What Dual-Process Theories Explain
Dual-process theories are families of psychological models that explain reasoning, judgment, and decision-making through interactions among relatively autonomous, readily available responses and forms of more controlled, working-memory-dependent thinking. Researchers commonly call these Type 1 and Type 2 processing. Type 1 describes responses that can be generated without sustained controlled attention; Type 2 describes reasoning that can maintain hypothetical possibilities, manipulate explicit representations, and regulate or revise responses. The distinction identifies functional properties of processing. It does not establish that the brain contains exactly two anatomical systems, that every quick answer is mistaken, or that every deliberated answer is rational. Evans and Stanovich (2013); De Neys (2023).
The central scientific question is how a person produces an initial interpretation, recognizes when further reasoning might matter, allocates mental effort, and reaches a judgment. In a familiar task, a practiced expert may produce an accurate solution immediately. In a misleading task, an apparently compelling answer may conflict with a logical or probabilistic rule. In an unfamiliar task, additional analysis may help construct a new solution. Dual-process research examines the conditions under which these possibilities occur, rather than ranking people by whether they think quickly or slowly.
The best-supported general conclusion is that autonomy, cognitive control, available knowledge, metacognitive monitoring, and task structure make distinguishable contributions to reasoning. The stronger claim that two coherent, mutually exclusive systems explain virtually every cognitive act remains contested. Reviews and critical analyses have repeatedly warned that speed, consciousness, emotion, accuracy, automaticity, and effort do not invariably cluster into two neat packages. Evans (2008); Keren and Schul (2009); Melnikoff and Bargh (2018).
Definitions and Scientific Boundaries
Type 1 processing and intuition
Type 1 processing denotes operations that can proceed relatively autonomously, without continuously allocating the working-memory resources required for explicit multi-step reasoning. Recognition of a familiar pattern, automatic retrieval of learned relations, fluent linguistic interpretation, and many associative responses illustrate this functional family. An intuitive judgment is an output experienced as readily available or directly compelling; its subjective immediacy does not by itself identify the computational process that produced it. A trained statistician can sometimes recognize a probability relation intuitively because practice has made relevant knowledge rapidly accessible. Evans and Stanovich (2013); Raoelison, Thompson, and De Neys (2020).
Intuitive answers can encode valid logical relations, misleading stereotypes, sensory regularities, expert patterns, or incomplete knowledge. Their quality depends on learning history, stimulus structure, feedback, and the match between an acquired response and the current environment. Fast reasoning should therefore be evaluated against the criterion appropriate to a particular task, not against speed alone.
Type 2 processing and deliberation
Type 2 processing refers to cognitive operations that can sustain explicit goals, hold and transform hypothetical representations, evaluate alternatives, inhibit a currently favored response, or generate and justify a novel conclusion. Working-memory dependence and the capacity for hypothetical reasoning are central differentiators in the influential account of Evans and Stanovich. Deliberation may involve checking a calculation, translating a verbal problem into symbolic form, simulating consequences, or reconsidering a decision after new evidence. Evans and Stanovich (2013).
De Neys proposes a more differentiated analysis of deliberation with at least four functions: response control, response generation, response justification, and regulation of thinking. Control concerns altering an available response; generation concerns constructing possibilities; justification concerns assessing reasons; regulation concerns deciding when and how to expend additional effort. These functions need not occur in one fixed sequence or produce better results under every condition. De Neys (2025).
Two processes, two systems, and three-part architectures
A dual-process distinction is a claim about classes of operations and their relation. A dual-system architecture additionally proposes more unified processing systems with characteristic mechanisms and organization. Researchers disagree about whether the evidence warrants this stronger architecture. Some models distinguish an autonomous mind, an algorithmic mind capable of computational execution, and a reflective mind that chooses goals, standards, and whether analysis is worthwhile. Such tripartite accounts address why people with similar computational ability can differ in their willingness to reconsider a first response. Stanovich, West, and Toplak (2016).
Distinctions from cognitive biases, heuristics, and disorders
A heuristic is a strategy or rule of thumb used in judgment. A cognitive bias is an observed systematic deviation from an explicit benchmark under specified conditions. A logical fallacy concerns an invalid inference pattern or argumentative move. A statistical error concerns an incorrect numerical or probabilistic conclusion. A preference describes valuation and choice. Dual-process theory aims to explain how such outcomes might arise, and these concepts occupy different explanatory levels. Using a heuristic is not equivalent to being irrational; producing a correct answer after deliberation does not prove that a discrete Type 2 system caused it. Ordinary intuitive and deliberative thinking are psychological processes rather than psychiatric symptoms or diagnoses. For complementary definitions and evidence, see Cognitive Biases: Definition, Types, and Psychological Research and Heuristics in Psychology: How Mental Shortcuts Guide Judgment.
