The Mythical Number Two

Every behavioural training programme a Canadian business owner is likely to encounter rests on the same architecture: a fast, automatic, error-prone system and a slow, deliberate, reliable one. The instruction that follows is always the same, which is to engage the second. This article is about what the cognitive science literature has been saying about that architecture, in public, in its own journals, for well over a decade.

The Verdict, Stated First

Five claims, in descending order of confidence, because this article is long.

One. The version of System 1 and System 2 taught in business settings is not defended by anyone in this dispute. The critics say the two-type framework is a myth. The defenders say the features it is built from were never definitional. The popular model uses exactly those features as the definition, which means it is the thing the critics attacked and not the thing the defenders defended.

Two. The critics' central claim is countable, and we counted it. The typology sorts processes on four binary features and asserts they align into two clusters. If the features were independent, only 12.5 percent of processes would be pure types and 87.5 percent would be mixtures the framework does not name. For pure types to dominate, the features would need to correlate with one another at around 0.8. The critics say those correlations have never been demonstrated.

Three. The defence is technically strong and rhetorically expensive. Saying that intentionality, controllability, consciousness and efficiency were incidental correlates rather than defining features rescues the theory and simultaneously concedes that almost everything a lay reader believes about it is not part of it.

Four. Even "fast and slow" is contested in the primary literature. Two papers cited within the exchange are titled to challenge the time-course assumption directly, one asking whether logic can be fast and one asserting that rule-based reasoning is fast while belief-based reasoning can be slow.

Five. Institutional policy was built on this. The critics record that the World Bank and the Institute of Medicine issued reports in 2015 endorsing the distinction and urging decision makers and medical practitioners around the globe to rely on Type 2 thinking. That is the fact that moves this from an academic quarrel to something a professional should know about.

Our Grades For These Claims

Applying the scheme from the first article in this series.

Grade A that a serious, public, unresolved dispute exists. A target article, at least three commentaries and a reply, all in the same journal within months, and we obtained the abstract or opening of each.

Grade A for the critics' four-part argument as stated, which we obtained verbatim in the authors' own summary.

Grade A for the defenders' distinction between defining features and incidental correlates, obtained verbatim from the reply's own glossary.

Grade B for the institutional endorsements, which come from the critics' highlights rather than from the reports themselves, which we did not obtain.

Grade C for the challenges to the fast-slow assumption, where we have titles and citations and none of the papers.

Ungraded for whether dual-process theory is correct, which is a live question in cognitive science and which this article does not attempt to settle.

Our position: this is the best-documented live dispute this series has covered, and the reason is simply that the participants argued in print rather than in corridors.

A Note On Method

Everything here is verified to August 2026.

We obtained the 2018 critique's abstract verbatim from the authors' own laboratory copy of the paper, plus its highlights from the publisher's record[1][3]. We did not obtain the body of that paper, which runs to fourteen pages, and report no argument from it beyond what the abstract and highlights state.

We obtained the opening and glossary of the reply from the publisher's record[2], and the critics' rejoinder's opening paragraph containing their own four-point summary[4].

We did not obtain any of the papers in the surrounding exchange, including the two further commentaries, and report them from citation records.

We did not obtain Kahneman's 2011 book, and this article does not quote it or characterise its contents.

We did not obtain the World Bank or Institute of Medicine reports and report their existence and content only as the critics' highlights describe them.

All arithmetic and simulation is ours, uses invented parameters, and cannot settle any empirical question.

This article discusses a dispute in cognitive science. It is not training, learning and development, or management advice.

The Framework As You Received It

The proposition under dispute, stated in the form the critics attack.

"It is often said that there are two types of psychological processes: one that is intentional, controllable, conscious, and inefficient, and another that is unintentional, uncontrollable, unconscious, and efficient."[1]

The reply restates the same proposition in its own words: "the idea that cognitive processing is either unintentional, uncontrollable, unconscious, and efficient (Type 1) or intentional, controllable, conscious, and inefficient (Type 2)."[2]

Three observations, ours.

Both sides agree on what is being disputed, which is rarer than it should be and makes this exchange unusually easy to follow. The disagreement is about whether that proposition is the theory.

Note the structure carefully. Four features, each binary, asserted to travel together. This is not a claim that thinking is sometimes fast and sometimes slow. It is a claim that four distinct properties are correlated so tightly that they define two categories.

And notice what is absent from both statements: any mention of speed. The words fast and slow do not appear in the formal proposition at all, though they dominate the popular version. We return to that.

The World Bank Endorsed It

The fact that makes this consequential rather than academic.

The critique's published highlights record: "The distinction between Type 1 and Type 2 processing has grown more popular each decade. The World Bank and Institute of Medicine issued reports in 2015 endorsing the Type 1/Type 2 distinction, and urging decision makers and medical practitioners around the globe to rely on Type 2 thinking."[3]

We did not obtain either report and report this as the critics' characterisation of them.

