Bounded rationality and dual-process thinking

Working Definition

Bounded rationality is the principle that real cognitive agents—humans, institutions, and machines—never optimize decisions under complete knowledge, but make choices within the constraints of available time, incomplete information, and finite computational power by searching for solutions that are “good enough” (satisficing). Dual-process theory, popularized as the division between System 1 and System 2, describes the internal cognitive architecture that implements this adaptation: the interplay between a fast, automatic, associative mode of intuition (System 1) and a slow, serial, effortful mode of analytical deliberation (System 2). In popular usage, the two systems are often treated as two separate brain regions (“emotion versus reason”); in technical usage, they are two functional families of information-processing operations in which most judgments arise by automatically substituting an easier question for a harder one.

Genesis and Global Lineage

American economist, political scientist, and artificial intelligence pioneer Herbert A. Simon introduced bounded rationality in his 1955 and 1956 articles and his book Models of Man (1957), challenging the neoclassical model of homo oeconomicus. Simon coined the portmanteau satisficing (blending satisfy and suffice): rather than computing every branch of a decision tree, a human sets an aspiration threshold and selects the first encountered option that meets it. Two decades later, Israeli psychologists Amos Tversky and Daniel Kahneman, in a series of studies culminating in their 1974 Science paper (Judgment under Uncertainty: Heuristics and Biases) and their 1979 prospect theory, transformed Simon's framework into a map of systematic cognitive biases. In parallel, dual-process models matured in the psychology of reasoning and attention (Peter Wason, Jonathan Evans, Steven Sloman), receiving the neutral labels “System 1” and “System 2” from Keith Stanovich and Richard West in 2000. In 2011, Kahneman united both traditions in his bestseller Thinking, Fast and Slow, making the tension between fast heuristics and lazy analytical control the dominant language of contemporary behavioral economics.

Central European Lineage

Both the modern concept of a heuristic in cognitive science and the environmental understanding of bounded rationality have documented roots in interwar Budapest and Vienna. The modern scientific career of the term “heuristic”—understood as a practical, fallible rule of thumb that narrows a search space when an exhaustive algorithm is unavailable—was launched by Hungarian mathematician György (George) Pólya, educated at the universities of Budapest and Vienna, in his 1945 book How to Solve It. Herbert Simon, who knew Pólya personally at Stanford and regarded his book as a breakthrough in understanding human problem-solving, adopted the concept for the earliest AI programs (Logic Theorist, 1956), and through Simon's work it entered the vocabulary of Tversky and Kahneman. Meanwhile, Simon's second key source—the metaphor of “scissors” whose two blades are the computational limits of the mind and the probabilistic structure of the environment—grew out of the work of Viennese psychologist Egon Brunswik. A student of Karl Bühler at the University of Vienna and a participant in the Vienna Circle, Brunswik formulated “probabilistic functionalism” in Vienna in 1934 (Wahrnehmung und Gegenstandswelt), which Simon cited in his 1956 environmental paper and on which Gerd Gigerenzer later built the school of ecological rationality. Finally, the very benchmark of “unbounded rationality” against which Simon and Kahneman rebelled had been formalized by Budapest-born John von Neumann and Vienna-trained Oskar Morgenstern in Theory of Games and Economic Behavior (1944).

Why Now

Between 2018 and 2026, the pairing of bounded rationality and dual-process thinking reached a turning point for two reasons. First, in the social sciences, the years 2020–2024 brought a replication crisis in parts of popular behavioral economics (including high-profile retractions of dishonesty studies in 2023 and meta-analyses showing near-zero population-scale effects of minor nudges), while Daniel Kahneman's death in March 2024 prompted researchers to separate the durable foundations of the Tversky–Kahneman program from pop-psychological simplifications. Second, between 2023 and 2026, the distinction between System 1 and System 2 became the central architectural framework in artificial intelligence (popularized by Yoshua Bengio and Andrej Karpathy, among others). A classical large language model generating the next token in the fixed time of a single forward pass corresponds to associative System 1; by contrast, the breakthrough of reasoning models in 2024–2026 (such as OpenAI's o1/o3 series, DeepSeek-R1, and extended-thinking Claude), which allocate a variable budget of test-time compute to an internal chain of verification and search, represents an engineering realization of System 2.

Explanatory Power

This framework explains a phenomenon that traditional theories of education and management find baffling: why people with very high IQs and vast domain expertise so often make disastrous investment, medical, or political decisions. As Keith Stanovich demonstrated by introducing the concept of “dysrationalia,” high algorithmic computing power in System 2 in no way guarantees that System 2 will actually be engaged to audit a first impulse. On the contrary, because sustaining analytical attention is metabolically and cognitively expensive, the human mind operates as a “cognitive miser.” When a highly intelligent person falls prey to confirmation bias or framing, they use their above-average analytical skill not to refute System 1's flawed intuition, but to construct an exceptionally sophisticated defensive wall around it. Understanding this mechanism shifts the burden of reform from appealing to individuals to “be more mindful” toward designing institutional procedures—surgical checklists, boardroom pre-mortems, and blind peer review—that force System 2 participation from the outside.

Axes of Debate

Underappreciated Angle

Popular reception of Thinking, Fast and Slow merged the vision of Herbert Simon and the vision of Daniel Kahneman into a single story about “human irrationality,” obscuring a deep philosophical rift between the two Nobel laureates. For Simon, bounded rationality was not evidence that humans are flawed or foolish—it was evidence that the classical model of full optimization is computationally impossible in a universe of finite resources (what computer scientists later formalized as undecidability or NP-hard complexity). From Simon's standpoint, an agent confronted with a combinatorial problem who abandons the search for a global optimum and applies a simple satisficing threshold is behaving in the only rational way permitted by the physics of information. When AI engineers today design real-time decision systems that dynamically halt deep tree search (System 2) once the computational cost of the next step exceeds the expected gain in accuracy, they are not “imitating human frailty”—they are rediscovering Simon's fundamental law of computational economics.