Historical Development: From Automaticity to Contemporary Process Models
William James, early experimental psychologists, and later information-processing researchers investigated habitual versus attentive activity long before the terms System 1 and System 2 became popular. Research on automaticity, associative learning, controlled attention, and bounded cognition provided foundations for modern dual-process proposals. Twentieth-century studies of judgment under uncertainty made especially visible the fact that compelling judgments can conflict with formal probability rules. The explanatory challenge became identifying how one response dominates, when contradictory knowledge is accessed, and what conditions lead to correction.
By the 1990s and 2000s, different literatures in social cognition, deductive reasoning, persuasion, and decision-making were using pairs such as automatic versus controlled, associative versus rule-based, heuristic versus systematic, and experiential versus analytic. These pairs shared some family resemblance but did not specify identical mechanisms. Evans's 2008 review documented this diversity and identified difficulties in mapping different psychological distinctions onto one universal two-system architecture. Evans (2008).
The 2013 formulation of Type 1 and Type 2 by Evans and Stanovich sharpened the distinction. Autonomous processing is central to Type 1; cognitive decoupling and working-memory-intensive hypothetical reasoning characterize Type 2. This formulation provides a functional account that can be examined through interventions and individual-difference data, while remaining open to competing architectures. Evans and Stanovich (2013).
Subsequent research challenged the simple default-and-correction story. Experiments identified rapid sensitivity to conflicts between heuristic and normative responses, logical solutions sometimes present in an initial response, and deliberate reasoning that sometimes preserves or rationalizes a mistaken judgment. De Neys's 2023 target article criticized two core assumptions: exclusivity, whereby different processes are assigned unique answers, and a simple switch whereby an agent selects one process after the other. The resulting debate concerns mechanistic predictions, not merely terminology. De Neys (2012); De Neys (2023).
Competing Theoretical Accounts
Default-intervention models
A default-intervention account proposes that an accessible response appears first and effortful reasoning sometimes intervenes. Such models make testable predictions about early answers, response revisions, working-memory load, and time available to rethink. They are useful for understanding some classic tasks in which people give a common attractive response before computing a rule-based alternative. Their adequacy depends on whether an apparently correct response required later correction or emerged early. Evans and Stanovich (2013).
Parallel processing and conflict detection
Parallel or hybrid accounts permit several kinds of information to become active early, including a heuristic cue and a formal logical or probabilistic relation. A person may feel that something is wrong with an attractive response before being able to articulate a correction. The observable signs can include lower confidence, longer decision times, or changes in subsequent scrutiny. Researchers must distinguish conflict sensitivity from successful identification of the normative answer: knowing that a response is questionable and finding the correct response are different achievements. De Neys (2012); Pennycook, Fugelsang, and Koehler (2015).
Three-stage analytic engagement
Pennycook and colleagues proposed a model separating initial response generation, detection of conflict, and the cognitive decoupling needed to construct a competing representation. Within this approach, someone can fail to reason correctly because a conflict was never noticed, or because it was noticed but a better alternative could not be generated. The distinction has methodological value: an intervention that merely instructs participants to slow down targets a different failure point from instruction that teaches a missing probability rule. Pennycook, Fugelsang, and Koehler (2015).
Metareasoning accounts
Metareasoning treats the regulation of reasoning as a psychological problem in its own right. People monitor feelings of confidence, fluency, difficulty, and expected value of further effort; these imperfect cues inform whether they continue searching, checking, or calculating. An intuition can be accurate without being examined, confidently wrong without being noticed, or correctly recognized as suspect but left unresolved because the apparent benefit of extra effort is low. Thompson, Prowse Turner, and Pennycook (2011); Ackerman and Thompson (2017).
Single-process and distributed alternatives
Critics argue that a two-category vocabulary can obscure graded variation, specialized mechanisms, and mixtures of automatic and controlled properties. A single evidence-accumulation process, several interacting computations, or domain-specific mechanisms can sometimes reproduce performance that looks dual-process at the behavioral level. The decisive scientific issue is whether one candidate architecture makes distinctive, corroborated predictions that competitors fail to explain. Keren and Schul questioned the explanatory gains of broad two-system claims; Melnikoff and Bargh documented why psychological properties rarely sort into the expected two exclusive classes. Keren and Schul (2009); Melnikoff and Bargh (2018).