Three observations, ours.

This is the pattern the fiftieth article named as the most costly in the series: a finding travels into policy faster than the debate about it travels anywhere. Reports issued in 2015 urging global decision makers toward Type 2 thinking; a paper in 2018 arguing the distinction lacks empirical support.

The direction of the advice is worth pausing on. Rely on Type 2 thinking presupposes not only that the two types exist but that one is reliably better, which is a further claim on top of the disputed one.

And it is precisely the advice every corporate behavioural training deck gives. Slow down, engage the deliberate system, override the automatic one. That instruction is downstream of a framework whose own literature was, three years later, publicly disputing whether the two categories exist.

A Convenient And Seductive Myth

The critique, in full, because it is short and every clause carries.

Melnikoff, D. E., and Bargh, J. A. (2018), The Mythical Number Two, Trends in Cognitive Sciences, 22(4), 280–293[1][5].

The abstract: "It is often said that there are two types of psychological processes: one that is intentional, controllable, conscious, and inefficient, and another that is unintentional, uncontrollable, unconscious, and efficient. Yet, there have been persistent and increasing objections to this widely influential dual-process typology. Critics point out that the 'two types' framework lacks empirical support, contradicts well-established findings, and is internally incoherent. Moreover, the untested and untenable assumption that psychological phenomena can be partitioned into two types, we argue, has the consequence of systematically thwarting scientific progress. It is time that we as a field come to terms with these issues. In short, the dual-process typology is a convenient and seductive myth, and we think cognitive science can do better."[1]

Four observations, ours.

Three separate charges are laid, and they are not the same charge. Lacks empirical support is a claim about missing evidence. Contradicts well-established findings is a claim about conflicting evidence. Internally incoherent is a claim that does not depend on evidence at all.

The fourth sentence is the strongest and the one most likely to be resisted: not merely that the framework is wrong but that it is "systematically thwarting scientific progress." That is an accusation about opportunity cost, and it is why the reply was as sharp as it was.

Note the hedge that is genuinely there: "we argue." The authors mark the thwarting claim as their own argument rather than an established fact, which is more careful than the surrounding rhetoric suggests.

And "persistent and increasing objections" tells you this was not a novel attack in 2018. The critics position themselves as joining a queue, and their rejoinder later names five prior works they were joining[4].

The Four-Part Argument

The critics' own numbered summary of their case, from their rejoinder, which is the clearest statement of it we obtained.

"We recently joined other researchers in challenging the popular assumption that 'consciousness', 'efficiency', 'intentionality', and 'controllability' are correlated such that they form two clusters, those typically called 'Type 1' and 'Type 2', which we refer to collectively as the dual-process typology. More specifically, we argued that (i) these processing features have never been shown to correlate with one another, (ii) there are good reasons to assume that the features are, in fact, completely uncorrelated, and (iii) the features are incoherent, therefore (iv) the dual-process typology should be abandoned."[4]

Four observations, ours, one per clause.

Claim (i) is the strongest and most checkable. It does not say the correlations are absent; it says nobody has shown them. That is a claim about the literature, and it is the kind of claim that can be refuted by producing a citation.

Claim (ii) is much stronger and much harder to establish. Asserting good reasons to think the features are completely uncorrelated is a theoretical argument, and we did not obtain the body of the paper where it is made.

Claim (iii), incoherence, is independent of the other two. If the features are conceptually incoherent, no amount of correlational evidence would rescue them.

And (iv) follows only if you accept the premise that the typology is the theory. That premise is exactly what the defenders deny, which is why the whole exchange turns on a definitional question rather than an empirical one.

We Counted The Combinations

Because claim (i) has arithmetic consequences that nobody in the exchange spells out. Our own calculation, and it involves no empirical assumption at all.

Four binary features generate 2 to the power of 4, which is 16 possible combinations.

The typology recognises two of them.

If the four features were statistically independent, each combination would be equally likely, so pure types would account for 2 in 16, or 12.5 percent of processes, and 87.5 percent would be mixtures the framework does not name.

Three observations.

Seven of every eight processes would fall outside both categories. That is what claim (ii) means if taken at face value, and it is a considerably more vivid statement of the problem than "the features are uncorrelated."

This is arithmetic, not evidence. It does not show the features are uncorrelated; it shows what would follow if they were, and that is the disagreement's shape rather than its resolution.

And it clarifies what the defenders need. They do not need the features to be perfectly aligned. They need them aligned enough that the two named categories cover most of the interesting cases, or they need to give up the features, which is what they do.

How Aligned Would They Have To Be

Putting a number on "enough". Our own simulation, five hundred thousand processes, each of the four features modelled as a noisy reading of one underlying dimension. The loading controls how tightly they travel together. No source states any of these parameters and this cannot settle anything.

At a loading of zero, implying no correlation between features: 12.5 percent pure types, matching the arithmetic above.

At a loading of 0.5, implying pairwise correlations of about 0.25: 25.3 percent pure.

