Entered by Anatol Rapoport in Robert Axelrod's computer tournament (1980) and analyzed by Axelrod (The Evolution of Cooperation, 1984).
Cooperates on the first move and then copies the other side's last move, a clear and forgiving rule that sustains cooperation in repeated dealings.
Two freight carriers that pass each other overflow loads, for twenty months: pick a rule for each and read the score
Each month each carrier either shares its good overflow loads with the other or holds them back and passes on only the poor ones. Both sharing earn 3 each; both holding back, 1 each; one holding back against one sharing takes 5 and leaves 0. Pick our rule and the partner's; the line gives the twenty-month score and the rule's total against all four partners. It opens on tit for tat against a partner who holds back once by mistake: switch our rule to forgiving the first slight. Illustrative.
Our rule
The partner's rule
- Pick a rule in each row to see the twenty-month score.Twenty months of shared loads: 60 to 60. Neither holds back first, so neither has a slight to copy. Across the four partners: 189.Shared throughout: 60 to 60. Across the four partners: 189.Exploited in month one only, then matches every month: 19 to 24, the most it can lose by. Across the four partners: 189.The month-6 slip is copied, copied back, and echoes every other month to the end: 50 to 55. Across the four partners: 189. Try forgiving the first slight.Shared throughout: 60 to 60. Against a fair partner forgiving never comes up. Across the four partners: 195, the most.Shared throughout: 60 to 60. Across the four partners: 195.Forgiving costs one extra month, then it matches every month: 18 to 28. Across the four partners: 195.The slip goes unanswered, the partner shares again, and cooperation resumes at once: 57 to 62. Across the four partners: 195.Shared throughout: 60 to 60. Across the four partners: 177.Shared throughout: 60 to 60. Across the four partners: 177.Never answering invites every slight: 0 to 100. This one match sinks the rule. Across the four partners: 177.The slip passes and sharing goes on: 57 to 62. Across the four partners: 177, because the partner who always holds back takes everything.Wins month one, then both hold back for nineteen months: 24 to 19. It wins the match and earns far less than two sharers. Across the four partners: 168, the least.Takes everything from a partner who never answers: 100 to 0. Across the four partners: 168.Nobody shares from month one: 20 to 20, a third of what two sharers earn. Across the four partners: 168.Same as against tit for tat: 24 to 19, a win at a low score. Across the four partners: 168.Shared throughout: 60 to 60. Across the four partners: 172.Shared throughout: 60 to 60. Across the four partners: 172.Exploited once, then holds back forever: 19 to 24. Across the four partners: 172.One slip starts a standoff for the rest of the twenty months: 33 to 33, bad for both. Across the four partners: 172.
- Twenty months of shared loads: 60 to 60. Neither holds back first, so neither has a slight to copy. Across the four partners: 189. (ours: Tit for tat, rival: Tit for tat)
- Shared throughout: 60 to 60. Across the four partners: 189. (ours: Tit for tat, rival: Always share)
- Exploited in month one only, then matches every month: 19 to 24, the most it can lose by. Across the four partners: 189. (ours: Tit for tat, rival: Always hold back)
- The month-6 slip is copied, copied back, and echoes every other month to the end: 50 to 55. Across the four partners: 189. Try forgiving the first slight. (ours: Tit for tat, rival: Holds back once)
- Shared throughout: 60 to 60. Against a fair partner forgiving never comes up. Across the four partners: 195, the most. (ours: Forgive the first slight, rival: Tit for tat)
- Shared throughout: 60 to 60. Across the four partners: 195. (ours: Forgive the first slight, rival: Always share)
- Forgiving costs one extra month, then it matches every month: 18 to 28. Across the four partners: 195. (ours: Forgive the first slight, rival: Always hold back)
- The slip goes unanswered, the partner shares again, and cooperation resumes at once: 57 to 62. Across the four partners: 195. (ours: Forgive the first slight, rival: Holds back once)
- Shared throughout: 60 to 60. Across the four partners: 177. (ours: Always share, rival: Tit for tat)
- Shared throughout: 60 to 60. Across the four partners: 177. (ours: Always share, rival: Always share)
- Never answering invites every slight: 0 to 100. This one match sinks the rule. Across the four partners: 177. (ours: Always share, rival: Always hold back)
- The slip passes and sharing goes on: 57 to 62. Across the four partners: 177, because the partner who always holds back takes everything. (ours: Always share, rival: Holds back once)
- Wins month one, then both hold back for nineteen months: 24 to 19. It wins the match and earns far less than two sharers. Across the four partners: 168, the least. (ours: Always hold back, rival: Tit for tat)
- Takes everything from a partner who never answers: 100 to 0. Across the four partners: 168. (ours: Always hold back, rival: Always share)
- Nobody shares from month one: 20 to 20, a third of what two sharers earn. Across the four partners: 168. (ours: Always hold back, rival: Always hold back)
- Same as against tit for tat: 24 to 19, a win at a low score. Across the four partners: 168. (ours: Always hold back, rival: Holds back once)
- Shared throughout: 60 to 60. Across the four partners: 172. (ours: Hold a grudge, rival: Tit for tat)
- Shared throughout: 60 to 60. Across the four partners: 172. (ours: Hold a grudge, rival: Always share)
- Exploited once, then holds back forever: 19 to 24. Across the four partners: 172. (ours: Hold a grudge, rival: Always hold back)
- One slip starts a standoff for the rest of the twenty months: 33 to 33, bad for both. Across the four partners: 172. (ours: Hold a grudge, rival: Holds back once)
An example
Two regional trucking firms had an old arrangement: when one had more freight than trucks, it passed the overflow to the other. Over a year the deal had soured. Each suspected the other of keeping the profitable loads and handing over only the short-notice, low-rate ones, and each had started doing the same in return.