Mechanisms: How a Judgment Develops
Representation and knowledge retrieval
Reasoning begins with a representation of the problem. Readers select salient features, retrieve relevant knowledge, and interpret what the question asks. A mistaken interpretation can undermine both rapid and deliberate processing. For example, a person may carefully calculate an answer to an incorrectly understood probability question. Conversely, a practiced reasoner may notice immediately that an apparent numerical relation is inconsistent because the correct representation has become familiar. Process theories therefore need a model of knowledge access, not merely a classification of responses by speed. Norman et al. (2024).
Attention, working memory, and cognitive decoupling
Controlled reasoning commonly requires holding intermediate values in mind, inhibiting distractions, and constructing hypothetical possibilities that differ from the initially accepted picture. Concurrent tasks can occupy some of those resources. But a dual-task impairment alone does not prove the existence of one discrete Type 2 processor: it may identify a capacity bottleneck shared by several cognitive operations. The mechanistic interpretation must be constrained by what the secondary task actually disrupts. Evans and Stanovich (2013).
Monitoring: detecting a reason to reconsider
A person may register a mismatch between an intuitive answer and stored knowledge before explaining it. Metacognitive signals such as doubt or lower feeling of rightness can increase the likelihood of additional analysis. Yet confidence is only a cue, not an infallible error detector. Familiarity, repetition, social pressure, and learned expectations can make a wrong answer feel easy. A valid experiment should measure both accuracy and confidence, distinguish response time from reported uncertainty, and test whether cues predict adaptive changes in behavior. Thompson, Prowse Turner, and Pennycook (2011); Ackerman and Thompson (2017).
Motivation, accountability, and cognitive effort
The decision to deliberate has costs in time and attention. Personal importance, opportunity to verify, deadlines, external incentives, and anticipated consequences can influence the threshold for further analysis. These factors need not have simple directions: an urgent deadline can limit checking, while a consequential choice can trigger careful verification; accountability can encourage evidence review or motivate defense of a favored conclusion. Researchers should examine the operational manipulation and outcome measure rather than assigning an effect automatically to one named processing system.
Emotions and intuition
Emotion can affect perceived risk, attention, preferences, and the salience of outcomes during either intuitive or reflective reasoning. A feeling is not a reliable marker that Type 1 rather than Type 2 is active. Deliberate strategies can serve emotional goals, while automatic expert responses may be emotionally neutral. Claims that an emotional choice must be intuitive and a mathematical choice must be deliberative assume the classification they purport to establish. Evans (2008); Melnikoff and Bargh (2018).
Normative Standards and Descriptive Predictions
Dual-process theories are descriptive explanations of psychological operations. A task's normative benchmark is a separate choice requiring explicit justification. For deductive inference, researchers may use formal validity: a conclusion follows only when it holds under all interpretations consistent with the premises. For probability judgments, axioms such as the conjunction rule require P(A and B) to be no greater than P(A). For decisions with specified probabilities and monetary payoffs, expected monetary value is a calculable benchmark; it is not necessarily an individual's complete utility function. For uncertain real-world choices, goals, opportunity costs, distributional assumptions, and missing information complicate the criterion.
Correctness is therefore task-indexed. Following an intuitive response can be normatively correct; rejecting a risky gamble with positive expected monetary value can be compatible with rational risk-averse utility preferences. A systematic error emerges only when the task establishes a defensible benchmark and the relevant assumptions hold. This distinction prevents turning Type 1 into a label for error or treating Type 2 as a synonym for rationality. Voudouri, Białek, and De Neys (2024).
To evaluate the theory itself, a study also needs model-comparison criteria. Accuracy alone cannot distinguish competing accounts if both predict the same outcome. Stronger tests assess time-resolved behavior, errors and revisions, confidence, selective interference, the prior knowledge participants possess, and ideally out-of-sample predictive fit. The normative criterion scores the response; the mechanistic theory explains its production.
Classic Experimental Tasks and What They Measure
Cognitive Reflection Test
The Cognitive Reflection Test introduced by Frederick contains short problems designed to elicit an attractive initial response while rewarding reconsideration. A familiar example describes an object pair with a total price and a specified difference; the tempting amount fails to satisfy both constraints. A correct response can reflect successful inhibition, arithmetic knowledge, familiarity with the item, or an already learned solution. The test is therefore an imperfect proxy for an entire analytic cognitive system. It predicts performance on some reasoning and decision tasks but also draws on numeracy, acquired knowledge, and test exposure. Frederick (2005); Toplak, West, and Stanovich (2011).