At 0.7, pairwise about 0.49: 39.3 percent pure, so a clear majority still mixed.

At 0.9, pairwise about 0.81: 63.9 percent pure.

At 0.99: 88.3 percent pure.

Three observations.

The result is unforgiving. Even with the four features correlating at roughly 0.49 with one another, which would be a strong finding in psychology, a majority of processes are still mixed types.

To get the two categories to describe most of what happens, you need pairwise correlations around 0.8. Correlations of that size between four distinct psychological properties would be a remarkable empirical result, and it is precisely the result the critics say has never been demonstrated.

And we want to be exact about what this does and does not establish. It establishes that the alignment claim is demanding. It does not establish that the alignment is absent, because we have measured nothing.

Very Few Processes Fit

The critics' own empirical claim on this point, from their published highlights.

"Since the Type 1/Type 2 distinction first emerged, researchers have discovered that very few processes fit into either category, most possess some mixture of Type 1 and Type 2 features."[3]

Two observations, ours.

That sentence is the empirical version of our arithmetic, asserted by the critics as a research finding rather than a derivation. Very few fit; most are mixtures. Our independence calculation gives 12.5 percent and 87.5 percent, which is the shape their sentence describes.

We did not obtain the evidence behind it. It appears in a highlights box, not in a results section we can read, and the reply disputes the interpretation rather than the observation.

The Defence

The reply, published in the same journal months later.

Pennycook, G., De Neys, W., Evans, J. St. B. T., Stanovich, K. E., and Thompson, V. A. (2018), The Mythical Dual-Process Typology, Trends in Cognitive Sciences, 22(8), 667–668[2][6].

It characterises the attack fairly: "Melnikoff and Bargh offer a challenge to what they term the dual-process typology... The authors argue that no one has ever tested this proposition and they provide examples of thoughts that do not abide by the proposed featural configuration of the typology (e.g., that processing might be unconscious but also intentional). These examples, it is argued, invalidate the common 'two-types framework' and the authors conclude that distinguishing between two types of processes is 'systematically thwarting scientific progress'."[2]

Three observations, ours.

The example given is concrete and damaging: processing that is unconscious but also intentional. Under the typology, unconscious belongs to Type 1 and intentional to Type 2, so such a process belongs to neither and the categories do not exhaust the space.

Note that the defenders restate the charge accurately and do not dispute the example. That is the mark of a serious reply and it narrows the disagreement usefully.

And note the phrase "what they term the dual-process typology." The defenders are signalling, in their first sentence, that the thing being attacked is a construction of the attackers. That is the entire defence, and the next section is about whether it works.

Defining Features Versus Incidental Correlates

The distinction the whole dispute turns on, obtained from the reply's own glossary.

The reply's glossary defines dual-process typology as "a term introduced by Melnikoff and Bargh to represent the idea that cognitive processes can be sorted into two types with aligned characteristics: (i) Type 1 processes, which are unintentional, uncontrollable, unconscious, and efficient; (ii) Type 2 processes, which are intentional, controllable, conscious, and inefficient."[2]

And it defines incidental correlates as "also introduced by Evans and Stanovich; the various characteristics that have been associated with Type 1 and Type 2 processes (e.g., intentionality, controllability, consciousness, efficiency) but do not define the distinction for a given" dual-process theory[2].

Four observations, ours, and this is the most important passage in the article.

The defence is that the four features were never the definition. They are things that tend to accompany the two types without constituting them, and the term for that was introduced by the theorists themselves, not invented for this argument.

A citing source states the defenders' positive position: "Type 1 or intuitive processes with automaticity as their defining feature and Type 2 or deliberative processes with dependence on working memory and cognitive decoupling from irrelevant features as their necessary features."[7] So the real definition is automaticity on one side and working memory dependence on the other.

The defenders' logical move is stated crisply by another citing source: "As long as there exists a qualitative difference along one single dimension, there is no need for such a difference along multiple features."[7]

And that is a valid argument. If the theory only ever claimed one defining dimension, then demonstrating that four other properties fail to align is beside the point, however vivid the demonstration.

Is That A Good Defence

Our assessment, stated directly because a reader deserves one.

Four points.

Technically, yes. A theory is entitled to specify its own defining criteria, and if the published record shows Evans and Stanovich distinguishing defining features from incidental correlates before the attack, the defence is not invented after the fact. The glossary attributes the term to them[2], and a 2012 paper by Stanovich and Toplak is titled Defining features versus incidental correlates of Type 1 and Type 2 processing[8], six years before the exchange.

Rhetorically, it is expensive. The four features are what makes the framework memorable, teachable and useful to a lay audience. Conceding they are incidental is conceding that the version in circulation is not the theory.

For our purposes it is decisive in an unexpected direction. If the four features are incidental, then a training programme that teaches them as the framework is teaching something its own proponents do not defend.

