Aggression — Stability and the Selfish Machine

Core idea

Why don't animals fight to the death over every meal? The orthodox answer — that they restrain themselves "for the good of the species" — falls to the same group-selection objection as in topic 1. Dawkins reaches instead for John Maynard Smith's evolutionary stable strategy (ESS) to give a gene-level explanation of why animal conflicts are usually ritualized rather than lethal.

Dawkins's argument: Animals restrain their aggression not from any species-level wisdom but because, in the long run, no individual that fought to the death every contest left more descendants than rivals who were more strategic. An ESS is a strategy that, when most of the population adopts it, cannot be invaded by any rare alternative. Restraint, in many situations, is the ESS — not because it is "best for the species" but because it is uninvadable.

The topic introduces the formal apparatus of evolutionary game theory and walks through several worked games: hawk vs. dove, hawk-dove-retaliator, and the more elaborate hawk-bully-retaliator-prober-retaliator. Each shows the same lesson: the stable mix of strategies — not the optimal one — is what evolution finds.

Why it matters

From "for the good of the species" to "uninvadable"

The fundamental shift the topic accomplishes is in what counts as an explanation. The old framing asks: what outcome would be best for the species, and how is this behavior arranged to produce it? The ESS framing asks: given that every animal acts in its own (gene's) interest, what is the stable outcome? The two need not coincide.

Indeed they often do not. Population biology shows many cases where the ESS is worse for everyone than some hypothetical cooperative arrangement — but the cooperative arrangement is not stable, because any individual who defected from it would prosper. Stability beats optimality.

Hawk and dove

The book's central worked example is the simplest two-strategy game. Two animals contest a resource of value V. A hawk always fights and risks injury of cost C. A dove threatens but retreats if challenged. The payoff matrix:

• Hawk vs. Hawk: each wins half the time, costs are shared — average payoff (V − C)/2
• Hawk vs. Dove: hawk wins; dove costs nothing — hawk gets V, dove gets 0
• Dove vs. Dove: long display, eventually one wins — each gets V/2 on average

If V > C, hawks win cleanly and a pure hawk strategy is stable. But typically V < C (the cost of serious injury outweighs the value of one meal), and then neither pure strategy is stable. A population of doves is invaded by a single hawk (who beats every dove and reaps V each time). A population of hawks is invaded by a dove (who avoids the high-cost hawk-vs-hawk fights). The ESS is a mixed strategy — a stable proportion of each strategy in the population, or, equivalently, every individual playing hawk with some probability p.

Retaliator: the conditional strategy

Once Dawkins introduces a third strategy — the retaliator, who plays dove unless attacked and then plays hawk — the analysis gets more interesting. A pure retaliator population is uninvadable by either hawk (who pays the high cost of hawk-vs-hawk when retaliators defend) or dove (whose costs and benefits are matched). Retaliator is a conditional strategy, and conditional strategies often beat pure ones because they extract the benefits of dove behavior in safe contexts and hawk behavior in dangerous ones.

This is a recurring theme of the book: the most successful strategies are if-then rules, not blanket policies. "If big and strong, then fight; if small or wounded, then retreat" outperforms either "always fight" or "always retreat."

Asymmetric contests

Real-world contests rarely involve identical contestants. One is bigger, one is older, one is the prior resident of a territory, one approached first. Each asymmetry can be used as a signal that resolves the contest without combat — if both contestants use the same convention. The convention "resident wins, intruder retreats" is an ESS so long as resident-recognition is reliable; in real species (Speckled Wood butterflies, certain spiders) it is observed to hold. The opposite convention ("intruder wins, resident retreats") is also a theoretical ESS — and is rarely if ever observed, presumably because it is harder to maintain.

The deeper point: many of the rules animals use to resolve disputes are arbitrary conventions maintained by selection because they are stable, not because they are fair or efficient.

Why animals "limit" their weapons

The topic closes by addressing the question that troubled Konrad Lorenz: why don't animals evolve more lethal weapons? The answer in ESS terms is straightforward. Imagine a species in which contests are settled by display. A mutant who evolved a slightly more lethal weapon would, in any given contest, defeat the displayers. But other displayers would respond — by evolving their own more-lethal weapons, or by adjusting their assessment rules. The arms race drives both sides toward more dangerous contests, with worse outcomes for all. Stable evolution does not always find this peaceful equilibrium — but when it does, neither side fighting harder is the ESS, not because the species "agreed" to restraint but because, given everyone else's strategy, you do worse by escalating.

Key takeaways

Mental model

Copy sourceZoom

Practical application

Stop asking "what's best for everyone" — ask "what's stable"

The ESS framing is a general tool, well beyond biology. Whenever you analyze a multi-agent situation — markets, politics, organizations, traffic — the question "what would be best for all parties together?" is the wrong question if no party has the means to enforce that outcome. The right question is: "given that each party acts in its own interest, what is the stable mix of strategies?" That is almost always different from, and worse than, the cooperative ideal.

Look for the conditional rule

In any situation where outcomes look surprisingly restrained or surprisingly aggressive, ask: what is the conditional rule that makes this stable? "If resident, defend; if intruder, retreat." "If above some threshold, escalate; if below, accommodate." The conditional rule, once you find it, usually reveals why the equilibrium holds — and what would break it.

Note the arbitrariness of conventions

Many social conventions — driving on the left or right, addressing seniors first, who pays for dinner — are arbitrary in the same sense the topic describes. They are not optimal; they are merely stable. Knowing this matters when you propose to change one. The right intervention is rarely "convince everyone the new convention is better." It is "tilt the immediate incentives just enough that the new convention becomes the locally stable one."

Example

Consider queueing. In an orderly queue, the convention is "first arrival, first served." This is an ESS so long as most queuers follow it: a single cheater who jumps the line will be (usually) shamed back; if many people start cheating, the convention collapses into a mob.

The opposite convention, "loudest voice, first served," is also an ESS in principle. There are real societies and contexts where it prevails (some traffic situations, certain markets). Neither convention is fairer in any deep sense; both are uninvadable so long as enough people follow them. What stabilizes one over the other is local enforcement, expectations, and small punishments that change the individual payoff of defecting.

This is the topic's deep lesson moved out of biology: stable conventions are not chosen for their fairness; they persist because, given that most others follow them, no individual gains by deviating.

Related lessons