Contribution
- A model of selection for altruism between non-relatives, including members of different species, sustained by withdrawal of help from cheaters (The Model, pp. 35–39).
- Three applications: host restraint in cleaning symbioses (pp. 39–43), a non-reciprocal, self-interested account of bird warning calls (pp. 43–45), and human reciprocal altruism (pp. 45–54).
- A functional analysis of the human “psychological system underlying human reciprocal altruism” (pp. 47–54), which is the first place the moral emotions are given adaptive functions in this literature.
Key insight
Reciprocity pays only if cheating is punished by the loss of future help, and once it pays, selection builds a regulating psychology around it: warm emotions to motivate giving, anger to protect the giver, guilt to repair a detected breach, and then counterfeits of each and the suspicion to catch them. The instability is the engine: “once moralistic aggression has been selected for to protect against cheating, selection favors sham moralistic aggression as a new form of cheating”, and so on (p. 54).
Assumptions
- Cost below benefit. Each altruistic act costs the giver less than it benefits the recipient, in inclusive-fitness terms (p. 36).
- Discrimination. Altruists respond to a non-altruistic act by curtailing future help (case 3, p. 37).
- Conditions (p. 37): many altruistic situations per lifetime, repeated interaction with a small set of individuals, and symmetric exposure. Favoured by long life, low dispersal, mutual dependence, long parental care, weak dominance hierarchies, and aid in combat.
- Genetics. One altruism allele a₂ versus a₁ for exposition; the argument is said not to require that every altruistic act be controlled by the same locus.
- Human section. That reciprocal altruism was important in recent human evolution and that “the underlying emotional dispositions … have important genetic components” is assumed, not shown (p. 48).
Key results
- The model. Random altruism is eliminated; kin-directed altruism is Hamilton’s case; altruism dispensed by reference to the recipient’s altruism spreads if the net benefit to altruists exceeds that to non-altruists, which holds when non-altruists are cut off (p. 37).
- Prisoner’s Dilemma (pp. 38–39). Hamilton’s communicated formulation: over 2n situations two altruists each get nR; against a non-altruist, an altruist who switches after being cheated gets S + (n − 1)P. A rare non-altruist cannot spread unless T ≫ R; a rare altruist faces a barrier that weakens as n grows.
- Cleaning symbioses. Hosts do not eat cleaners, signal departure and chase off threats. Evidence: Hediger’s grouper raised alone for six years that let a cleaner into its mouth; cleaners’ site fidelity and longevity; hosts returning to the same stations. Cleaners most dependent on cleaning are territorial, solitary and rarely eaten.
- Warning calls. Reciprocation is rejected as the explanation (“difficult to visualize how one would discover and discriminate against the cheater”). Instead a caller benefits because a predator that eats a neighbour is more likely to learn, specialize on, or stay near the caller’s species and place.
- Human psychology (items numbered (1)–(10) in the text): a complex regulating system; friendship and liking; moralistic aggression; gratitude and sympathy sensitive to cost/benefit; guilt and reparative altruism, stronger after public than private transgression; mimics of each; detection of subtle cheaters; overtures to strangers; multiparty interactions, generalized altruism and rules of exchange; developmental plasticity.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Altruism among non-kin can be selected when cheaters lose future help | moderate (verbal, with a sketched payoff comparison) | The Model; Hamilton’s PD sketch |
| C2 | Host restraint toward cleaners is reciprocal altruism | moderate | natural-history evidence for each precondition; no measure of costs or of individual re-use beyond anecdote |
| C3 | Warning calls are explained by the caller’s own interest in keeping predators from specializing | weak–moderate | argument; Trivers says discriminating evidence is “almost impossible to gather” |
| C4 | Moralistic aggression, gratitude, sympathy and guilt are adaptations regulating reciprocity | weak–moderate | functional argument fitted to social-psychology findings (Krebs 1970 review, Wallace & Sadalla 1966, Pruitt 1968, etc.) and hunter-gatherer ethnography; no test designed for it |
| C5 | Selection favours sham versions of these emotions and the ability to detect them | assertion | argument; “I know of no psychological studies on the detection of sham moralistic aggression and sham guilt” (p. 51) |
| C6 | Language-enabled multiparty systems favour norms of reciprocal conduct | assertion | one paragraph (p. 52) with ethnographic citations |
| C7 | Selection for the altruistic system may have contributed to hominid brain growth | speculation, so labelled | closing sentence (p. 54) |
Concepts
- Altruistic behaviour — benefits another organism, “not closely related”, at an apparent cost to the actor, both in inclusive-fitness terms (p. 35).