And a citing source suggests the defenders themselves see the concession's value: the non-rigid alignment "alleviates many of the critiques that have been leveled against dual-process theories, particularly that they posit an" alignment, with our source truncating[7]. Loosening the claim is what makes it survivable.

Who Signed The Reply

A detail that determines how to weigh the exchange, and which is easy to miss.

The reply's authors are recorded at Yale University, Paris Descartes University, the University of Plymouth, the University of Toronto, and the University of Saskatchewan[2][6].

Three observations, ours.

The fourth author is Keith Stanovich, whose work with Richard West is credited by multiple sources as the origin of the System 1 and System 2 terminology[9]. The person who named the systems is defending them.

The reply is therefore not a neutral referee's verdict, and we do not present it as one. It is an interested party's rebuttal, and it should be read as such and taken seriously anyway, because the argument stands or falls independently of who made it.

And two of the five are at Canadian institutions, in Toronto and Saskatoon, which is a small point but a real one for this publication's readers: the defence of the framework in every Canadian corporate training deck was partly written in Canada.

And Then There Is Fast And Slow

The part that should trouble a business reader most, because speed is the entire popular framing and it is not in either side's formal statement of the theory.

Neither the critics' proposition nor the defenders' restatement mentions speed. The four features are intentionality, controllability, consciousness and efficiency[1][2]. Fast and slow are not among them.

And the time-course assumption is separately contested. Reference lists in the exchange identify Bago, B., and De Neys, W. (2017), Fast logic? Examining the time course assumption of dual process theory, Cognition, and Newman, I. R., and colleagues (2017), Rule-based reasoning is fast and belief-based reasoning can be slow: challenging current accounts, with the title truncating in our source[4].

We obtained neither paper and report titles and citations only.

Three observations, ours.

Both titles assert, in the title itself, that the speed ordering can run the other way. Rule-based reasoning fast, belief-based reasoning slow, is the exact inversion of the popular model.

A modern citing source states the general position: "Defining intuition and deliberation solely by surface features such as speed or effort has been criticized as inadequate, as these features are not consistently observed, do not reliably co-occur, and reveal little about the underlying processes."[10]

And this is the sharpest practical point in the article. The popular framework is named after a property that neither side treats as definitional and that a body of work is titled to contradict.

The Consensus Sentence

One line from the critics' highlights that we think is the single most consequential claim in the whole exchange, and which we handle carefully because it is contested by construction.

"A consensus is emerging among the critics and top proponents of the dual-process typology that the classic distinction between Type 1 and Type 2 processing is flawed."[3]

Three observations, ours.

If accurate, this is remarkable: the critics claim the proponents agree with them that the classic distinction is flawed, with the remaining dispute being about what replaces it.

It is also a claim by one side about the other side's views, published in a highlights box, and the reply's existence demonstrates that the proponents did not accept the characterisation wholesale.

And yet the reply's own defence is consistent with a weaker version of it. Conceding that four widely taught features are incidental correlates rather than defining criteria is agreeing that the classic distinction, as commonly stated, is flawed. The disagreement is over whether that is a correction or a refutation.

Persecuted By An Integer

A detail of scholarly form, recorded because it tells you what kind of paper this is.

The critique opens with an epigraph: "'My problem is that I have been persecuted by an integer'. – George A. Miller."[5]

Two observations, ours.

The line is from the opening of Miller's celebrated paper on the limits of short-term memory, whose title contains the number seven, and the joke is that a number in a title acquires a life independent of the argument.

The critics are making a claim about how numbers in titles behave, and applying it to the number two. Whether or not one accepts the substantive argument, the observation about intellectual history is sound and this series has documented it repeatedly: the number survives, the qualifications do not.

What The Exchange Itself Demonstrates

Stepping back from the content to the form. Ours.

Within one journal and roughly four months, the record contains a target article at 22(4), 280–293; a commentary by Osman at 22(4), 276–277 titled Persistent Maladies: The Case of Two-Mind Syndrome; a commentary by Chater at 22(5), 369–371 titled Is the Type 1/Type 2 Distinction Important for Behavioral Policy?; the five-author reply at 22(8), 667–668; and the critics' rejoinder at 22(8), 668–669[11].

We obtained none of the two commentaries and report their titles, authors, journal, volume, issue and pages from a citation record.

Three observations, ours.

This is the correction process working at speed and in public, and it stands in sharp contrast to most of what this series has documented. The thirty-fourth article covered a debunking that stood uncorrected for thirty-three years.

Chater's title is the one a business reader should notice: Is the Type 1/Type 2 Distinction Important for Behavioral Policy? Someone asked precisely the question this article is about, in the same journal, within two months, and we could not obtain the answer. That is a real gap and we would rather name it than write around it.

And the rejoinder appearing in the same issue as the reply, at consecutive page numbers, tells you the journal ran them together deliberately.

Time To Move On

Where the field appears to have gone, from a later source.

