Tit For Tat Vs Pavlov

The original IteratedPrisonersDilemma contest was held by RobertAxelrod?, and was discussed in his TheEvolutionOfCooperation (ISBN 0465021212 ). Very briefly, the finding was that in many environments, cooperation will emerge.

The stunning winner of the first competition was TitForTat. You can find its strategy on the PrisonersDilemma page. It begins by cooperating, then copies its opponent's action in the previous turn. If the opponent always cooperates, so does TitForTat, and both players make the optimal move. If the opponent ever defects, TFT will defect on its next turn; if the opponent ever switches back to cooperation, TFT also switches back on its next turn. It is a very simple strategy, "An eye for an eye, a tooth for a tooth." TFT's main advantage in this competition is that it assumed cooperation but did not put up with any crap if the opponent did not cooperate. It was never more than a small number of points below each of its opponents scores, but whenever it played itself it always got the highest score possible.

Later, another competition was held in which there was a small probability for communication error (PleaseComment: I don't have the source). Occasionally, the player would see a defection when the opponent actually cooperated, or vice versa. In this case, another strategy won the day, PavlovStrategy. Like TitForTat, it too is very simple: if your opponent cooperated with you on the last round, repeat your last move; otherwise, switch. PAVLOV's biggest advantage was that, like TFT, it assumed cooperation. However, if cooperation failed, like TFT, it would automatically punish its opponent. But then, unlike TFT, once both sides were defecting, it would try to switch again to cooperation. If both players were PAVLOV, this behaviour acted like error correction, and both players would resume cooperation. TFT on the other hand, when playing against TFT, would slip into mutual defection.

One last result before I get into its relevance today: In a further study, where mutation was introduced into a population of strategies, it was found that there is no one best strategy. The best strategy always depends on what other strategies are present. Some strategies will be the 'dominant' strategies, while certain others will guard against uprising 'invading' strategies. (PleaseComment: I don't have the source for this one either).

Axelrod writes in TheEvolutionOfCooperation that no strategy is universally best, even when communication is perfect. Fitness always depends on the mix of strategies present. While TitForTat's performance is usually excellent, it does not always spread throughout a population.

Anyway, I know it's simplistic, but I see the terrorism of today as representing a TFT strategy that has been ingrained in a TFT culture. The motto, "An eye for an eye, a tooth for a tooth," says it all. When there is the possibility for miscommunication, as there very much is in this world, it results in a holy war with no end. It's going to take more than a US counterstrike to stop terrorism. It will take a shift in the predominant cooperation strategies of the world. Something like a Gandhi.

Sounds like PAVLOV turns the other cheek...

View edit of January 31, 2003 or FindPage with title or text search