Loading field map
When two people bargain over a price, a firm decides whether to enter a market, or a species evolves a cooperative behavior, each faces the same underlying challenge: the outcome depends not only on one's own choices but on the choices of others. This strategic interdependence is the raw material of game theory. Since its modern formulation in the mid-twentieth century, the field has branched into several distinct frameworks, each built on different assumptions about who the players are, what they know, how they reason, and what counts as a solution. The history of applied game theory is the story of these frameworks responding to the limitations of their predecessors—sometimes by relaxing rationality, sometimes by changing the unit of analysis, and sometimes by inverting the question entirely.
Game theory's first systematic framework appeared in John von Neumann and Oskar Morgenstern's 1944 book Theory of Games and Economic Behavior. Classical Game Theory focused on two-player zero-sum games—situations where one player's gain is exactly the other's loss. Von Neumann's minimax theorem showed that in such games there is a value that each player can guarantee through a mixed strategy, and that rational play leads to that value. This framework assumed perfectly rational, utility-maximizing players with complete information about the game structure. It was a powerful starting point, but its scope was narrow: most real strategic interactions are not zero-sum, and the minimax solution does not generalize easily to games with more than two players or to cooperative possibilities.
Published in the same 1944 volume, Cooperative Game Theory took a different approach. Instead of modeling the detailed strategic moves of individuals, it asked what outcomes groups of players (coalitions) could achieve if they could make binding agreements. The central question became: how should the total payoff from cooperation be divided among the players? The core (the set of payoff allocations that no coalition can improve upon) and the Shapley value (a fair division based on each player's marginal contribution) became the framework's signature tools. Cooperative Game Theory coexisted with the classical framework from the start, but it deliberately abstracted away from the bargaining process itself. It described what coalitions could achieve, not how strategic interaction would lead to that achievement. This made it useful for problems like cost allocation or voting power, but it left open the question of whether the predicted outcomes would actually arise from self-interested behavior.
John Nash's work in the early 1950s transformed the field by providing a unified solution concept for all games, not just zero-sum ones. The Nash equilibrium—a profile of strategies where no player can benefit by unilaterally changing their own strategy—became the cornerstone of Non-cooperative Game Theory. Unlike the cooperative framework, non-cooperative theory insisted on modeling the full strategic interaction, including the moves available to each player and the information they have. It did not assume binding agreements; instead, it asked what outcomes could be sustained when each player acts in their own interest, anticipating the actions of others. This framework was more general than the classical one (it subsumed zero-sum games as a special case) and more detailed than the cooperative one (it explained how outcomes emerge from individual choices). Over time, non-cooperative theory largely absorbed the classical framework and provided a foundation for analyzing bargaining, oligopoly, and auctions. Yet it retained the assumption of perfect rationality—players are assumed to have unlimited computational ability and common knowledge of each other's rationality—which later frameworks would challenge.
By the 1970s, researchers began to invert the non-cooperative question. Instead of taking the rules of a game as given and predicting the outcome, Mechanism Design Theory asked: given a desired social outcome, what rules (or "mechanism") would lead self-interested players to produce that outcome? This framework, pioneered by Leonid Hurwicz, Eric Maskin, and Roger Myerson, treats the game designer as a player who chooses the structure of the game. The revelation principle—that any outcome achievable through a complex mechanism can also be achieved through a direct truthful-revelation mechanism—became a central tool. Mechanism design extended non-cooperative theory by applying its equilibrium logic to the design of institutions, from auctions to voting rules to public-good provision. It did not replace non-cooperative theory but rather used it as infrastructure, while shifting the focus from prediction to engineering.
Also in the 1970s, a different challenge to the rational core emerged from biology. John Maynard Smith and George R. Price introduced Evolutionary Game Theory, which replaced the rational, calculating agent with a population of individuals whose strategies are inherited and subject to natural selection. The key solution concept, the evolutionarily stable strategy (ESS), is a strategy that, once established in a population, cannot be invaded by any alternative strategy. This framework did not assume any conscious reasoning; instead, it modeled how behaviors spread through differential reproduction. Evolutionary game theory coexists with non-cooperative theory but operates on a different level of analysis. While non-cooperative theory explains equilibrium as a product of rational deliberation, evolutionary theory explains it as a product of population dynamics. The two frameworks sometimes agree on the same equilibrium (e.g., the Hawk-Dove game yields a mixed-strategy Nash equilibrium that is also an ESS), but they diverge when multiple equilibria exist or when learning and mutation matter. Evolutionary game theory also opened the door to studying norms, conventions, and cooperation without assuming rationality—a theme that behavioral game theory would later take up from a psychological angle.
By the 1990s, the rise of the internet and large-scale computational systems created new pressures. Algorithmic Game Theory emerged at the intersection of computer science and economics, asking how strategic agents interact in computational environments. Its central concerns include the complexity of computing equilibria (e.g., the PPAD-completeness of finding a Nash equilibrium), the design of algorithms that are robust to strategic manipulation (e.g., truthful auction mechanisms), and the analysis of inefficiency in games through concepts like the price of anarchy—the ratio between the worst-case equilibrium outcome and the socially optimal outcome. This framework does not reject non-cooperative theory; rather, it adds computational constraints and tractability as first-class concerns. It narrows the scope of classical rationality by showing that even if players are perfectly rational, they may not be able to compute an equilibrium in practice. Algorithmic game theory is now central to the design of online markets, routing protocols, and blockchain mechanisms.
At roughly the same time, Behavioral Game Theory began importing insights from psychology and experimental economics to modify the assumptions of non-cooperative theory. Instead of assuming perfect rationality, behavioral models incorporate bounded rationality, social preferences (e.g., fairness, reciprocity), and systematic cognitive biases. Colin Camerer's work on quantal response equilibrium, for example, replaces deterministic best response with probabilistic choice, capturing the idea that players make mistakes but learn from experience. Behavioral game theory does not abandon the equilibrium concept; it refines it by adding psychological realism. It coexists with non-cooperative theory as a parallel tradition, often testing where the rational model succeeds and where it fails. In the laboratory, behavioral models better explain outcomes in games like the ultimatum game, public goods game, and centipede game, where fairness and trust matter. The framework has also begun to merge with mechanism design (behavioral mechanism design) and with evolutionary theory (studying how social preferences evolve).
Today, no single framework dominates applied game theory. Non-cooperative game theory remains the lingua franca, providing the baseline model for most economic and political analysis. Cooperative game theory continues to be used in cost allocation, voting power, and network formation. Mechanism design is the backbone of market design and auction theory. Evolutionary game theory is standard in biology and increasingly in social science. Algorithmic game theory drives the analysis of internet platforms and cryptocurrencies. Behavioral game theory informs policy interventions (nudges) and experimental economics.
The frameworks agree on the fundamental importance of strategic interdependence and equilibrium analysis. They disagree, however, on what kind of equilibrium to use and what assumptions about agents are justified. The deepest tension is between the rational core (non-cooperative theory and mechanism design) and the behavioral/evolutionary alternatives. Rational models assume that agents are consistent, forward-looking, and computationally unbounded; behavioral models argue that real agents are not, and that the models must be adjusted accordingly. Algorithmic game theory adds a third dimension: even if agents are rational, computational limits matter. The field is now moving toward synthesis—behavioral mechanism design, evolutionary models of learning in algorithms, and experiments that test both rational and behavioral predictions. The result is a pluralistic discipline where the choice of framework depends on the question: are you designing an auction, explaining a biological convention, or predicting how people bargain in a lab? Each framework offers a different lens, and the history of the field shows that the most productive advances come from understanding what each lens reveals and what it leaves out.