Skip to content
TheChatArchiveThe Chat Archive
GPT 5.1 Codex Max· openaiHugging Face Frontier Reasoning Traces

Explain iterated prisoner's dilemma. Why does tit-for-tat outperform always defect in…

  1. User

    Explain iterated prisoner's dilemma. Why does tit-for-tat outperform always defect in infinite horizon? Explain the discount factor condition for cooperation to be Nash equilibrium.

  2. Assistant

    Detailing iterated prisoner's dilemma dynamics

    Explains that in infinitely repeated Prisoner's Dilemma, tit-for-tat can outperform always defect when the discount factor exceeds (T-R)/(T-P), enabling mutual cooperation as a Nash equilibrium sustained by future punishments. Highlights that tit-for-tat's reciprocity and punishment foster cooperation yielding higher average payoffs than always defect, especially under long-term interaction horizons. The stage game in a prisoner’s dilemma has the familiar ordering of pay‑offs [ T>R>P>S, ] where each player gets (R) for mutual co‑operation, (P) for mutual defection, (T) for “taking the temptation” and defecting against a co‑operator, and (S) for being the sucker and co‑operating against a defector. Played once, ((D,D)) is the unique Nash equilibrium.

    In an iterated or repeated prisoner’s dilemma the same stage game is played over and over, and players are allowed to condition their current move on the history so far. This opens up a much richer set of strategies than “always co‑operate” and “always defect”; “tit‑for‑tat” is the simplest of the contingent strategies: start by co‑operating, and thereafter do whatever your opponent did last time.

    The reason tit‑for‑tat beats “always defect” in an environment with ongoing interaction is twofold. First, against itself tit‑for‑tat achieves (R) in every round, whereas two “always defect” types only ever achieve (P<R). Second, because tit‑for‑tat punishes defection immediately, an “always defect” opponent only ever extracts the one‑off temptation payoff in the first round and then is stuck at (P) thereafter. When payoffs are averaged over a long or infinite horizon, a population of reciprocators earns a higher average payoff than a population of unconditional defectors.

    Formally, with an infinite horizon and discounting, a co‑operative outcome can be supported as an equilibrium if the “shadow of the future” is sufficiently long. Under a simple trigger strategy such as “co‑operate unless my opponent has defected in the past, then defect forever”, a player who contemplates deviating once compares [ \text{payoff from co‑operating forever} = \frac{R}{1-\delta} ] with [ \text{payoff from a one‑shot deviation} = T + \delta,\frac{P}{1-\delta}, ] where (\delta\in(0,1)) is the discount factor. Co‑operation is a best response if the former is at least as large as the latter, which reduces to [ \frac{R}{1-\delta} \geq T + \delta,\frac{P}{1-\delta} \quad\Longleftrightarrow\quad \delta \geq \frac{T-R}{T-P}. ]

    When (\delta) exceeds this threshold, the loss of future co‑operative payoffs outweighs the one‑time temptation to defect, and mutual co‑operation can be sustained as a Nash or even subgame‑perfect equilibrium. Tit‑for‑tat and other reciprocal strategies rely on exactly this logic: by making defection costly in terms of foregone future co‑operation, they make co‑operation self‑enforcing.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.