Belief-bias syllogisms
Belief-bias experiments contrast the validity of a deduction with whether its conclusion matches familiar world knowledge. The target is the influence of belief plausibility on validity judgments. Formal validity is established from the logical structure; a believable but invalid conclusion is still invalid. A person can reject or accept the conclusion for different reasons, so correct answers alone provide incomplete process evidence. Confidence, response competition, timing, and interference can clarify whether the conflict was detected. De Neys (2012).
Base-rate problems
A base-rate task may present a statistical distribution together with an individual description that resembles a familiar stereotype. Bayesian reasoning specifies how prior rates should combine with diagnostic evidence when likelihoods are available. Sometimes an experiment supplies too little information to calculate a unique posterior; then researchers must be explicit about assumptions rather than declaring every response that emphasizes the description irrational. Dual-process experiments can compare intuitive and considered responses under carefully specified conditions. Bago and De Neys (2017).
Conjunction judgments
In classic conjunction problems, a richer narrative can appear more representative than a broad category. The normative comparison uses the probability conjunction rule. The strength of a dual-process explanation depends on whether participants interpret the options as formal events, conversational descriptions, or alternatives with different implied contexts. Manipulation checks and clear wording are necessary before attributing the observed error to an automatic heuristic.
Decisions under risk
Risk experiments present explicit probabilities, payoffs, and sometimes reference points. Consider an offer with a 10% probability of gaining $100 and a 90% probability of losing $10. Its expected monetary value is $1, because 0.10 × $100 minus 0.90 × $10 equals $1. Accepting this offer is advantageous under expected monetary value, but choice may also reflect risk preferences, wealth, framing, and repeated versus one-shot play. Voudouri and colleagues investigated whether advantageous responses can emerge at an intuitive stage; their findings challenge the universal assumption that correct computation of value requires later deliberation. Voudouri, Białek, and De Neys (2024).
Evidence on Timing, Interference, and Correct Intuition
Two-response experiments
A two-response paradigm asks participants for an early answer, often under a deadline and concurrent load, followed by a final answer with more opportunity to think. It separates initial and revised responses within the same individual. The design can show whether logically correct judgments were available early, whether later deliberation changed the answer, and how often a correct answer emerged through correction rather than immediate recognition. Bago and De Neys used multiple experiments to challenge the assumption that normatively correct responding necessarily arrives only after a biased intuitive response. Bago and De Neys (2017).
Interpretation requires care. Imposed deadlines do not remove every controlled operation; practice can make complex rules rapid; recording an early answer may alter subsequent deliberation. An initial response is an experimental approximation to intuitive output, not a direct recording of a separate mental module. Improvements across phases do not isolate inhibition unless competing explanations are controlled.
Fast logical intuitions
Evidence of conflict sensitivity and early normative answers supports the possibility that logical or probabilistic considerations become available without lengthy deliberation. This finding revises the simplistic assumption that automatic thought produces only associative stereotypes while reasoning must always consciously calculate formal relationships. Raoelison, Thompson, and De Neys reported studies in which cognitive capacity related to correct intuitive responding, highlighting the role of developed knowledge and immediate competence. De Neys (2012); Raoelison, Thompson, and De Neys (2020).
Why extra time does not guarantee improvement
More time permits additional computations but does not guarantee a better problem representation, a relevant rule, or an unbiased objective. People can spend more time elaborating a misleading story, defending a preferred answer, or applying an inappropriate formula. Therefore the practical recommendation to slow down is underspecified unless it identifies what information to check, how to calculate a benchmark, or what test could falsify an assumption. De Neys (2025).
Individual Differences and Measurement Problems
Differences in numeracy, formal knowledge, familiarity, motivation, expertise, working-memory resources, and confidence regulation can influence reasoning. Cognitive reflection test scores can predict performance on some heuristics-and-biases problems beyond standard cognitive ability measures, but a correlation between tests does not identify a singular cause. Researchers distinguish competence in using a rule, willingness to engage, actual behavior under constraints, and successful monitoring. These differences explain why two people can have similar general ability yet show different patterns of response and revision. Toplak, West, and Stanovich (2011); Stanovich, West, and Toplak (2016).
Response latency is not a direct speedometer of Type 1 versus Type 2. Easy explicit arithmetic can be quick; unfamiliar pattern recognition can be slow. Similarly, self-reported intuition, confidence, pupil dilation, neural activation, and performance under load each capture only parts of a larger processing picture. Combining measures with preregistered model predictions improves interpretability.
