Foundations: The Philosophical and Mathematical Underpinnings of Bounded Rationality

The edifice of neoclassical economics rests upon a singular, elegant assumption: that of *Homo Economicus*, a being of unbounded rationality who maximizes utility under constraint with flawless computational prowess. This agent, equipped with stable preferences and perfect probabilistic reasoning, navigates the decision landscape via the axiomatic compass of expected utility theory. For decades, this framework provided a mathematically pristine foundation for microeconomics, finance, and social choice theory. Yet, its descriptive validity has long been besieged by a quiet insurgency from experimental psychology and cognitive science. The systematic, predictable departures from its predictions are not mere noise; they are the signature of a different computational architecture—one built not on optimization but on *satisficing*, not on exhaustive calculation but on *heuristics*, not on logical consistency but on *bounded rationality*. This chapter establishes the formal and philosophical rupture with the *Homo Economicus* paradigm. It critically contrasts Herbert Simon’s computational constraint thesis with Daniel Kahneman and Amos Tversky’s heuristics-and-biases program, two genealogies that converged on the same destination via distinct routes. We then introduce the formal mathematical language used to model these adaptive tools: fast-and-frugal trees, sequential sampling models, and the economics of information processing costs. Finally, we derive the expected utility framework in its most rigorous form—the Savagean or von Neumann-Morgenstern axiomatic system—not as a prescriptive ideal but as the essential *null hypothesis* against which all observed biases are measured. A technical appendix integrates the modern understanding of probability weighting functions, from the original inverse-S-shape of Tversky and Kahneman to the elegant one-parameter functional form of Prelec, and demonstrates their incorporation into cumulative prospect theory, the most successful descriptive model of decision under risk.

1.1 The Axiomatic Fortress: Von Neumann-Morgenstern and the Birth of Expected Utility

The expected utility hypothesis did not emerge fully formed; it was axiomatized. The 1944 treatise by von Neumann and Morgenstern provided the deductive skeleton, proving that any rational agent whose preferences satisfy four plausible-seeming axioms must possess a utility function such that choices maximize its mathematical expectation. The axioms—completeness, transitivity, continuity, and independence—are deceptively simple. Independence, the most powerful and controversial, states that if an agent prefers a lottery $L_1$ to $L_2$, they must also prefer a compound lottery $pL_1 + (1-p)L_3$ to $pL_2 + (1-p)L_3$ for any third lottery $L_3$ and any probability $0 < p < 1$. This axiom effectively rules out certain kinds of risk attitudes, forcing them into the multiplicative scaling of the utility function.

```mermaid graph TD A[Decision under Risk] --> B[Von Neumann-Morgenstern Axioms] B --> C1[Completeness] B --> C2[Transitivity] B --> C3[Continuity] B --> C4[Independence] C4 --> D[Expected Utility Representation Theorem] D --> E[EU = Σ p_i u(x_i)] E --> F[Optimization: Choose act with max EU] ```

The elegance of this system is its logical inevitability. Accept the axioms, and the utility representation follows. The agent becomes a pure optimizer, a solipsistic calculator of discounted outcomes. The *Homo Economicus* is thus born not as an empirical claim but as a logical consequence of a particular definition of rationality—one that prizes internal coherence above all else. This definition would reign supreme for a generation, providing the bedrock for general equilibrium theory, the capital asset pricing model (CAPM), and the efficient market hypothesis. Its power lies in its parsimony and its ability to generate sharp, testable predictions about market-level outcomes, even if individual behavior often strayed.

1.2 Empirical Cracks: Allais, Ellsberg, and the First Breaches

The first major empirical fissures appeared not in the marketplace but in the laboratory. Maurice Allais (1953) presented his famous paradox, a pair of choice problems that starkly violated the independence axiom. In the first problem, most people prefer a certain $1 million (1A) over a 10% chance at $5 million and a 89% chance at $1 million (1B). In the second, they prefer a 11% chance at $1 million (2B) over a 10% chance at $5 million and a 1% chance at nothing (2A). Logically, 1A and 2A, and 1B and 2B, share identical outcomes in the common consequence ($1 million at 89% vs. 1%). A true expected utility maximizer, whose utility for $1 million is $u(1)$, should treat the common consequence as a constant that cancels out, making the choice between 1A/1B and 2A/2B identical. The systematic reversal—preferring certainty in the first frame but the higher probability in the second—demonstrates that people are sensitive to how a sure outcome is *framed*, violating independence. This is the Allais paradox.

Daniel Ellsberg (1961) subsequently demonstrated violations of another cornerstone: subjective expected utility (SEU), which extends the model to ambiguous probabilities. In his urn paradox, an urn contains 90 balls: 30 are red, and the remaining 60 are either black or yellow, with unknown proportions. Subjects prefer to bet on “red” (known 1/3 probability) over “black” (unknown probability between 0 and 2/3) in one task, but in a second task, prefer betting on “yellow” (unknown) over “red” (known). This ambiguity aversion—a preference for known risks over unknown risks—cannot be accommodated within SEU, which requires probabilities to be precise, single numbers. The axiom of *dynamic consistency*, which underpins intertemporal choice and planning, also faced experimental refutation.

These paradoxes were not mere curiosities. They revealed that human judgment under risk and ambiguity systematically deviates from the axiomatic model in patterned ways. The response from the economic orthodoxy was often to dismiss the experiments as “small stakes” artifacts or to argue that the axioms defined *rationality*, so any violation was simply a mistake, not a challenge to the model’s normative force. This is where Kahneman and Tversky entered, with a radically different program: not to patch the axioms, but to *replace* the underlying model of the decision-maker.

1.3 Two Genealogies of Bounded Rationality: Simon’s Scissors vs. Kahneman-Tversky’s Architecture

The concept of bounded rationality has two primary intellectual ancestors, whose insights, while compatible, stemmed from different disciplinary preoccupations and employed different methodological toolkits.

Herbert A. Simon (1955, 1956) approached the problem from the perspective of *administrative behavior* and *computer science*. His central metaphor was the scissors: one blade represents the cognitive limitations of the human mind (its finite processing speed, memory, and attention), the other blade represents the structure of the external environment. Rationality is *bounded* because it is shaped by the interaction of these two constraints. Simon’s key contribution was the concept of satisficing—searching for an option that is “good enough” rather than optimal. He argued that in complex, real-world environments, the computational cost of identifying the true optimum is often astronomically high, exceeding any conceivable cognitive budget. Therefore, a rational agent employs *procedures* (e.g., sequential search with an aspiration level) that are adapted to the environment and yield satisfactory outcomes with manageable effort. Simon’s work was profoundly influential in organizational theory and artificial intelligence, emphasizing *procedural rationality*—the rationality of the decision-making *process*—over the substantive rationality of the outcome. His models were often