Contribution
- A probabilistic treatment of repeated interaction in biology: after each move the same pair meets again with probability w.
- An account of the whole “chronology” of cooperation: robustness among many strategies, stability once established, and initial viability in an asocial population.
- Applications at the microbial level, including speculative ones on disease and chromosomal nondisjunction (the authors’ own list, p. 1391).
Key insight
Defection is the only solution of the one-shot game and of a game of known length, but when the end is uncertain a strategy that cooperates first and then copies its partner cannot be beaten once established, provided the future matters enough. Getting there from universal defection needs a foothold: relatives, whose payoffs are partly shared, or a cluster of reciprocators who mostly meet each other.
Assumptions
- Game. Two-player Prisoner’s Dilemma with T > R > P > S and R > (S + T)/2, payoffs in fitness; numerical values R = 3, T = 5, P = 1 in the tournaments (Fig. 1).
- Continuation. A fixed probability w that the pair meets again; payoffs discounted by w^(n−1). Choices simultaneous and discrete, said to be equivalent for most purposes to continuous or sequential interaction (note 21).
- Recognition. Players can recognize a previous partner and remember at least the last move.
- No commensurability of the two sides’ payoffs is needed, so host–symbiont pairs qualify.
- Clustering. A fraction p of a cluster member’s interactions are with other members; the cluster is a negligible share of the residents’ interactions.
Key results
- Robustness. First tournament: 14 entries and a random strategy, round robin, 200 moves; TIT FOR TAT (Rapoport) scored highest. Second: 62 entries from six countries, expected length 200 (w = .99654); TIT FOR TAT won again. Its success is attributed to being nice, provocable and forgiving. In an ecological replay it displaced all others and went to fixation.
- Stability. Against TIT FOR TAT the best reply is either ALL D or alternation of D and C. TIT FOR TAT v. itself earns R/(1 − w); ALL D earns T + wP/(1 − w); alternation earns (T + wS)/(1 − w²). Hence no strategy does better than TIT FOR TAT iff w ≥ (T − R)/(T − P) and w ≥ (T − R)/(R − S) (condition 1).
- ALL D is evolutionarily stable for every w.
- Initial viability. Kinship: reckoning payoffs in inclusive fitness can remove T > R and P > S, so cooperation can begin between relatives and spread as conditionality is acquired. Clustering: a cluster of TIT FOR TAT earns p[R/(1 − w)] + (1 − p)[S + wP/(1 − w)], which beats ALL D’s P/(1 − w) for large enough p and w.
- Ratchet. A nice strategy that is evolutionarily stable cannot be invaded by a cluster of any other strategy.
- Applications. Fig wasps and figs (trees abort figs whose wasps under-pollinate), cleaner mutualisms tied to fixed sites, territorial birds that tolerate neighbours’ song more than strangers’, face recognition and prosopagnosia, symbionts turning parasitic when a host sickens or ages, and two labelled speculations (cancer onset, maternal-age nondisjunction).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | TIT FOR TAT is robust in a variegated environment | moderate | two tournaments with submitted strategies and an ecological replay; the environment is whatever was submitted |
| C2 | TIT FOR TAT is evolutionarily stable iff condition (1) holds | moderate as “no strategy does better”; weak as strict ESS | the reduction to ALL D and alternation is sound; ALL C ties, so stability is neutral |
| C3 | ALL D is evolutionarily stable for all w | strong | one-line argument |
| C4 | Kinship or clustering can establish reciprocity from ALL D | strong for clustering (explicit payoff), moderate for kinship (verbal) | Initial Viability section; formal proofs referred to Axelrod’s forthcoming APSR paper (note 19) |
| C5 | A nice ESS resists invasion by clusters | strong | weighted-average argument |
| C6 | Symbionts should turn parasitic as w falls; nondisjunction may be chromosomal defection | speculation, labelled so | Applications |
Concepts
- Nice strategy — never the first to defect.
- Robustness / stability / initial viability — the three questions the paper separates.
- Evolutionarily stable — here, “cannot be invaded by a rare mutant adopting a different strategy”, applied as “no mutant does better”.
Connections
- Trivers 1971 (LIT-510) is the “pioneer account” of reciprocation; Hamilton had given Trivers the Prisoner’s Dilemma framing.
- Hamilton 1964 (LIT-497) supplies the kinship route to initial viability; note 3 cites Part I.
- Maynard Smith & Price 1973 (LIT-505) supply the ESS concept (note 11), and note 16 points to their contest model as another game with gains for cooperation.
- Alexander is cited (note 27) for uncertain paternity, not for moral systems.
- Nowak 2006 (LIT-501) restates direct reciprocity as w > c/b for a donation game and reports TIT FOR TAT’s weakness under error.
Bearing on the record
- Exchange domain. Grounded: reciprocity is stable only with retaliation, recognition, and a sufficient chance of meeting again.
- Kin domain. Used, not added to: kinship is the starter.
- Group domain. Clustering is a population-structure mechanism, not coordination for mutual benefit; it does not ground MAC’s group domain.
- Other MAC domains. Not grounded. The paper is not about morality.
- LIT-101. The w of condition (1) is the formal content of the “shadow of the future” behind the delay-of-gratification premise in moral disciplining (NOTE-132, A1).
- No THEORY is indicated. No instruction for ML practice.
Note of 2026-10-16: THEORY-672 now cites LIT-498 as a source.
Limitations
- Strategies are deterministic and error-free; noise is not considered.
- The tournament environment is the set of strategies people submitted.
- Stability is shown against single mutants and clusters, not against neutral drift followed by invasion.
- Formal proofs are deferred to another paper.
Open questions
- Is TIT FOR TAT, or any pure strategy, stable once neutral mutants such as ALL C can drift in? (Answered in the negative by Boyd & Lorberbaum 1987, not in the record.)
- Does w measurably fall with a partner’s illness or age, and does cooperation fall with it, as the Applications predict?
Corrections
- none to a seeded skim (there was no seed)
- “Evolutionarily stable” is neutral stability here. The proof establishes that no strategy does better than TIT FOR TAT against TIT FOR TAT. A strategy that always cooperates does exactly as well, so by Maynard Smith and Price’s two-part definition (LIT-505) TIT FOR TAT is not strictly an ESS. Boyd & Lorberbaum (1987, DOI-10.1038/327058a0) proved that no pure strategy is evolutionarily stable in the repeated Prisoner’s Dilemma; checked on Crossref, not read.