defects, meaning it wins HUGE. On the other hand, if a slave detects it is playing against someone other than the "master" program, it starts always defecting, forcing the opposing program into taking a non-optimal path of either losing big time, or defecting also (still a fairly big loss compared to ideal).
So by having this mob of slave programs sabotaging all "normal" programs playing the game by always defecting, and massively boosting the one "master" program by always cooperating with it, the "master" program was able to achieve a far higher average score in the end than tit for tat. So while to be fair the programs there were gaming the system as much as they were the actual prisoner's dilemma, tit for tat is not always undefeatable. |
