Cooperation, Temptation, and Escalation in Iterated Prisoner’s Dilemma Tournaments

Abstract

This study investigates the emergence, stability, and breakdown of cooperation in repeated

strategic interactions using a series of custom Iterated Prisoner’s Dilemma (IPD) tournaments.

Building on the foundational work of Robert Axelrod, nine original strategies, including Tit for

Tat, Generous Tit for Tat, Grim Trigger, and Always Cooperate, were evaluated under multiple

rule variations designed to model real world incentives such as compounding rewards, random

errors, early mistrust, and escalating temptation.

Each tournament was implemented as an AI powered computational simulation through

OpenAI’s ChatGPT-5 in a python execution environment, consisting of 200 round round-robin

matches using Axelrod’s classic payoff structure. Under a dynamic reward structure in which

mutual cooperation increased the reward payoff, Always Cooperate achieved the highest

cumulative score (46,398.5), followed closely by forgiving reciprocal strategies. Introducing

execution noise (7.5% random move errors) reduced the performance of strict retaliators, while

Tit for Two Tats ranked first with a total score of 29,856.3. When temptation payoffs increased

beyond a threshold, cooperative strategies still outperformed defect-oriented strategies, though

overall score variance declined. In a final tournament with uniformly increasing payoffs and a

rule forcing defection whenever it yielded a relative advantage, all nine strategies converged to

identical cumulative scores (51,750), demonstrating an arms-race dynamic in which escalation

eliminated strategic differentiation. These results show that cooperation is highly sensitive to

incentive design and that excessive temptation undermines relative advantage. This research was

conducted as part of the 2025Summer Undergraduate Research Experience project at Shepherd

University.