A citing paper records the state of play: "The dual vs single process model debate has been around for a long time and continues to fill the pages of prominent journals... My goal in the present paper is to convince the community that it is time to move on and leave the debate behind us."[7]

The same source records the defenders' unbowed position: "Dual process theories are still the best model for describing and explicating the nature of human reasoning."[7]

We did not obtain the paper making either statement and report both from a repository page carrying citing text.

Two observations, ours.

Those two sentences appear on the same page and describe the same field. One says the debate is exhausted; the other says the theory has won. Both are positions held by serious researchers, which is the honest state of the question.

And a proposal to move on is not the same as a resolution. It is a judgment that further argument along the current lines has low expected value, which is a reasonable thing to conclude and is not evidence about who was right.

What Actually Survives

Our own reading, stated without hedging, of what a careful person should retain.

Five things.

Some processing is automatic and some depends on working memory. This is the defenders' actual definitional claim and nothing in the critique touches it. Automaticity is a real, measurable property.

Cognitive load impairs the effortful kind. This follows from the working memory criterion and is among the better-established findings in cognitive psychology, though we have not surveyed that literature here and do not grade it.

The four popular features do not reliably travel together. Both sides now agree on something close to this. The critics call it a refutation; the defenders call it the reason those features were never definitional.

Speed is the weakest of the popular criteria. It is absent from both formal statements and two papers in the exchange's own reference lists are titled to contradict it.

And the binary is the part to discard. On our own arithmetic, four independent binary features generate sixteen combinations and the framework names two. Whatever the true correlations, the assumption that every mental act belongs to one of two boxes is the assumption that does the damage.

What The Training Deck Gets Wrong

The application, and we are going to be blunt about it. Ours.

The standard corporate version of this framework makes four claims. Assessed against the exchange above:

"There are two systems." Disputed in print by one side and defended by the other only in a much weaker form, where the two types are defined by automaticity and working memory dependence rather than by the features usually listed.

"System 1 is fast, System 2 is slow." Speed appears in neither side's formal statement, and two papers in the exchange's reference lists are titled to challenge the ordering directly.

"System 1 causes bias, System 2 corrects it." This is a further claim the exchange does not establish and which the World Bank and Institute of Medicine reports are described as having promoted.

"Therefore slow down and engage System 2." An instruction that inherits every weakness above and adds the assumption that deliberation improves accuracy, which this series has found contested elsewhere: the forty-second article reported that the most persistent form of overconfidence is excessive certainty, which is not obviously cured by thinking longer.

One observation. Every one of those four claims is weaker than the deck presents it, and the fourth is weaker than the first three because it compounds them.

The Specific Harm

Why this matters commercially rather than just intellectually. Ours, and offered as reasoning.

Four costs.

It misdirects remedy design. If the model is that a fast system produces errors a slow system fixes, the intervention is always to slow people down. This series has repeatedly found that the effective interventions were structural: independent estimates before discussion, contemporaneous records, blinding the measurer. None of those works by slowing anyone down.

It licenses a false confidence in deliberation. A manager who has deliberated feels they have engaged the reliable system. The forty-fourth article found that subjects who explicitly believed outcomes should not affect their judgments were affected anyway, and deliberating did not save them.

It provides a vocabulary for dismissing disagreement. Describing a colleague's objection as System 1 is an argumentative move dressed as a diagnosis, and the framework's contested status makes it an unearned one.

And it is unfalsifiable in practice. Any good decision can be attributed to deliberation and any bad one to intuition, after the fact, which is the forty-third article's problem operating on a framework rather than on a file.

What To Say Instead

The replacement vocabulary we would use, derived from the defenders' actual definitional criteria rather than from the popular features. Ours, untested.

Four substitutions.

Instead of "that's System 1", say "that judgment was automatic", which is a claim about a measurable property and does not smuggle in three other assertions.

Instead of "engage System 2", say "this needs working memory, so remove the load", which points at an actionable condition rather than an act of will.

Instead of "slow down", say "write it down", because on the evidence in this series a contemporaneous record does work and additional deliberation time has a weaker case.

And instead of "avoid bias", say "change who measures what, and when", which is the structural family of remedies the fiftieth article's audit found to be the durable residue of fifty articles.

What To Do

Stop using System 1 and System 2 as an explanation. The four features that make the framework teachable are described by its own defenders as incidental correlates that do not define the distinction.

Drop the speed framing first. Speed appears in neither side's formal statement of the theory, and two papers in the exchange's reference lists are titled to challenge the time-course assumption directly.

Keep automaticity and working memory. Those are the defenders' actual criteria, they are measurable, and nothing in the critique touches them.

Discard the binary. Four independent binary features generate sixteen combinations; the framework names two; the critics report that most processes are mixtures.

Treat "slow down and think harder" as unsupported by this exchange. It is a further claim built on top of a disputed one, and the structural remedies this series has found effective do not work by slowing anyone down.

