A person is asked to solve a reasoning problem alone.
They make an error.
Put the same person into an exchange where someone else challenges the answer, and something changes. Reasons become visible. Weak justifications can be attacked. Better arguments can move the group.
For Hugo Mercier, that difference is not a side effect of reasoning.
It is close to the center of what reasoning may be for.
Mercier's research programme has spent more than a decade challenging a familiar picture of reason as primarily an internal tool for solitary truth-seeking.
Current research identity
As of 23 September 2026, Hugo Mercier is a Directeur de Recherche at the CNRS and a statutory member of the Institut Jean Nicod at ENS-PSL in Paris, where his work is connected to social cognition, collective intelligence and epistemic vigilance.
His institutional profile describes two overlapping themes: reasoning and epistemic vigilance.
That pairing is important.
Reasoning concerns how people produce and evaluate reasons.
Epistemic vigilance concerns how people decide whether information communicated by others deserves acceptance.
Together they form a research programme about how minds think with and about other minds.
The argumentative theory changed the question
Mercier and Dan Sperber's 2011 Behavioral and Brain Sciences paper, Why do humans reason? Arguments for an argumentative theory, proposed that the evolutionary function of reasoning is primarily argumentative.
On this account, reasoning helps people produce arguments to justify claims and evaluate arguments offered by others.
This was a deliberately revisionary proposal.
Instead of asking why a supposedly truth-maximizing reasoning system is so biased, Mercier and Sperber asked whether some apparent flaws make more sense if reasoning evolved for social exchange.
The theory later developed into what they call an interactionist account of reason.
Confirmation bias looks different under this theory
If reasoning is mainly a solitary truth engine, searching for arguments that support one's own view looks like a design failure.
If reasoning also evolved to defend claims in interaction, that same tendency becomes less surprising.
Mercier and Sperber use this reframing to explain why people can be biased when generating reasons for their own position yet more discriminating when evaluating other people's arguments.
This is not the same as saying confirmation bias is good.
It is an attempt to explain why a directional asymmetry might exist.
The normative question remains separate: does the process lead to more accurate judgments in a given environment?
Groups are not automatically wiser
The interactionist account is sometimes caricatured as "reasoning works in groups."
That is too simple.
Groups can polarize, conform, defer to status, repeat errors and exclude dissent.
The theory's stronger claim is that exchanging reasons can create conditions in which argument production and argument evaluation complement one another.
Whether that produces better outcomes depends on conversational structure, incentives, diversity of information, power and willingness to revise.
The Institute should therefore treat interaction as a mechanism opportunity, not a guarantee of collective intelligence.
Epistemic vigilance extends the programme beyond argument
The 2010 paper Epistemic Vigilance, written by Dan Sperber, Fabrice Clément, Christophe Heintz, Olivier Mascaro, Hugo Mercier, Gloria Origgi and Deirdre Wilson, begins from a basic problem.
Human beings benefit enormously from communication.
That same dependence exposes them to error and deception.
The authors propose a suite of mechanisms for evaluating both sources and content.
People attend to competence, benevolence, plausibility and arguments.
Mercier's later work on gullibility develops the idea that humans may be much less passively credulous than common narratives suggest.
The gullibility question matters now
In a 2017 review, Mercier asked How Gullible Are We?
He argued that evidence from psychology and social science does not support a picture of humans as indiscriminately receptive to communicated information.
People often check messages against prior beliefs, assess source competence and motives, and evaluate arguments.
This claim has practical consequences for misinformation research.
If people are not simply empty vessels waiting to be infected by falsehood, then interventions should examine why particular claims overcome vigilance rather than assuming a universal gullibility deficit.
Prior beliefs can protect and trap
Epistemic vigilance uses existing knowledge to assess new claims.
That is efficient.
It also creates a familiar difficulty: a false prior belief can cause accurate corrective information to look implausible.
A person's defense against misinformation can therefore become a defense of misinformation.
Mercier's framework does not remove this problem.
It helps state it more precisely.
Vigilance is adaptive when its criteria are well calibrated to source quality and evidence. It becomes conservative when plausibility is judged against a distorted prior model of the world.
Argument quality is not rhetorical success
A persuasive argument can be weak.
A strong argument can fail because it is poorly communicated, arrives from a distrusted source or threatens identity.
Mercier's work is most useful when argument evaluation is treated as a cognitive process that can be separated from mere persuasion.
For the Institute, this distinction protects discernment from becoming communication strategy.
The aim is not to produce reasons that win.
It is to improve the conditions under which reasons can be inspected, challenged and revised.
