Game Theory

Imagine navigating a thick fog near the edge of a cliff. You must decide whether to step forward or stay put, but you have no competitor trying to push you, nor an opponent trying to trick you. Nature is entirely indifferent to your survival. This is the domain of **standard decision theory**: the formal study of how an isolated, rational agent *should* make choices when facing uncertainty, without the complications of strategic human conflict. While game theory models an interactive chess match against an intelligent opponent, standard decision theory models a solitary game of solitaire played against probability. ## The Triad of Choice Formalized by Leonard J. Savage in his foundational 1954 work, [*The Foundations of Statistics*](https://en.wikipedia.org/wiki/The_Foundations_of_Statistics), classical decision theory assumes that any choice can be broken down into three distinct components: 1. **Acts**: The options available to the agent (e.g., carrying an umbrella). 2. **States**: The possible conditions of the world, which are outside the agent's control (e.g., rain or shine). 3. **Outcomes**: The consequences that result from a specific act paired with a specific state (e.g., staying dry while carrying a heavy burden). Because the agent does not know which state will materialize, they must assign a **subjective probability** to each state and a numerical value—known as **utility**—to each outcome. The normative rule of decision theory is deceptively simple: always choose the act that maximizes **Expected Utility**. > "If a person is faced with a choice between two actions... he will choose that action whose expected utility is the greater." > — Leonard J. Savage, *The Foundations of Statistics* (1954) ## The Paradox of Human Irrationality Standard decision theory is strictly **normative**—it describes how a perfectly logical entity *ought* to act. However, when applied as a **descriptive** model of actual human behavior, the framework shatters. Consider the famous [Allais Paradox](https://en.wikipedia.org/wiki/Allais_paradox), formulated by Maurice Allais in 1953. When offered a choice between a guaranteed $1 million or an 89% chance at $1 million alongside a 10% chance at $5 million, most people choose the certainty. Yet, when the probabilities are shifted slightly across a second pair of choices, those same individuals violate the foundational mathematical axioms of expected utility theory. This tension between normative elegance and real-world behavior led Daniel Kahneman and Amos Tversky to develop [Prospect Theory](https://plato.stanford.edu/entries/behavioral-economics/#ProsTheo). They demonstrated that humans do not evaluate outcomes in absolute utility; instead, we evaluate them relative to a neutral reference point, fearing losses far more than we value equivalent gains. ## Parametric vs. Strategic Uncertainty The fundamental dividing line between standard decision theory and game theory lies in the nature of the uncertainty involved: - **Parametric Uncertainty (Decision Theory)**: The world is passive. The probabilities of states (like the chance of rain) may be unknown or difficult to estimate, but nature is not actively adjusting its parameters to defeat you. - **Strategic Uncertainty (Game Theory)**: The environment reacts. The outcome depends on other conscious agents who are actively predicting your move to optimize their own results. Standard decision theory provides the baseline calculus for individual choice. It powers modern artificial intelligence risk assessment, medical diagnostics, and automated financial trading—domains where an agent must optimize decisions against a chaotic, yet indifferent, universe.