Notice when the framework is used to dismiss rather than to explain. Calling an objection intuitive is an argumentative move, not a diagnosis.

Ask what a programme built on this actually measures. The thirty-eighth article found a 492-study meta-analysis in which changes to a target measure did not carry through to behaviour, and the same question applies here.

Do not conclude that dual-process theory is dead. Five researchers including the terminology's co-originator published a defence, one later source says the theory remains the best available model, and this article does not settle the question.

The Limits Of This Analysis

Several caveats matter and this article's scope is narrower than its subject. This article discusses a dispute in cognitive science and is not training, learning and development, or management advice. Everything is verified to August 2026. We obtained the critique's abstract verbatim from the authors' own laboratory copy and its highlights from the publisher, and we did not obtain the body of the paper, which runs to fourteen pages; we report no argument from it beyond what the abstract and highlights state, and in particular we did not read the evidence behind the claim that very few processes fit either category. We obtained the reply's opening and glossary and not its argument. We obtained the rejoinder's opening paragraph only. We did not obtain either of the two commentaries in the exchange, including the one asking directly whether the distinction matters for behavioural policy, which is the question closest to this article's purpose and whose answer we cannot report. We did not obtain the World Bank or Institute of Medicine reports and describe them solely as the critics' highlights characterise them. We did not obtain Kahneman's 2011 book and this article neither quotes nor characterises it. We did not obtain the papers challenging the time-course assumption, the 2012 paper on defining features versus incidental correlates, the 2009 critical evaluation, or the 2013 exchange in another journal, and report all of these from citation records. Two statements about the current state of the field are taken from a repository page carrying citing text rather than from the papers making them. All arithmetic and simulation is ours; the sixteen-combination calculation is a derivation and not a finding, the loading model uses parameters stated by no source, and neither can establish what the true correlations between the four features are. This article does not attempt to determine whether dual-process theory is correct, and its practical sections are our own reasoning, untested.

Related Reading

No. 54 Three Questions Half a sample had already seen the questions and scored 2.36 against 1.48. Then six studies with 2,500 participants found the test predicted just as... Behavioural Finance 492 Studies And A Broken Chain What to ask of any training programme built on a contested construct. Behavioural Finance Fifty In The audit finding that the durable remedies in this series are structural, not motivational.

Frequently Asked Questions

Is System 1 and System 2 wrong?

The version taught in business settings is not defended by anyone in this dispute. Critics called the two-type framework a convenient and seductive myth. Defenders replied that the four features it is built from are incidental correlates that do not define the distinction. The popular model uses those features as the definition.

What is the critics' actual argument?

Four numbered claims in their own words: the features have never been shown to correlate, there are good reasons to think they are completely uncorrelated, the features are incoherent, and therefore the typology should be abandoned. The first is checkable by producing a citation; the second is a theoretical argument we did not obtain.

Why does the number of features matter?

Because four binary features generate sixteen combinations and the framework names two. On our own arithmetic, if the features were independent only 12.5 percent of processes would be pure types. For the two categories to describe most cases, the features would need to correlate with each other at around 0.8.

Is the defence any good?

Technically yes. A theory may specify its own defining criteria, and a 2012 paper titled "Defining features versus incidental correlates" predates the attack by six years. But it is expensive: the four features are what make the framework teachable, and conceding they are incidental concedes that the circulating version is not the theory.

What about fast and slow?

Speed appears in neither side's formal statement of the theory. The four features are intentionality, controllability, consciousness and efficiency. And two papers in the exchange's reference lists are titled to challenge the time course directly, one asking whether logic can be fast and one asserting that rule-based reasoning is fast while belief-based reasoning can be slow.

So is dual-process theory dead?

No, and we do not claim otherwise. Five researchers including the terminology's co-originator published a defence, one later source records the position that dual process theories remain the best available model of human reasoning, and another argues the debate itself should be set aside. This article reports the exchange; it does not settle it.

What should replace it in practice?

Automaticity and working memory dependence, which are the defenders' actual criteria and are measurable. Practically: say a judgment was automatic rather than that it was System 1, remove cognitive load rather than instructing people to engage a system, and write the number down rather than slowing down.

