A Predictive Strategy for the Iterated Prisoner's Dilemma

TL;DR

This paper introduces PREDICTOR, a simple yet effective strategy for the iterated prisoner's dilemma that learns to predict opponents' moves and adapt its cooperation or defection accordingly, achieving high scores and winning tournaments.

Contribution

The paper presents PREDICTOR, a novel strategy that models opponents and predicts future actions without tags or complex mechanisms, demonstrating strong performance in simulations.

Findings

PREDICTOR achieves high average scores in tournaments.

It wins various parameter setting tournaments.

The strategy evolves morality from selfish behavior.

Abstract

The iterated prisoner's dilemma is a game that produces many counter-intuitive and complex behaviors in a social environment, based on very simple basic rules. It illustrates that cooperation can be a good thing even in a competitive world, that individual fitness needs not to be the most important criteria of success, and that some strategies are very strong in a direct confrontation but could still perform poorly on average or are evolutionarily unstable. In this contribution, we present a strategy -- PREDICTOR -- which appears to be "sentient" and chooses to cooperate when playing against some strategies, but defects when playing against others, without the need to record "tags" for its opponents or an involved decision-making mechanism. To be able to operate in the highly-contextual environment, as modeled by the iterated prisoner's dilemma, PREDICTOR learns from its experience to choose optimal actions by modeling its opponent and predicting a (fictive) future. It is shown that PREDICTOR is an efficient strategy for playing the iterated prisoner's dilemma and is simple to implement. In a simulated and representative tournament, it achieves high average scores and wins the tournament for various parameter settings. PREDICTOR thereby relies on a brief phase of exploration to improve its model, and it can evolve morality from intrinsically selfish behavior.