Tit for Tat¶
Summary¶
The simplest and most effective strategy for the iterated prisoner's dilemma, winner of Robert Axelrod's 1980 tournaments. Starts with cooperation, then copies the opponent's last move — immediately retaliating and immediately forgiving.
The Strategy¶
- Start with cooperation — always
- Copy opponent's last move — if they cooperated, cooperate; if they defected, defect
- No memory beyond last move — forgive immediately when opponent cooperates again
Why It Wins¶
Robert Axelrod identified four properties:
| Property | What It Does |
|---|---|
| Nice | Never defects first — prevents unnecessary trouble |
| Retaliatory | Immediately defects back — discourages persistent defection |
| Forgiving | Returns to cooperation when opponent does — restores mutual benefit |
| Clear | Predictable and intelligible — elicits long-term cooperation |
Tournament Results¶
- 14 programs competed in iterated prisoner's dilemma (200 rounds each)
- Tit for Tat never won any individual game (can only lose or draw one-on-one)
- Won the tournament by cooperating with enough players to get highest total score
- Won again in second tournament (62 programs, unknown round count)
- Generous Tit for Tat (occasionally forgiving defections) proved even more effective in chaotic conditions
Life Lessons¶
- Being nice is a strength — leading with cooperation prevents escalation
- Holding grudges is a weakness — forgiveness restores mutual benefit
- Consequence must be proportional, consistent, and clear
- A strategy always focused on winning can be the least effective at winning
- "What makes cooperation possible is the fact that the players might meet again."
In AI Agent Design¶
Tit for Tat principles apply to multi-agent systems: - Agents that always defect end up in mutual destruction - Agents that always cooperate get exploited - Tit for Tat agents establish stable cooperation with other cooperative agents while defending against defectors