References

1. Melnikoff, D. E., & Bargh, J. A. (2018). The Mythical Number Two. Trends in Cognitive Sciences, 22(4), 280–293. Copy hosted on the authors' own university laboratory website, reproducing the paper's front matter and abstract in full: on it often being said that there are two types of psychological processes, one intentional, controllable, conscious and inefficient, and another unintentional, uncontrollable, unconscious and efficient; on there having been persistent and increasing objections to this widely influential dual-process typology; on critics pointing out that the two types framework lacks empirical support, contradicts well-established findings, and is internally incoherent; on the untested and untenable assumption that psychological phenomena can be partitioned into two types having, the authors argue, the consequence of systematically thwarting scientific progress; on it being time that the field come to terms with these issues; and on the dual-process typology being, in short, a convenient and seductive myth, with the authors holding that cognitive science can do better. Note: the authors' own laboratory copy. We obtained the front matter and abstract verbatim and did not obtain the body of the paper, which runs to fourteen pages. acmelab.yale.edu
2. Pennycook, G., De Neys, W., Evans, J. St. B. T., Stanovich, K. E., & Thompson, V. A. (2018). The Mythical Dual-Process Typology. Trends in Cognitive Sciences, 22(8), 667–668. Publisher record reproducing the letter's opening and glossary, recording the authors' affiliations at Yale University, Paris Descartes University, the University of Plymouth, the University of Toronto and the University of Saskatchewan; on Melnikoff and Bargh offering a challenge to what they term the dual-process typology, specifically the idea that cognitive processing is either unintentional, uncontrollable, unconscious and efficient (Type 1) or intentional, controllable, conscious and inefficient (Type 2); on those authors arguing that no one has ever tested this proposition and providing examples of thoughts that do not abide by the proposed featural configuration, such as processing that might be unconscious but also intentional; on those examples being argued to invalidate the common two-types framework, with the conclusion that distinguishing between two types of processes is systematically thwarting scientific progress; and reproducing glossary definitions of the dual-process typology as a term introduced by Melnikoff and Bargh, and of incidental correlates as a term also introduced by Evans and Stanovich, being the various characteristics associated with Type 1 and Type 2 processes, such as intentionality, controllability, consciousness and efficiency, but which do not define the distinction for a given dual-process theory. Note: the publisher's record. We obtained the opening and glossary and not the letter's full argument. cell.com
3. Publisher record for the 2018 critique reproducing its published highlights: on the distinction between Type 1 and Type 2 processing having grown more popular each decade; on the World Bank and Institute of Medicine having issued reports in 2015 endorsing the Type 1/Type 2 distinction and urging decision makers and medical practitioners around the globe to rely on Type 2 thinking; on a consensus emerging among the critics and top proponents of the dual-process typology that the classic distinction between Type 1 and Type 2 processing is flawed; and on researchers having discovered, since the distinction first emerged, that very few processes fit into either category, with most possessing some mixture of Type 1 and Type 2 features. Note: the publisher's highlights box. The claims about the World Bank and Institute of Medicine reports, and about an emerging consensus among proponents, are the critics' characterisations; we obtained neither report and note that the reply at reference 2 demonstrates the characterisation was not accepted wholesale. sciencedirect.com
4. Melnikoff, D. E., & Bargh, J. A. (2018). The Insidious Number Two. Trends in Cognitive Sciences, 22(8), 668–669. Publisher record reproducing the rejoinder's opening: on the authors having recently joined other researchers in challenging the popular assumption that consciousness, efficiency, intentionality and controllability are correlated such that they form two clusters, those typically called Type 1 and Type 2, referred to collectively as the dual-process typology; and on the authors having argued that these processing features have never been shown to correlate with one another, that there are good reasons to assume the features are in fact completely uncorrelated, that the features are incoherent, and that therefore the dual-process typology should be abandoned; together with its reference list identifying Keren, G., and Schul, Y., Two is not always better than one: a critical evaluation of two-system theories, Perspectives on Psychological Science, 2009, 4, 533–550; Evans, J. St. B. T., and Stanovich, K. E., Dual-process theories of higher cognition: advancing the debate, Perspectives on Psychological Science, 2013, 8, 223–241; Moors, A., and De Houwer, J., Automaticity: a theoretical and conceptual analysis, Psychological Bulletin, 2006, 132, 297–326; Bago, B., and De Neys, W. (2017), Fast logic? Examining the time course assumption of dual process theory, Cognition; and Newman, I. R., and colleagues (2017), Rule-based reasoning is fast and belief-based reasoning can be slow: challenging current accounts, title truncating. Note: the publisher's record. We obtained the opening paragraph and reference list, and none of the works cited within it. cell.com
5. Hosted copy of the 2018 critique's opening pages, reproducing the article type as an Opinion piece, its running head, its authors' affiliation at the Department of Psychology, Yale University, and its opening epigraph attributed to George A. Miller, that his problem is that he has been persecuted by an integer, followed by the section heading "A Tale of Two Types". Note: a hosted copy of the paper's first page. Our source for the epigraph and article type. cse.buffalo.edu
