Two minds in one head. In 2011, Daniel Kahneman — Nobel laureate, founder with Amos Tversky of the heuristics-and-biases programme — published Thinking, Fast and Slow. Human cognition has two systems: System 1 is fast, automatic, parallel, effortless, associative, hard to modify; System 2 is slow, deliberate, serial, effortful, rule-following, capacity-limited. The biases Kahneman and Tversky catalogued are System 1 doing its work, uncorrected by a lazy System 2. The dichotomy was not original to Kahneman — Keith Stanovich, Jonathan Evans, and Steve Sloman had developed dual-process theories since the 1990s — but Kahneman's framing made it the dominant folk theory of cognition of the 2010s.
Dual-process theories posit two qualitatively different modes of cognitive processing in the same mind. System 1 (Type 1) is autonomous — running without intentional engagement and impossible to switch off — fast, parallel, low-load on working memory, evolutionarily older, implicit; examples include recognising a face, understanding a sentence, driving a familiar route, forming a first impression. System 1 generates most of the contents of consciousness. System 2 is deliberate, serial, capacity-limited (bottlenecked by working memory), slow, effortful, evolutionarily newer, explicit; examples include multiplying 17 × 24, parking in a tight spot, evaluating a logical argument, resisting an impulse. System 2 monitors and can override System 1 but is metabolically and attentionally expensive, so the default disposition is to let System 1 outputs stand. The famous biases (anchoring, availability, representativeness, substitution — replacing a hard question with an easier one) are System 1 pattern-matching that is usually approximately right but systematically wrong in specific cases. Stanovich's cognitive miser picture is that people are not stupid but cognitively miserly, defaulting to System 1 because System 2 is expensive. The framework is not universally accepted — single-process accounts argue the distinction is a continuum mischaracterised as a dichotomy; Evans and Stanovich (2013) tightened the defining criteria to autonomy and working-memory load alone. Neural correlates fit roughly but imperfectly: prefrontal cortex (especially dorsolateral PFC) supports System-2-like deliberation, working memory, and conflict resolution; striatal and amygdalar circuits support System-1-like automaticity, habit, and emotional reaction.
Dual-process theory has escaped psychology and become a general-purpose cultural framework. Behavioral economics (Thaler, Sunstein, the Nudge programme) designs choice architecture to either engage System 2 (cooling-off periods, mandatory disclosures) or enlist System 1 (default options, salient prompts, frictionless paths). AI alignment discussions increasingly invoke the framing for what is missing in current LLMs — fast pattern-matching is what they do well, while slow deliberate multi-step reasoning is what chain-of-thought, tree-of-thought, and reasoning models (OpenAI o1/o3, Anthropic's extended thinking, DeepSeek-R1) explicitly try to add. Rationalist communities treat System-2 calibration — debiasing, base-rate reasoning, Bayesian updating — as a core practice. The replication crisis has dented some specific claims, but the broad architecture has held up.