- Cheating — failure to reciprocate, “solely for convenience … no conscious intent or moral connotation is implied” (p. 36).
- Gross vs subtle cheating — not reciprocating, versus always giving back a little less than received (p. 46).
- Moralistic aggression — aggression provoked by “injustice, unfairness, and lack of reciprocity”, often “out of all proportion to the offenses” (p. 49).
- Reparative altruism — compensation by a transgressor, motivated in part by guilt (p. 50).
Connections
- Hamilton 1964 (LIT-497, LIT-506) is the foil: the model is built for cases where kin selection “can be ruled out”, and the alarm-call section argues against Hamilton’s kin explanation.
- Axelrod & Hamilton 1981 (LIT-498) formalize the same mechanism with a continuation probability w and an ESS analysis; Hamilton had already given Trivers the PD framing.
- Alexander 1987 (LIT-495) builds moral systems on indirect reciprocity, which Trivers’s multiparty section anticipates.
- Tomasello (LIT-125, NOTE-185) argues that tit-for-tat reciprocity yields gratitude, not obligation; Trivers’s guilt is the regulating emotion of exactly that reciprocity, so the two disagree on whether reciprocity is a source of moral emotion or only of liking and gratitude.
- Morality-as-cooperation. NOTE-034’s lineage paragraph and Curry 2016 (LIT-503) take reciprocity from here. Curry 2016’s exchange-domain list (trust, gratitude, punishment, revenge and forgiveness) is close to Trivers’s list.
Bearing on the record
- Exchange domain. Directly grounded, with its emotions.
- Moral emotions. The only one of the record’s cooperation sources that explains guilt, gratitude and moralistic anger. A THEORY about the function of moral emotions would start here.
- Fairness. Trivers’s “unfairness” is inequity in exchange. MAC’s division domain (dividing a disputed resource) is not grounded here, and a reader should not cite this paper for it.
- Group, hawk, dove, possession. Not grounded. Dominance appears only as an obstacle to reciprocity, aid in combat as a reciprocated service.
- No THEORY is indicated now. No instruction for ML practice.
Note of 2026-10-16: THEORY-663 now cites LIT-510 as a source.
Limitations
- The formal model is a sketch; stability against invasion is not analysed (Axelrod & Hamilton did that later).
- The human section is functional reasoning fitted to existing findings, many of them small 1960s laboratory studies of American students; no prediction is tested.
- Heritability of the emotional dispositions is assumed.
Open questions
- Do people expect “interest” on long-overdue reciprocation, highest just after sexual maturity, as the age-dependence argument predicts (p. 46)?
- Can sham guilt and sham moralistic aggression be detected, and is detection better than chance?
Corrections
- none to a seeded skim (there was no seed)
- Warning calls are not reciprocal altruism in this paper. The abstract lists “warning cries in birds” among three instances “the evolution of which the model can explain”, but the section rejects a reciprocation account and offers a self-interest model (pp. 43–45). Citing Trivers for reciprocity in alarm calls misreads him.
- Numbering. In the scan read, the psychological-system items run (1), (2), (4), (4), (5) …: moralistic aggression and gratitude are both numbered (4), and there is no (3). Checked on the rendered page image of p. 49: it is in the printed page, not an extraction artefact.