Zaratustra above mentioned the Prisoner's Dilemma computer contests held -- after the rounds, the highest score wins. But in game theory, this doesn't work out too well. The reason for this is that the principles of game theory are based on your payoffs being yours alone. You shouldn't even validly be able to compare yours to someone else's.

Exploiting this failure, Tit for Tat wins overall on a large scale. Why? Two reasons. One is because it never lets the other guy get _ahead_ by more than k (which is my variable for "the difference between the payoffs of the two prisoners when one confesses and the other defects"). The other reason is that it acquires its enormous point advantage from programs similar to it which are likely to cooperate.

In any case, all this is invalid when you realize that, theoretically, your only goal is to get the highest possible score, not to "beat" your opponent. You can have a matrix like this and it is still a "Prisoners' Dilemma:"

Non-comparative PD matrix
(Payoffs listed in Row, Column form)

           C         D
Coop     (3, 999)  (1, 9999)
Defect   (4, 9)    (2, 99)

(If you haven't seen payoffs before, this is called a payoff matrix. One player chooses one of the rows, the other player chooses a column, and you intersect them. The row player gets "paid" the first ordinate in the pair, and the column gets paid the second.)

Anyway, in this game, one player gets paid a lot more than the other, but that should mean nothing to either player. They're playing for abstract goodness, not comparative goodness. Even if you use a more "standard" PD matrix, you shouldn't compare them. I'll use ordinal payoffs to represent that: 1 is the worst outcome and 4 is the best, no matter what their actual value to you is.

Standard PD game matrix with ordinal payoffs

           C         D
Coop     (3, 3)    (1, 4)
Defect   (4, 1)    (2, 2)

This is a human nature thing -- we like to compete and we judge ourselves by what others do. One way to think of it is that the payoff is not an amount of money--but an "abstract happiness quotient". How happy are you with this outcome? Once "beating the other guy" is part of the "happiness quotient," the game is changed fundamentally.

Unfortunately, nobody has found a good solution to this. I don't believe there IS one, so we'll have to make do with what we have. It IS interesting, but doesn't fit with the original terms of the contract, so to speak.

See also chicken, w/u by caseyhb.