6. National medical literature database record for Pennycook, G., De Neys, W., Evans, J. St. B. T., Stanovich, K. E., and Thompson, V. A., The Mythical Dual-Process Typology, Trends in Cognitive Sciences, 22(8), 667–668, August 2018, DOI 10.1016/j.tics.2018.04.008, recording the authors' institutional affiliations and the article's keywords as analytic thinking, dual-process theory, intuition and reflection; and noting that no abstract is available for the item, it being a letter. Note: a bibliographic record, used to confirm the reply's authorship, journal, volume, issue, pages, date and DOI independently of the publisher's page. pubmed.ncbi.nlm.nih.gov
7. Repository page for the 2018 reply carrying scholarly citing text, on the dual versus single process model debate having been around for a long time and continuing to fill the pages of prominent journals, citing Evans and Stanovich (2013), De Houwer (2019), Keren (2013), Kruglanski and Gigerenzer (2011), Melnikoff and Bargh (2018), Osman (2018) and Pennycook and colleagues (2018); on one author's goal being to convince the community that it is time to move on and leave the debate behind; on the case for non-alignment not implying that the difference is quantitative, since as long as there exists a qualitative difference along one single dimension there is no need for such a difference along multiple features; on non-rigid alignment alleviating many of the critiques levelled against dual-process theories, particularly that they posit an alignment, our source truncating; on the position that dual process theories are still the best model for describing and explicating the nature of human reasoning; and on the account under which Type 1 or intuitive processes have automaticity as their defining feature and Type 2 or deliberative processes have dependence on working memory and cognitive decoupling from irrelevant features as their necessary features. Note: a repository page reproducing text from papers citing the reply. We obtained none of the citing papers, and the two contrasting statements about the state of the field come from different citing works reproduced on the same page. researchgate.net
8. Reference list carried on a hosted copy of the 2018 rejoinder, identifying Stanovich, K. E., and Toplak, M. E. (2012), Defining features versus incidental correlates of Type 1 and Type 2 processing; Evans, J. St. B. T., and Stanovich, K. E. (2013), Theory and metatheory in the study of dual processing: reply to comments, Perspectives on Psychological Science, 8, 263–271; Toplak, M. E., and colleagues (2014), Assessing miserly information processing: an expansion of the Cognitive Reflection Test, Thinking and Reasoning, 20, 147–168; Pennycook, G., and colleagues (2015), What makes us think? A three-stage dual-process model of analytic engagement, Cognitive Psychology, 80, 34–72; and De Neys, W. (ed.) (2017), Dual Process Theory 2.0. Note: a reference list on a hosted copy; citations only. We obtained none of these works. The 2012 title is cited here as evidence that the defining-features distinction predates the 2018 exchange by six years. wdeneys.org
9. Repository page for the 2018 critique carrying scholarly citing text, on a dominant paradigm in the psychology of reasoning postulating that human reasoning involves two distinct cognitive mechanisms with varying terminologies used to name them, citing Epstein (1994), Evans (2006), James (1890), Kahneman (2003), Sloman (1996) and Stanovich and West (2000); and on some having contested a strict two-system view of reasoning, citing Grayot and colleagues (2024), Keren and Schul (2009) and Melnikoff and Bargh (2018). Note: a repository page reproducing citing text. Our source for the attribution of the terminology's origin to Stanovich and West (2000), which we did not obtain. researchgate.net
10. Repository page carrying later citing text on the same critique, on defining intuition and deliberation solely by surface features such as speed or effort having been criticised as inadequate, since these features are not consistently observed, do not reliably co-occur, and reveal little about the underlying processes, citing Gawronski and colleagues (2014), De Neys (2021) and Melnikoff and Bargh (2018). Note: a repository page reproducing citing text from a later paper we did not obtain. researchgate.net
11. National medical literature database record for the 2018 critique, Trends in Cognitive Sciences, 22(4), 280–293, PMID 29571664, recording its keywords as automaticity, dual process, dual system, type 1 and type 2, and listing the related items in the exchange: Osman, M., Persistent Maladies: The Case of Two-Mind Syndrome, Trends in Cognitive Sciences, April 2018, 22(4), 276–277, DOI 10.1016/j.tics.2018.02.005, no abstract available; Chater, N., Is the Type 1/Type 2 Distinction Important for Behavioral Policy?, Trends in Cognitive Sciences, May 2018, 22(5), 369–371, DOI 10.1016/j.tics.2018.02.007; Melnikoff, D. E., and Bargh, J. A., The Insidious Number Two, August 2018, 22(8), 668–669; and Pennycook, G., and colleagues, The Mythical Dual-Process Typology, August 2018, 22(8), 667–668. Note: a bibliographic record, used to establish the full shape and chronology of the exchange. We obtained neither of the two commentaries, including the one asking whether the distinction matters for behavioural policy. pubmed.ncbi.nlm.nih.gov

This article discusses a dispute in cognitive science and is not training, learning and development, or management advice. The critique's abstract and highlights were obtained but not its fourteen-page body; the reply's opening and glossary were obtained but not its argument; the rejoinder's opening paragraph was obtained. Neither commentary in the exchange was obtained, including the one asking whether the distinction matters for behavioural policy. Kahneman's 2011 book was not obtained and is neither quoted nor characterised here. The World Bank and Institute of Medicine reports were not obtained and are described solely as the critics characterise them. All arithmetic and simulation is the authors' own, uses parameters stated by no source, and cannot establish the true correlations between the four features. This article does not attempt to determine whether dual-process theory is correct.