The 4th Humour
uninfluential words from an uninfluenced man
Bile humour Apathetic hemetic Fluent indifferent Emetic Phlegmatic

This page is powered by Blogger. Isn't yours?
Monday, November 15, 2004
Tit-for-Tat: The Dilemma

(very old draft, revived)

Anyone familiar with Axelrod's iterated Prisoner's Dilemma competitions also knows that the winning strategy has always been Tit-for-Tat. The Tit-for-Tat strategy, upon its first encounter with a competitor, cooperates. Thereafter it does whatever its competitor does; if its competitor defects, it defects; if its competitor cooperates again, it again cooperates.

I'll admit this seems like a good strategy because pride is not a factor. Good behavior is rewarded, bad behavior is punished. However, outside of a computational simulation--in real life--there is a problem if this simple tactic is followed: miscommunication.

What if some misunderstanding arose in the relationship that caused one agent to defect when it should have cooperated? A simple mistake of judgement. Two otherwise cooperative people could then be forever caught defecting against one another. Tragedy. I think this happens often, and the agents will often continue to defect, not as a matter of mechanism, but as a matter of "principle" or pride (I'm sure there's a name for this logical fallacy), even when later evidence suggests there was an initial misunderstanding. After all, the initial misunderstanding is likely the fault of one agent who does not want to admit its mistake and lose face.

Every once in a while, one of the agents will need to take a chance and start to trust again, to cooperate, measuring its potential future (mutual) gain against that of a potential temporary (amutual) loss. The game, then, isn't when to defect, as in a normal iteration, but when to again cooperate.