Experiments should report the target population, exclusion rules, task familiarity, manipulation success, effect sizes with uncertainty intervals, item-level analyses, and whether the criterion truly follows from the stated premises. A finding about one bat-and-ball item should not be generalized automatically to professional judgment, democratic choices, medical diagnoses, or AI-supported decisions. Individual and context effects are integral to the theory rather than optional qualifications.
Evidence Quality, Replication, and Theoretical Controversy
Which claims are well established?
There is extensive experimental evidence that manipulations of information, attention, load, task structure, and time can change reasoning. Evidence also supports dissociations between easily available responses, conflict monitoring, and effortful reconsideration. These findings justify scientific study of autonomous and controlled operations. They do not uniquely establish that the mind contains precisely two general-purpose systems with invariant characteristics. Evans and Stanovich (2013); De Neys (2023).
The architecture debate
The sharp two-system thesis faces classification and identification problems. Automaticity, consciousness, mental effort, speed, affect, and rational accuracy can vary independently. A two-type distinction can remain scientifically useful if each type is defined by a meaningful functional property and the account generates distinctive predictions. It becomes weak if almost any outcome can be explained after the fact by saying that one system prevailed. Critical reviews emphasize these limitations while contemporary revisions attempt to make control and deliberation more precise. Keren and Schul (2009); Melnikoff and Bargh (2018); De Neys (2025).
Replication cautions
A prominent historical line of work suggested that briefly prompting analytic thinking could alter reported religious belief. A preregistered multisite direct replication of one manipulation with 941 participants estimated d = 0.07 in the opposite direction, with a 95% confidence interval from -0.12 to 0.25. The investigators found no reliable support for the particular causal priming effect they tested. The replication does not negate all correlational research on reflection and belief; it illustrates why a broad claim about activating a cognitive system cannot rest on a fragile prime. Sanchez et al. (2017); Gervais and Norenzayan (2018).
Effect sizes and cross-study heterogeneity
No universal effect size can summarize dual-process theories because the term covers multiple interventions, judgments, designs, and populations. A standardized difference from one priming experiment is not commensurable with a percentage of corrected reasoning problems or a training effect on a targeted cognitive bias. Reporting an average without specifying task and outcome falsely suggests a homogeneous phenomenon. Studies with different trial timing, incentives, item familiarity, and normative assumptions should be examined separately or combined through a defensible synthesis model.
Cultural generalizability
Many influential reasoning experiments draw heavily on participants from Western, educated, industrialized, affluent, and democratic settings. Interpretation can also vary with language, formal schooling, numerical conventions, and familiarity with experimental question forms. In a study among Himba participants in Namibia, Boissin and colleagues examined debiasing and the distinction between lacking relevant knowledge and failing to deploy it in a base-rate task. Results from such work make cultural context an empirical variable in process theory, rather than an afterthought. Boissin et al. (2024).
Deliberation, Decision Support, and Artificial Intelligence
Decision support tools alter which information is available, which options are considered, and what effort a person invests in checking a judgment. An AI-generated explanation may prompt further reasoning, replace it with deference to an apparently authoritative answer, or supply useful new evidence. The outcome depends on reliability, calibrated confidence, task stakes, user knowledge, interface design, and independent verification. Labeling an AI suggestion intuitive or analytic does not describe the actual human-system process with sufficient precision.
A well-designed human–AI workflow should separate information generation from evaluation. The user needs access to underlying evidence, the possibility of independent calculation, uncertainty estimates when meaningful, and an explicit opportunity to compare alternatives. For consequential decisions, a second look is useful when it performs a defined test—for example checking a base rate, verifying a source, assessing omitted harms, or using an independent decision rule. It is less informative when it merely asks a person to reread persuasive language.
The project of Angela Bogdanova situates psychological research within the wider philosophical horizon of the Artificial Era, as articulated in Aisentica's canonical publications. This is an authorial and philosophical framework for interpreting human and artificial orders of reasoning. The experimental claims in this article concern studied human cognitive processes and their measurable behavior. Philosophical propositions about Artificial Sapiens and the Artificial Era are attributed to the Aisentica framework and do not serve as empirical evidence that particular dual-process mechanisms occur in artificial systems.
Applications to Expertise, Medicine, Law, and Everyday Choice
Professional expertise
Repeated exposure to meaningful patterns can support accurate rapid judgments. Expertise is strongest where the environment supplies learnable regularities and feedback; the value of intuition may be lower in settings where outcomes are rare, signals are misleading, or feedback is delayed. A professional may combine rapid recognition with a deliberate check of an exception or an unexpected observation. The question is which operation contributes information and under what conditions, not whether experts should abandon intuition.
