Hacker News Reader

I remember that in older formal game theory tournaments, a punish-once single-retaliation strategy won out. It was unconditional on copying, simply that if the opponent cheated, you cheat back once and then forgive until another cheat. Another form of Golden Rule approach. But I think those tournaments were under simpler conditions than the one here.

I like the incorporation of miscommunication, and being able to change the parameters.

View on HN

Once weakness in those tournaments is that every interaction has the same stakes, so it doesn't model the possibility of "con artists" who behave generally at low stakes, but cheat at high stakes.

Varying the stakes would be interesting.

My understanding is this: in these computerized multi-player repeated games, each agent has no knowledge of previous games nor identities of other players. So every game a particular agent plays must follow the same algorithm. This means tournament results can easily be tabulated by tallying performance of each strategy type.

I remember that as well, but only because it's mentioned in book 8 of The Expanse.

Reply on news.ycombinator.com