Game theory is the mathematical study of strategic decision-making — situations where the outcome for each participant depends not only on their own choices, but also on the choices made by others. Founded as a rigorous mathematical discipline by John von Neumann and Oskar Morgenstern in 1944, it has become an indispensable tool across economics, evolutionary biology, political science, and computer science.
Imagine two suspects arrested separately by police, each secretly offered the same deal: betray the other and go free (while the other gets a harsh sentence), or stay silent. If both stay silent, both get a light sentence. If both betray each other, both get a moderate sentence. This is the famous Prisoner's Dilemma — and the mathematics of exactly how a rational person should decide, given that the other person's choice is unknown but strategically relevant, is what game theory studies.
In game theory, a "game" doesn't necessarily mean anything playful — it refers formally to any strategic interaction involving multiple decision-makers ("players"), specific choices available to each ("strategies"), and consequences (outcomes or "payoffs") that depend jointly on everyone's combined choices. Business competition, international diplomacy, evolutionary competition between species, and auctions are all rigorously analysable as "games" in this precise mathematical sense.
Formally, a game specifies: the players involved, the strategies (possible choices) available to each player, and a payoff function assigning a numerical value (representing utility, profit, or similar) to each player for every possible combination of strategies chosen by all players.
In a zero-sum game, one player's gain is always exactly another's loss (the payoffs always sum to zero) — like most standard board games (chess, poker). In a non-zero-sum game, players can potentially both gain or both lose together, as in the Prisoner's Dilemma, opening the door to genuine cooperation as a rational strategic possibility.
Von Neumann's minimax theorem (1928) proves that in any finite, two-player, zero-sum game, there always exists an optimal strategy for each player that guarantees the best possible worst-case outcome — minimising the maximum possible loss. This foundational result, predating von Neumann and Morgenstern's landmark 1944 book, is generally regarded as marking the true birth of game theory as a rigorous, self-contained mathematical discipline.
| B stays silent | B betrays | |
|---|---|---|
| A stays silent | Both: 1 year | A: 10 years, B: free |
| A betrays | A: free, B: 10 years | Both: 5 years |
Regardless of what B does, A is always individually better off betraying — this is called a dominant strategy. Since the same logic applies symmetrically to B, both rationally betray, landing at "both: 5 years" — worse for both than the unreachable cooperative "both: 1 year" outcome.
When the same game is played repeatedly between the same players (rather than just once), cooperation can become a genuinely rational strategy, since players can reward good behaviour and punish betrayal in future rounds. Robert Axelrod's famous computer tournaments (1980) found that a remarkably simple strategy called "Tit-for-Tat" (cooperate first, then simply mirror whatever the opponent did last round) consistently outperformed far more complex strategies submitted by expert game theorists.
Non-cooperative game theory (the focus of most classical analysis, including the Prisoner's Dilemma and Nash equilibrium — see that entry) assumes players cannot make binding agreements and must act in their own individual self-interest. Cooperative game theory instead studies situations where groups (coalitions) of players can form binding agreements, focusing on how the resulting joint gains should be fairly divided among coalition members — with the Shapley value (Lloyd Shapley, 1953) providing one particularly influential and mathematically elegant general solution to this fair-division problem.
Mechanism design, sometimes informally called "reverse game theory," starts from a desired social or economic outcome and works backward to design the specific rules of a game (such as an auction format) that will reliably induce self-interested, rational players to actually produce that desired outcome through their own strategic behaviour. This field, pioneered substantially by Leonid Hurwicz, Eric Maskin, and Roger Myerson (jointly awarded the 2007 Nobel Memorial Prize in Economic Sciences), underlies the careful design of real-world auctions, voting systems, and resource-allocation mechanisms.
John Maynard Smith and George Price (1973) adapted game theory to evolutionary biology, introducing the concept of an evolutionarily stable strategy — a strategy that, once adopted by the overwhelming majority of a population, cannot be successfully invaded or displaced by any small, initially rare alternative strategy. This framework has proven successful in explaining a wide range of otherwise puzzling animal behaviours, including patterns of aggression, cooperation, and specific ritualised contest behaviours observed across numerous species.
Many real-world strategic situations involve genuine uncertainty about other players' available strategies, preferences, or payoffs — not just uncertainty about which strategy they will actually choose. John Harsanyi (1967–68, also a co-recipient of the 1994 Nobel Memorial Prize alongside Nash and Selten) developed a rigorous framework for analysing such "games of incomplete information," substantially extending the scope and real-world applicability of classical game theory to situations involving imperfect or asymmetric knowledge between the players.
While Nash proved that an equilibrium always exists in any finite game (see the Nash Equilibrium entry), actually computing that equilibrium can be computationally very difficult. The problem of finding a Nash equilibrium has been proven to be PPAD-complete (Daskalakis, Goldberg and Papadimitriou, 2009) — a specific complexity class believed, though not proven, to lack any generally efficient algorithm, connecting game theory directly to deep open questions in computational complexity theory (see the P vs NP entry).