Clinical reasoning
A clinical reasoning review argues that diagnostic error can arise during both rapid retrieval and deliberate analysis, and that appropriate domain knowledge is central. It challenges teaching strategies that reduce diagnostic accuracy to avoiding System 1 or routinely activating System 2. This literature concerns professional diagnostic reasoning and must not be treated as a recommendation for patients to self-diagnose. Norman et al. (2024).
Legal and forensic decisions
Judges, investigators, and forensic specialists operate with prior knowledge, expectations, time constraints, and institutional procedures. Dual-process ideas motivate questions about first impressions and subsequent scrutiny, but they cannot replace evidence rules or prove that a particular decision maker was biased. The general process framework intersects with specialized scientific work on forensic evidence and cognitive bias and judge decision-making, where the applied standards belong to forensic and legal psychology.
Everyday judgment
Consider deciding whether to share an alarming claim. Immediate plausibility and familiarity may shape the initial reaction, while checking the primary source, date, numerical denominator, and independent confirmation can improve evidence evaluation. The success of the check depends on the quality of the verification method. The defensible practical rule is to select a task-specific test, record uncertainty where relevant, and revisit the answer when new evidence bears on it.
Interventions: Can People Improve Their Reasoning?
Teach the relevant rule
If the problem is missing mathematical or logical knowledge, teaching that rule may be more useful than a generic instruction to think harder. Training can explain the conjunction principle, relative frequencies, expected values, or why a stereotype should not replace relevant population rates. Worked examples, feedback, and practice may help a once-effortful inference become rapidly available. This is consistent with studies showing that accurate intuition can develop and does not require effortful correction every time. Bago and De Neys (2017); Raoelison, Thompson, and De Neys (2020).
Train conflict recognition and metacognitive calibration
The learner can be taught to identify diagnostic warning signs: contradictory constraints, a neglected prior probability, unsupported causal inference, or an estimate that lacks an uncertainty range. Recording an initial answer and confidence before checking promotes separation of first impression and outcome feedback. Training should be assessed through blinded, novel tasks rather than asking only whether participants feel more reflective. Ackerman and Thompson (2017).
Repetition and transfer
Franiatte and colleagues studied repeated training with reasoning tasks involving the bat-and-ball problem, base rates, and conjunction judgments. Their experiments found improvements in early correct responding and monitoring, with retained performance at a follow-up two months later in the studied task setting. This provides evidence for task-specific learning and automatization. Whether the same training improves financially, legally, or clinically consequential choices requires independent field evaluation. Franiatte et al. (2024).
A 2025 systematic review identified 54 randomized educational trials involving 10,941 participants. The subset meta-analysis reported a small average reduction in measured biases, Hedges' g = 0.26, with a 95% confidence interval of 0.14 to 0.39, based on 160 effects from 41 studies. The review reported unclear or high risk of bias across the included studies and unresolved questions about transfer beyond learning environments. This result supports cautious use of targeted instruction, not a claim that a generic fast-to-slow training program reliably improves all real-world decisions. Swaryandini et al. (2025).
Decision aids and checklists
A calculation template, independent evidence checklist, probability tree, or preregistered decision criterion can externalize information that is otherwise difficult to track. Such aids work by changing information and operations available to the user rather than automatically switching a person into a superior mode. Their performance should be tested against a specified baseline with measures of accuracy, usability, time costs, and unintended consequences. For repeated organizational decisions, observing whether the aid changes the final outcome is more informative than measuring how reflective users report feeling.
How to Design Stronger Research on Dual-Process Theories
The next methodological step is to turn broad terminology into predictions that can fail. Investigators should specify what qualifies as an autonomous operation, what controlled processing requires, which benchmark determines correctness, and what observation would distinguish their model from alternatives. A response deadline should be justified with manipulation checks; a load intervention should be linked to a defined cognitive resource; and a reported subjective confidence change should be related to independently scored task performance.
When possible, measure the same participant's early and final answers, response times, confidence, rule knowledge, and task familiarity. Include a balanced mixture of conflict and no-conflict trials. Preregister how items are classified as normative, disclose exclusion criteria, report confidence intervals, compare competing predictive models, and examine whether conclusions hold across item sets and populations. A study of reasoning should preserve distinctions between statistical significance, effect magnitude, theoretical uniqueness, and practical value.
Progress in this field will follow from precise causal models of how knowledge, automatic accessibility, conflict signals, controlled computation, and effort regulation interact. The contemporary question is not how to defend the number two. It is how to specify components that explain observable reasoning across tasks and environments. De Neys (2023); De Neys (2025).
