News for hackers. Stuff that matters.
Re:Show HN: Prisoner's dilemma tournament with bots that can... by philh on Thursday July 31, 2014 @12:50PM (#8115225)
> The tit-for-tat is very effective in the classical iterated prisoner's dilemma, but it is easily exploitable in the program equilibrium verion of the game. This is because the bot's decision is based entirely on the history of previous moves, so an opponent can always predict when the tit-for-tat bot will cooperate and then defect against it.

Unless I'm missing something, doing this naively doesn't let you exploit TitForTat, it just decreases your reward from (3 every round) to (5 every other round).

You can still exploit TitForTat, e.g. by defecting in the final round. But that's not a massive win for you. CooperateBot is much more exploitable.

Re:Show HN: Prisoner's dilemma tournament with bots that can... by spimta on Thursday July 31, 2014 @12:54PM (#8115267)
Yep, you're exactly right--I didn't actually add up the payoffs when I wrote that. You could potentially do better against tit-for-tat variants that have a certain chance of forgiving after a defection, but not against the naive, deterministic TitForTat.