Transparent tit-for-tat is a decent enough strategy, but it is easily beaten 1-on-1 by “cooperate on every turn except the very last one”, and against many other simple bots by “TFT but defect on the last turn”.
I agree, I was simply giving an example of a strategy which would not lead to the sort of infinite regress envisioned by ThisSpaceAvailable.
I do think it’s worth noting that “TFT but defect on the last turn” would be beaten 1-on-1 by “TFT but defect on the last two turns.” And both of them would be beaten 1-on-1 by “TFT but defect on the last 50 turns.”
In fact, in a 1-on-1 game, it seems to me the best strategy would seem to be to always defect and hope the other fellow is stupid enough to cooperate at least once. Actually it seems pretty much pointless to talk about 1-on-1 prisoner’s dilemma, because it turns things back into a zero sum game.
If you think about it as win/lose, then yes, of course. However, it’s still instructive to ask the question “what’s the best response strategy”? In other words, “what strategy will maximise my utility against this given, fixed opponent strategy?”.
In that sense, the best response to TFT is indeed “TFT but defect on the last turn”, because “TFT but defect on the last two turns” does strictly worse when playing against TFT.
Fortunately, you can indeed do significantly better than TFT in this game!
In that sense, the best response to TFT is indeed “TFT but defect on the last turn”, because “TFT but defect on the last two turns” does strictly worse when playing against TFT.
But what if you are playing against “TFT but defect on the last turn”? In that case, the best strategy is TFT-2. And if the other fellow is playing TFT-2, the best strategy is TFT-3. And so on. Agreed?
Fortunately, you can indeed do significantly better than TFT in this game!
Seems to me it depends on what strategies the other contestants are playing.
Oh, yes, of course; the best response to TFT-1 is clearly TFT-2, and so on.
As for how well strategies do, while it’s clear that it will depend on the strategies of other contestants and in that sense there cannot be a “best strategy”, I think one can do better—for example, if there’s a Nash Equilibrium strategy that isn’t simply (Defect, Defect).
At a bare minimum, you can improve upon TFT by also making sure it defects against CooperateBots, and doesn’t wait until the second turn to defect against DefectBots. Of course, there may indeed be JusticeBots out there who punish you for defecting against CooperateBots...
At a bare minimum, you can improve upon TFT by also making sure it defects against CooperateBots,
That’s assuming that the new algorithm can correctly identify its opponents. Also, if other algorithms correctly sense that yours is opportunistic, they might change their strategy to your detriment.
I agree, I was simply giving an example of a strategy which would not lead to the sort of infinite regress envisioned by ThisSpaceAvailable.
I do think it’s worth noting that “TFT but defect on the last turn” would be beaten 1-on-1 by “TFT but defect on the last two turns.” And both of them would be beaten 1-on-1 by “TFT but defect on the last 50 turns.”
In fact, in a 1-on-1 game, it seems to me the best strategy would seem to be to always defect and hope the other fellow is stupid enough to cooperate at least once. Actually it seems pretty much pointless to talk about 1-on-1 prisoner’s dilemma, because it turns things back into a zero sum game.
If you think about it as win/lose, then yes, of course. However, it’s still instructive to ask the question “what’s the best response strategy”? In other words, “what strategy will maximise my utility against this given, fixed opponent strategy?”.
In that sense, the best response to TFT is indeed “TFT but defect on the last turn”, because “TFT but defect on the last two turns” does strictly worse when playing against TFT.
Fortunately, you can indeed do significantly better than TFT in this game!
But what if you are playing against “TFT but defect on the last turn”? In that case, the best strategy is TFT-2. And if the other fellow is playing TFT-2, the best strategy is TFT-3. And so on. Agreed?
Seems to me it depends on what strategies the other contestants are playing.
Oh, yes, of course; the best response to TFT-1 is clearly TFT-2, and so on.
As for how well strategies do, while it’s clear that it will depend on the strategies of other contestants and in that sense there cannot be a “best strategy”, I think one can do better—for example, if there’s a Nash Equilibrium strategy that isn’t simply (Defect, Defect).
At a bare minimum, you can improve upon TFT by also making sure it defects against CooperateBots, and doesn’t wait until the second turn to defect against DefectBots. Of course, there may indeed be JusticeBots out there who punish you for defecting against CooperateBots...
That’s assuming that the new algorithm can correctly identify its opponents. Also, if other algorithms correctly sense that yours is opportunistic, they might change their strategy to your detriment.