If you would like to discuss strategies in a more conversational way, you can join this WhatsApp group. Comments are enabled on this doc but I’m also happy to open it up for direct editing to anyone that would like to contribute more. Just let me know!
Turn 1 strategies:
Which strategies should our 5 players play?
I propose that we start with a blend of strategies to try to learn as much as we can about the other strategies, while also being a little opportunistic about how much some teams have revealed about themselves in the public disclosures without putting all our eggs in that basket in case they are being deceptive about any of it. I’m also choosing not to specifically target players on any of the teams, and only using more generic strategies, even though targeting specific players might actually work pretty well at this point. Opting to keep it simple for now, even though we might regret it later.
Strategy first draft:
- Player 1: A strategy I wrote called, internally, Trickster, which attempt to identify if a strategy is friendly or unfriendly, and specifically exploits friendliness as much as it can without risking a more severe retaliatory response. Will share final code for this once I’ve worked out a few more bugs in it.
- Buster 6/29: Having 2nd thoughts about this… might be too tricky for it’s own good.
- Player 2: Tit for Two Tats: This one will help us identify exploitative strategies on the other teams without risking too many points by being aggressive itself. This strategy has historically “won” iterated prisoner dilemma tournaments, and is a safe bet to have on our team.
- Player 3: Gradual: Another safe bet that does well generally in friendly and semi-hostile environments. Cooperates until opponent defects. Then punishes with N defections where N is the total number of times opponent has defected. After punishment, plays two cooperations to signal willingness to cooperate.
- Player 4: A slight variant of Joss that I’ll call Noisy Tit for Tat where it’s always copies the other opponent but 10% of the time plays the opposite move that the opponent played. We can use this to identify strategies that have learning components to them.
- Player 5: Grim Trigger, to protect against the craftiest and most exploitative strategies. This one cooperates until the opponent defects, even once, and then defects for the rest of the match. It’s basically an insurance policy for our team if we’ve just been outsmarted by the other teams, and also a kind of pressure on them to never defect proactively.
Strategy second draft:
- Player 1: Trickster, as above
- Player 2: Tit for Tat
- Player 3: Gradual
- Player 4: Grim Trigger