The interactionist account is influential and contested
Mercier and Sperber's theory has generated extensive discussion across cognitive science, philosophy and social epistemology.
Critiques have questioned the evolutionary claims, the breadth of the proposed function, the interpretation of individual reasoning failures and the extent to which reasoning is best understood through argumentation rather than problem solving, planning or other cognitive roles.
Those disagreements are not peripheral.
A researcher profile should not turn theoretical influence into consensus.
The theory is valuable partly because it generates testable predictions about when reasoning should improve under interaction.
Evidence should be parsed by level
Several types of claim can become entangled in discussion of Mercier's work.
One is descriptive: people often reason differently when generating versus evaluating arguments.
Another is functional: reasoning is adapted for social argumentation.
A third is evolutionary: selection favored this architecture because of communication and coordination problems.
A fourth is normative: argumentative exchange improves truth-seeking.
Evidence for one does not automatically establish the others.
This layering is essential for scholarly restraint.
The Enigma of Reason broadens the theory
Mercier and Sperber's 2017 book The Enigma of Reason develops the interactionist theory for a wider audience while retaining its cognitive-science ambition.
The book argues that reason makes more sense when understood as a social competence used in giving and evaluating reasons.
It also emphasizes how individual biases can coexist with collective benefits when others are available to challenge weak arguments.
The book is intellectually important.
It is not itself a substitute for primary empirical studies testing the component predictions.
Mercier's later work questions pessimistic stories about mass persuasion
In Not Born Yesterday and related research, Mercier develops a broader argument that people are not easily manipulated by messages that conflict with their interests, prior knowledge and social expectations.
This stance pushes against popular accounts in which propaganda or misinformation automatically "brainwashes" passive audiences.
The corrective is useful.
But the opposite overclaim should also be avoided.
People can be deceived, coordinated into false beliefs, influenced by repetition and manipulated under specific conditions.
The empirical question is which conditions defeat or redirect ordinary vigilance.
This matters for AI because fluent reasons are cheap now
Generative AI can produce arguments, counterarguments and justifications almost instantly.
That changes a scarcity condition.
Historically, generating coherent reasons required human effort.
Now the user can receive ten arguments for a position before they have evaluated the first one.
Mercier's programme suggests a useful question for AI-mediated reasoning:
Does abundant argument improve evaluation, or does it overwhelm the very vigilance that arguments are supposed to recruit?
This is a research frontier, not a settled application of the interactionist theory.
AI can also externalize adversarial dialogue
One promising use of generative systems is structured opposition.
A model can be asked to identify assumptions, generate counterarguments or represent alternative hypotheses.
That resembles the social function Mercier assigns to reason: ideas become stronger when they encounter challenge.
But a model-generated counterargument is not an independent mind with separate evidence or incentives.
It may reproduce the same training-data errors as the original answer.
Artificial opposition can simulate disagreement without creating genuine epistemic independence.
That boundary should be explicit.
The European location of the programme matters substantively
Mercier's current work at the Institut Jean Nicod sits inside a French and European research environment linking cognitive science, philosophy, social cognition and AI-and-society questions.
In 2026 he also participated in French parliamentary discussions on AI in research and higher education.
The relevance for the Institute is not institutional prestige.
It is the continuity between his long-standing work on how people evaluate reasons and new questions about how machine-generated information enters human epistemic systems.
That continuity makes the programme especially pertinent to Inner Technology.
The theory makes disagreement a design variable
If reasoning has an important social function, then the quality of disagreement becomes part of the cognitive environment.
A meeting in which dissent is punished does not provide the same reasoning conditions as one in which competing explanations must be articulated. A group in which everyone received the same information does not gain much epistemic diversity merely by adding more voices.
Mercier's programme therefore has institutional implications even where the evolutionary theory remains contested.
Teams can ask:
- Are alternative hypotheses genuinely represented?
- Can lower-status members challenge higher-status claims?
- Are people rewarded for finding faults in their own side's argument?
- Is evidence available to everyone, or only rhetoric?
- Does the process record why a decision changed?
These are not consequences proved by the interactionist theory. They are reasonable design questions suggested by its emphasis on argument exchange.
Argument evaluation itself can be asymmetric
The interactionist account often highlights a difference between generating reasons for one's own view and evaluating arguments offered by others.
That asymmetry is psychologically plausible but normatively ambiguous.
A skeptical listener can detect flaws in another person's case while applying little scrutiny to assumptions they already accept. Conversely, social identity can make strong counterarguments feel threatening and weak in-group arguments feel persuasive.
This means collective reasoning needs more than argument production.
It needs symmetry rules: comparable evidentiary standards across favored and disfavored claims, explicit criteria before outcomes are known, and permission to revise.
