J. Maynard Smith and G. R. Price (1973), Nature 246 (5427): 15–18, 2 November 1973 — DOI-10.1038/246015a0
Key takeaways
- Conventional, non-injurious fighting can be favoured by individual selection, without group selection: in a population of retaliators an escalating “Hawk” does worse than a strategy that fights conventionally and retaliates (Table 1).
- The evolutionarily stable strategy (ESS) is defined: a strategy I such that E_I(I) > E_I(J) for all J, or, if equal, E_J(I) > E_J(J) (p. 17).
- In a contest decided by persistence alone, no pure strategy is stable; the ESS is a persistence cost drawn from p(x) = (1/v)exp(−x/v), so a stable population is polymorphic or individually variable (p. 17).
Standing in the record
Filed on 2026-10-02 at the owner’s request, as one of the evolutionary sources of morality-as-cooperation. The lineage paragraph of NOTE-034 credits “hawk–dove contests (Maynard Smith & Price)”, the concept list there glosses hawkish and dovish traits with the same citation, and Curry 2016 (LIT-503) writes that “animal conflicts are modelled … as nonzero-sum hawk–dove games … (Maynard Smith & Price, 1973)”.
NOTE-404 is the close reading of 2026-10-02, and it placed the work: Active, with a correction the MAC literature needs. This paper has no Hawk–Dove game. Its passive strategy is “Mouse”; the word “dove” does not occur. Its contestants have “identical fighting prowess”, so it does not model display of formidability followed by deference, and it has no owner–intruder asymmetry, so it does not ground possession. What it does show is that restraint is stable only when backed by retaliation (“contestants should respond to an ‘escalated’ attack by escalating in return”). In its informal “Real Animals” section it argues that conventional fighting carries information about prowess, so an animal facing a much superior opponent “will frequently retreat”, and that an unfakeable sign of recklessness (musth in elephants) can pay. That is the seed of MAC’s hawkish-display and dovish-deference domains, not their model. MAC’s possession domain rests, in Curry 2016, on Maynard Smith (1982) and Gintis (2007), neither in the record.
The stability claim also needs care. Table 1 gives Mouse 29.0 against Retaliator, the same as Retaliator against itself, and Retaliator 29.0 against Mouse, the same as Mouse against itself. By the second condition the paper itself states, Retaliator is therefore not an ESS: Mouse can drift in. The authors say “Retaliator is an ESS since no other strategy does better, though Mouse does equally well”. A comment by Gale & Eaves, Nature 254:463 (1975), DOI-10.1038/254463b0, re-examined the analysis; it was not read here.
Hamilton’s Part II (LIT-506, §6) explains restraint between relatives by relatedness. This paper explains it between non-relatives by retaliation, and says excessively dangerous weapons “would be opposed by kin selection”. Axelrod & Hamilton (LIT-498) take the ESS concept from here.
Topic. natural-sciences first: it is about animal behaviour.
mathematics for the evolutionary game theory, the “mathematical
frameworks applied to other fields” of that word’s blurb. Not
probabilistic-modeling, whose blurb is Bayesian inference and
statistical models; not moral-psychology, since the paper is not about
morality (its one human analogy is “fair and foul blows … in boxing”).
Anthology. No instruction for machine-learning practice; not held there.