The smaller firm's operations director decided to stop guessing the partner's mood and adopt a rule. She told her dispatchers in plain terms: we pass on good loads as long as they do; if they send us a month of poor loads, the next month we send them only what we would anyway; the month they go back to sharing, so do we. Within two quarters the partner had tested the rule twice and been matched twice. When a new dispatcher on their side sent a bad batch by mistake and the next month was fair again, she let it pass without answering. A year later the overflow arrangement was worth about $600,000 a year to each firm.
Why it works
In a single deal, keeping the good loads is tempting whatever the partner does. Across many months with the same partner, the shadow of the future changes the math: one greedy month buys one good month and an answered slight every month after. Tit for tat makes that shadow visible with the simplest possible rule.
Its strength is in four properties Axelrod drew from his tournaments. It is nice: it never defects first, so it never starts a feud. It retaliates: a defection is answered at once, so exploiting it pays only once. It forgives: the moment the rival returns to cooperating, so does it. And it is clear: a partner can read the rule from a few moves and see that cooperation is the best reply. The weakness is noise. A single mistake can echo back and forth indefinitely, which is why practitioners add a little forgiveness.
The thinking behind it
Anatol Rapoport Anatol Rapoport, a mathematical psychologist and game theorist, submitted tit for tat to both of Robert Axelrod's computer tournaments of the iterated Prisoner's Dilemma around 1980. It was the simplest strategy entered, and it won both. Rapoport had worked on the game for years before, including in Prisoner's Dilemma (1965), written with Albert M. Chammah. His entry made the case that a short, readable rule can beat far more elaborate ones.
Robert Axelrod Robert Axelrod, a political scientist, ran two computer tournaments of the iterated Prisoner's Dilemma around 1980, in which strategies submitted by others played one another over many rounds. In The Evolution of Cooperation (1984) he reported that tit for tat, which cooperates first and then copies the other player's last move, did best, and he used the results to explain how cooperation can take hold among players who each look after themselves.
How to run it
- Check that the dealings repeat. The rule works with partners or rivals met again and again; in a one-off deal it gives no protection.
- Cooperate on the first move. Share the good loads, the information, or the agreed terms first, so the relationship starts on the good outcome.
- Then match their last move. Answer a defection promptly and in proportion, on the same issue, and never escalate beyond it.
- Return to cooperation at once. When they go back to cooperating, resume cooperating in the next round, with no lingering penalty.
- Forgive the occasional slip. Let a single, apparently accidental defection pass once, so a mistake does not start an echo of answered cuts.
- Make the rule visible. State it to the team and act consistently, so the other side can read it from a few rounds.
Where it goes wrong
The classic failure is the echo. One misread move, a bad batch that was a new dispatcher's error, gets answered, the answer gets answered, and the two sides alternate punishment for a year. It shows up as defections that alternate between the parties. The fix is generosity: let one apparent mistake pass, or talk to the other side before answering.
A second failure is using the rule where dealings do not repeat, or near the known end of a relationship, when the shadow of the future is gone and defection pays. Check how many rounds remain. A third is disproportionate retaliation: answering one bad month by cutting the partner off entirely turns a correction into a feud. Match the move, not the grievance.
When to use something else
- Repeated Games. When the question is whether cooperation can hold at all, given how much each side values future rounds against a one-time gain.
- Game Theory. When the interaction is one-off or has different payoffs, and the first job is laying out the moves and their outcomes.
- Signaling Games. When the other side's intentions are hidden and the problem is reading or sending a credible signal of type.
Where it sits in SOLVE eX
Go to the source
- Effective Choice in the Prisoner's Dilemma, Journal of Conflict Resolution 24(1), 1980
- The Evolution of Cooperation, 1984