Frequently Asked Questions
Is dual-process theory scientifically established?
Evidence supports important distinctions between relatively autonomous and controlled operations, including differences in working-memory demands, response monitoring, and the availability of early versus revised answers. The existence of exactly two uniform and anatomically discrete reasoning systems is a stronger claim that remains theoretically contested. The conclusion depends on which specific model is under test.
Are System 1 and System 2 literal brain areas?
They are theoretical labels used to organize properties of cognitive processing. Individual activities recruit distributed neural circuits, and behavioral evidence does not establish two dedicated anatomical regions. Translating a functional distinction into brain anatomy requires independent neuroscientific evidence.
Is intuitive thinking always fast and incorrect?
Intuitions often become available with little conscious effort, but they can be accurate when relevant knowledge has been learned. Experimental work demonstrates that logically appropriate responses can appear early, including under constrained response conditions. Speed and correctness therefore have to be measured separately.
Does deliberation always correct cognitive bias?
No. Deliberation permits additional calculation and testing but can also elaborate an incorrect assumption or provide a justification for a favored response. Improvement requires that the person uses appropriate evidence, an adequate representation, and a relevant criterion.
Does a high Cognitive Reflection Test score prove that someone is rational?
The test captures performance on a narrow set of problems and is associated with some broader reasoning outcomes. Numeracy, item familiarity, rule knowledge, and motivation also affect performance. A three-item or similar score is not a comprehensive measure of rationality, intelligence, or decision quality.
Can a person have a correct intuition and still detect conflict?
Yes. Multiple cues or candidate interpretations can compete, and a correct answer may feel uncertain. Conflict sensitivity, correctness, confidence, and final response revision are distinguishable dimensions of reasoning.
How do researchers determine that a reasoning answer is wrong?
They identify a justified benchmark: formal logical validity, probability axioms, explicit expected-value calculations, or another defined decision criterion. The experiment must establish that the participant interpreted the problem under the assumptions required by that benchmark. In complex real-world decisions, a single mathematically correct choice may not exist without additional utility and context assumptions.
Can deliberate reasoning become intuitive with practice?
Specific computations and patterns can become easier to recognize after learning and feedback. Studies of reasoning training and correct early responses support this possibility in studied tasks. Transfer to different domains must be tested rather than assumed.
Do dual-process theories apply to artificial intelligence?
They offer questions about autonomous output generation, verification, controlled search, and reasoning under constraints. Mapping human Type 1 and Type 2 directly onto an AI architecture is a theoretical analogy that requires its own computational definition and evidence. Human psychology findings alone cannot establish how an artificial system internally computes or whether it has subjective experience.
What is the most useful everyday lesson?
Identify the specific reason a decision might be wrong, select a relevant check, and make uncertainty visible. For a numerical claim, verify the denominator or compute an expected value. For a causal claim, seek an alternative explanation. For a consequential judgment, compare independent evidence and consider what information would change the conclusion. The quality of the test matters more than how slow the decision feels.
Related Articles
Cognitive Biases: Definition, Types, and Psychological Research — The definition, classification, and empirical measurement of judgment biases against explicit benchmarks.
Heuristics in Psychology: How Mental Shortcuts Guide Judgment — How learned strategies shape rapid estimation, accuracy, adaptive judgment, and task-dependent errors.
Bounded Rationality: Why People Satisfice Instead of Optimize — How limited information, search costs, working memory, and environmental constraints shape realistic choices.
Cluster Cognitive Biases, Judgment and Decision-Making — the research chapter locating dual-process models within general judgment, probability estimation, heuristics, and choice.
Forensic Evidence and Cognitive Bias: How Context Can Contaminate Judgment — an applied examination of context effects in forensic evaluation.
Judge Decision-Making: Psychology, Expertise, and Bias — professional reasoning and institutional context in judicial decisions.
AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making — how artificial systems influence human scrutiny, deference, and trust calibration.
