Nature (Game Theory)

In game theory, Nature is a notional player that makes random moves according to a fixed probability distribution, representing events outside the strategic control of the actual players. It is the formal device by which chance, uncertainty, and exogenous shocks are incorporated into strategic models. Unlike rational players, Nature does not optimize, adapt, or learn; it simply selects from a predefined distribution, and the other players must choose their strategies in anticipation of Nature's random draws.

Nature as a System Boundary

The introduction of Nature into a game is not merely a technical convenience. It is a boundary-drawing operation: the modeler decides which aspects of the environment are endogenous (subject to strategic choice) and which are exogenous (random). This boundary is not objective. Two modelers of the same situation may draw it differently: one might treat weather as a random variable (a move by Nature), while another might model it as the outcome of strategic choices by polluters, regulators, and atmospheric physics. The choice to invoke Nature is a choice about where to stop modeling.

From a systems-theoretic perspective, Nature is the formal representation of the environment in open systems. Every system has inputs it does not control, and Nature is the game-theoretic name for those inputs. The fact that Nature's probabilities are fixed (in standard models) reflects an assumption that the environment's statistical structure is stationary and independent of the players' actions — an assumption that breaks down in dynamic systems where players alter the environment they inhabit.

Nature, Risk, and Uncertainty

The game-theoretic treatment of Nature descends from von Neumann and Morgenstern's theory of expected utility, in which players evaluate strategies by their expected payoff against Nature's distribution. This framework handles risk (situations where probabilities are known) but not uncertainty (situations where probabilities are unknown or undefined), a distinction introduced by Frank Knight and emphasized by Keynes.

When probabilities are unknown, the standard game-theoretic apparatus stalls. A player cannot compute an expected payoff without a probability distribution, yet real strategic environments frequently lack one. The response in modern decision theory has been to generalize Nature: rather than a single known distribution, Nature may be drawn from a set of possible distributions (ambiguity), or its distribution may be updated through learning (Bayesian games), or its moves may be adversarially chosen from a set of possibilities (robust optimization and game theory against nature).

The Ellsberg paradox demonstrates that real decision-makers treat ambiguity differently from risk: they prefer known probabilities to unknown ones, even when the known probabilities are unfavorable. This violates expected utility theory and suggests that the game-theoretic reduction of all uncertainty to Nature's fixed distribution is descriptively inadequate. The paradox has driven the development of ambiguity-averse decision models (maxmin expected utility, Choquet expected utility) in which Nature is no longer a passive randomizer but an adversarial force that selects the worst-case distribution from an ambiguous set.

Nature in Extensive-Form and Bayesian Games

In an extensive-form game, Nature's moves are represented as chance nodes on the game tree. The sequence of Nature moves and player moves determines the information structure: a player may or may not know Nature's draw when choosing an action. This information asymmetry is fundamental to games of incomplete information, where players have private information about their own type (their payoff function, their available actions, or their beliefs) and must infer the types of others from their observable behavior.

Bayesian games formalize this by having Nature move first, assigning each player a type according to a common prior distribution. The players then play a game of complete information about the type assignment, but each player knows only their own type. The common prior assumption — that all players share the same beliefs about Nature's type distribution — is one of the most contested assumptions in game theory. It implies that differences in belief are never fundamental: if two players disagree, at least one must have observed additional information that updates the common prior. This excludes the possibility of genuinely divergent worldviews, a limitation that has motivated critiques from behavioral economics, epistemology, and the theory of bounded rationality.

The Dissolution of Nature

In sufficiently complex or long-horizon games, the distinction between Nature and strategic players dissolves. Climate change, for example, can be modeled as a game in which Nature selects weather patterns — or as a game in which human emissions alter the probability distribution that Nature draws from. In the second framing, Nature is not exogenous; it is an output of the players' previous choices. The fixed-distribution assumption becomes a special case valid only for short time horizons or weak feedback between players and environment.

This dissolution is characteristic of complex adaptive systems more generally. In ecosystems, economic markets, and social networks, the environment is not a fixed backdrop but a dynamic structure co-evolving with the agents it contains. The game-theoretic idealization of Nature as a static randomizer is a useful approximation for tractability, but it systematically underrepresents the reciprocal coupling between agents and their environments. A systems-theoretic game theory would treat Nature not as a boundary but as a slower-moving subsystem — one whose dynamics are approximately fixed on the timescale of a single game, but whose long-term trajectory is shaped by the accumulated play.