Home > Positioning > Persons > Axelrod

Robert Axelrod (1943–)

Axelrod showed that cooperation can emerge and sustain itself among self-interested agents without central enforcement, shared morality, or even mutual understanding. His iterated prisoner’s dilemma tournaments demonstrated that a simple strategy — tit-for-tat, which cooperates on the first move and then copies the other player’s previous move — outperforms sophisticated strategies in repeated interaction. The result is foundational for evolutionary game theory, political science, and complex adaptive systems research: cooperation does not require altruism; it requires repetition, recognition, and the possibility of future interaction.

Life

Born 1943. BA from the University of Chicago; PhD in political science from Yale (1969). Faculty at the University of California, Berkeley, then the University of Michigan, where he has spent the bulk of his career as Walgreen Professor for the Study of Human Understanding in the Department of Political Science and the Gerald R. Ford School of Public Policy. Member of the National Academy of Sciences. MacArthur Fellow (1987). Part of the BACH group (Burks, Axelrod, Cohen, Holland) at Michigan — the cross-disciplinary collaboration that prefigured the Santa Fe Institute’s approach.

The prisoner’s dilemma tournaments

The Evolution of Cooperation (Basic Books, 1984) reports the results of two computer tournaments. Axelrod invited game theorists, economists, psychologists, and others to submit strategies for the iterated prisoner’s dilemma — a game in which two players repeatedly choose to cooperate or defect, with payoffs that make mutual cooperation better for both than mutual defection, but unilateral defection better for the defector.

In both tournaments, the winner was tit-for-tat, submitted by Anatol Rapoport. The strategy is the simplest possible: cooperate on the first move; thereafter, do whatever the other player did on the previous move. It is nice (never defects first), retaliatory (punishes defection immediately), forgiving (returns to cooperation as soon as the other player does), and transparent (its logic is immediately legible to the other player).

The result is counterintuitive. Tit-for-tat never “beats” any opponent in a single game — it can at best tie. Its success comes from eliciting cooperation from other cooperative strategies and minimising losses against defectors. In a population of strategies competing over many rounds, tit-for-tat accumulates more total payoff than any other strategy. Cooperation wins not by dominating but by sustaining mutually beneficial interaction.

The conditions for cooperation

Axelrod distilled the tournament results into conditions under which cooperation can emerge among self-interested agents:

Repetition. The interaction must be repeated, with no known last round. If the game has a known endpoint, backward induction drives rational players to defect from the start. Open-ended repetition is what gives future consequences weight.

Recognition. Players must be able to identify their partners and remember past interactions. Without recognition, there is no basis for reciprocity — defectors cannot be distinguished from cooperators.

The shadow of the future. Future payoffs must be sufficiently valued relative to present ones. If agents discount the future heavily, the short-term gain from defection outweighs the long-term benefit of cooperation.

These conditions do not require altruism, trust, or central authority. They are structural: when the game has the right properties, cooperation is the self-interested strategy.

Complexity of Cooperation

The Complexity of Cooperation: Agent-Based Models of Competition and Collaboration (Princeton University Press, 1997) extends the programme. The book collects work on the evolution of norms, the dynamics of cultural dissemination, and the landscape of strategies in more complex games. Axelrod moves from the clean two-player iterated game to populations of agents with spatial structure, cultural traits, and evolving strategies — the territory that CAS occupies.

The cultural dissemination model is notable: agents on a grid have multiple cultural features, each with multiple traits. Agents interact more readily with similar neighbours and adopt traits from them. The result: convergence to homogeneous cultural regions, with the number of regions depending on the initial diversity and the number of features. The model shows how cultural boundaries can emerge from local interaction without pre-existing group identities.

The BACH group

Axelrod’s collaboration with Holland, Arthur Burks, and Michael Cohen at Michigan was a cross-disciplinary research group before SFI existed. Burks brought logic and automata theory, Holland brought genetic algorithms and adaptive systems, Cohen brought organisational theory, and Axelrod brought game theory and political science. The group’s work on cooperation, adaptation, and complexity fed directly into the CAS programme.

Where Axelrod stops

Axelrod’s agents are strategic but simple — they have fixed strategies (or strategies that evolve through selection, not learning). Tit-for-tat does not change its rule based on experience; it applies the same logic throughout. The evolution of cooperation operates through differential reproduction of strategies in a population, not through individual agents learning and adjusting their behaviour. The step from population-level selection of strategies to individual-level adaptive agents with internal models is the step that Holland’s CAS framework takes.

Key works

See also: Holland · Schelling · Complex Adaptive Systems · Mutualism