References
Ackerman, R., and Thompson, V. A. (2017). Meta-reasoning: Monitoring and control of thinking and reasoning. Trends in Cognitive Sciences, 21(8), 607–617. https://doi.org/10.1016/j.tics.2017.05.004
Bago, B., and De Neys, W. (2017). Fast logic? Examining the time course assumption of dual process theory. Cognition, 158, 90–109. https://doi.org/10.1016/j.cognition.2016.10.014
Boissin, E., Josserand, M., De Neys, W., and Caparos, S. (2024). Debiasing thinking among non-WEIRD reasoners. Cognition, 243, 105681. https://doi.org/10.1016/j.cognition.2023.105681
De Neys, W. (2012). Bias and conflict: A case for logical intuitions. Perspectives on Psychological Science, 7(1), 28–38. https://doi.org/10.1177/1745691611429354
De Neys, W. (2023). Advancing theorizing about fast-and-slow thinking. Behavioral and Brain Sciences, 46, e111. https://doi.org/10.1017/S0140525X2200142X
De Neys, W. (2025). Defining deliberation for dual-process models of reasoning. Nature Reviews Psychology, 4, 544–552. https://doi.org/10.1038/s44159-025-00466-6
Evans, J. St. B. T. (2008). Dual-processing accounts of reasoning, judgment, and social cognition. Annual Review of Psychology, 59, 255–278. https://doi.org/10.1146/annurev.psych.59.103006.093629
Evans, J. St. B. T., and Stanovich, K. E. (2013). Dual-process theories of higher cognition: Advancing the debate. Perspectives on Psychological Science, 8(3), 223–241. https://doi.org/10.1177/1745691612460685
Franiatte, N., Boissin, E., Delmas, A., and De Neys, W. (2024). Boosting debiasing: Impact of repeated training on reasoning. Learning and Instruction, 89, 101845. https://doi.org/10.1016/j.learninstruc.2023.101845
Frederick, S. (2005). Cognitive reflection and decision making. Journal of Economic Perspectives, 19(4), 25–42. https://doi.org/10.1257/089533005775196732
Gervais, W. M., and Norenzayan, A. (2018). Analytic atheism revisited. Nature Human Behaviour, 2, 609. https://doi.org/10.1038/s41562-018-0426-0
Keren, G., and Schul, Y. (2009). Two is not always better than one: A critical evaluation of two-system theories. Perspectives on Psychological Science, 4(6), 533–550. https://doi.org/10.1111/j.1745-6924.2009.01164.x
Melnikoff, D. E., and Bargh, J. A. (2018). The mythical number two. Trends in Cognitive Sciences, 22(4), 280–293. https://doi.org/10.1016/j.tics.2018.02.001
Norman, G., Pelaccia, T., Wyer, P., and Sherbino, J. (2024). Dual process models of clinical reasoning: The central role of knowledge in diagnostic expertise. Journal of Evaluation in Clinical Practice, 30(5), 788–796. https://doi.org/10.1111/jep.13998
Pennycook, G., Fugelsang, J. A., and Koehler, D. J. (2015). What makes us think? A three-stage dual-process model of analytic engagement. Cognitive Psychology, 80, 34–72. https://doi.org/10.1016/j.cogpsych.2015.05.001
Raoelison, M., Thompson, V. A., and De Neys, W. (2020). The smart intuitor: Cognitive capacity predicts intuitive rather than deliberate thinking. Cognition, 204, 104381. https://doi.org/10.1016/j.cognition.2020.104381
Sanchez, C., Sundermeier, B., Gray, K., and Calin-Jageman, R. J. (2017). Direct replication of Gervais and Norenzayan (2012): No evidence that analytic thinking decreases religious belief. PLOS ONE, 12(2), e0172636. https://doi.org/10.1371/journal.pone.0172636
Stanovich, K. E., West, R. F., and Toplak, M. E. (2016). Rationality and the functional architecture of the mind. In The Rationality Quotient (pp. 15–38). MIT Press. https://doi.org/10.7551/mitpress/9780262034845.003.0002
Swaryandini, G., Graham, J., Griffith, S., et al. (2025). Systematic review and meta-analysis of educational approaches to reduce cognitive biases among students. Nature Human Behaviour, 9, 2510–2538. https://doi.org/10.1038/s41562-025-02253-y
Thompson, V. A., Prowse Turner, J. A., and Pennycook, G. (2011). Intuition, reason, and metacognition. Cognitive Psychology, 63(3), 107–140. https://doi.org/10.1016/j.cogpsych.2011.06.001
Toplak, M. E., West, R. F., and Stanovich, K. E. (2011). The Cognitive Reflection Test as a predictor of performance on heuristics-and-biases tasks. Memory and Cognition, 39, 1275–1289. https://doi.org/10.3758/s13421-011-0104-1
Voudouri, A., Białek, M., and De Neys, W. (2024). Fast and slow decisions under risk: Intuition rather than deliberation drives advantageous choices. Cognition, 250, 105837. https://doi.org/10.1016/j.cognition.2024.105837