The Institute should treat those rules as hypotheses for better reasoning environments, not as automatic products of conversation.
Collective intelligence depends on partially independent cognition
Ten people repeating the same source are not ten independent confirmations.
The same principle applies to AI-supported groups.
If every participant asks the same model for a summary, apparent diversity can conceal a shared upstream dependency. The group may exchange many reasons that originated from one system.
Mercier's social account of reasoning makes this dependence especially important in the AI age.
The value of challenge comes partly from different information, different priors and different routes to the conclusion.
Artificially multiplying arguments without multiplying epistemic independence can create the appearance of deliberation without its strongest corrective property.
Research profiles should preserve falsifiability
There is a temptation to turn a generative theory into a worldview.
The interactionist account is more scientifically useful when translated into discriminating predictions: under what conditions should dialogue outperform isolated reasoning, which argument structures should be persuasive, and when should vigilance fail?
A profile should therefore preserve the places where the programme could be wrong.
That is not disrespect to the researcher.
It is the form of respect scientific ideas require.
The programme also changes how expertise should be heard
If argument evaluation is central to reasoning, expertise presents a subtle case.
Experts often have better reasons because they possess more evidence and richer causal models. A non-expert should not treat every expert claim as merely one opinion among many.
At the same time, deference becomes epistemically stronger when the grounds of expertise can be inspected: relevant training, track record, access to evidence, agreement or disagreement among peers, and the fit between expertise and the specific claim.
Mercier's work on epistemic vigilance helps explain why rational social learning requires both receptivity and filtering.
This is especially important in institutions. A culture that demands a full independent proof from every employee wastes expertise. A culture that treats status as sufficient reason disables scrutiny.
The design problem is calibrated deference: enough trust to learn from distributed knowledge, enough argument visibility to detect when authority has outrun evidence.
Persuasion research should distinguish acceptance from understanding
An argument can change a response without changing the listener's underlying model.
Someone may defer because the speaker is prestigious, because the conclusion is socially safe, or because the argument itself is compelling. These pathways have different implications for epistemic agency.
For Inner Technology, the strongest form of reason exchange is not mere compliance. It leaves the recipient better able to reconstruct why the conclusion deserves belief and what evidence would change it.
That standard connects Mercier's programme to education and science communication without assuming that every persuasive success is rational improvement.
What the evidence does not show
Mercier's work does not show that all reasoning is primarily argumentative in every context.
It does not show that groups automatically reason better than individuals.
It does not show that humans are immune to propaganda or misinformation.
Evidence of epistemic vigilance does not mean vigilance is always well calibrated.
The interactionist theory's evolutionary account should not be presented as settled fact simply because some behavioral findings are consistent with it.
And a research programme's influence does not determine whether each of its claims is correct.
What this changes for Inner Technology
Mercier's work shifts the design question from "How do we make individuals think harder?" to something more social:
What kinds of exchanges make reasons inspectable?
Can people state a claim clearly enough to be challenged?
Can someone else disagree without status deciding the outcome?
Are reasons evaluated symmetrically when they support or threaten prior beliefs?
Does the institution reward changing one's mind when the argument improves?
These are not merely communication questions.
They are design questions for collective cognition.
The reason across the table
The person who made the original error hears an objection.
It is specific.
The objection points to a hidden assumption.
For a moment, the social discomfort of being wrong is present too.
Then the argument can be examined.
Mercier's programme asks us to take that scene seriously.
Reason may not be most revealing when a mind is sealed inside itself.
It may be most revealing when one person's reasons become material another person can test.
Research references
- Sperber D, Clément F, Heintz C, Mascaro O, Mercier H, Origgi G, Wilson D. Epistemic vigilance. Mind & Language. 2010;25(4):359-393. https://doi.org/10.1111/j.1468-0017.2010.01394.x
- Mercier H, Sperber D. Why do humans reason? Arguments for an argumentative theory. Behavioral and Brain Sciences. 2011;34(2):57-74. https://doi.org/10.1017/S0140525X10000968
- Mercier H. How gullible are we? A review of the evidence from psychology and social science. Review of General Psychology. 2017;21(2):103-122. https://doi.org/10.1037/gpr0000111
- Mercier H, Sperber D. The Enigma of Reason. Harvard University Press; 2017.
- Mercier H. Not Born Yesterday: The Science of Who We Trust and What We Believe. Princeton University Press; 2020.
- Institut Jean Nicod. Hugo Mercier institutional profile. Current role verified September 2026. https://institutnicod.ens.psl.eu/membres/membres-statutaires/mercier-hugo/