A Working "Theory" of Rationality

It is tempting to think irrationality is primarily a failure of logic: the irrational person commits fallacies, ignores evidence, or draws invalid conclusions. Conspiracy thinking complicates that picture. A conspiracy theory can contain valid inferences, cite genuine data, and even remain internally coherent while something has still gone badly wrong with the inquiry that produced it.

That suggests that rationality must consist in more than possessing arguments or updating beliefs consistently. It must also concern the kind of inquiry we are engaged in, the standards governing disagreement, the way evidence is identified and structured, the comparison of rival explanations, the intellectual habits of the inquirer, and the information environment in which inquiry takes place.

My aim in this post is to assemble those pieces into a working model of the ideal inquirer. I am especially interested in how that model can help us understand conspiracy thinking. Rather than asking only whether a conspiratorial argument is formally valid, I want to ask a thicker set of questions: What is the dialogue trying to accomplish? What makes something evidence? How was that evidence obtained? How should it change our confidence? How does the whole explanatory package compare with its rivals? What virtues does responsible inquiry require? And what happens when our epistemic environment is actively exploiting our cognitive limitations?

The basic proposal: rational inquiry is a layered process. We first identify the goal and rules of the inquiry; compare competing explanatory models; determine how evidence bears on those models; update our confidence where quantification is appropriate; discipline the process with epistemic virtues; and remain alert to environmental forces that can distort every stage.

1. What are we trying to do? Rationality has a dialogical context

Before asking whether an argument is valid, we should ask what kind of argumentative activity is taking place. Arguments do not normally appear in a vacuum. They are moves within conversations in which people have different commitments, goals, burdens, and expectations. A move that is perfectly reasonable in one context may be irrelevant or even obstructive in another.

The pragma-dialectical approach developed by Frans van Eemeren and Rob Grootendorst is helpful here because it treats argumentation as a regulated critical discussion aimed at resolving a difference of opinion. Its ideal model moves through four stages. First, a disagreement becomes explicit. Next, the parties establish common ground, procedural rules, and burdens of proof. Then arguments and criticisms are exchanged. Finally, the discussion reaches a conclusion: a standpoint has either survived the criticism or it has not. The point is not merely to persuade an audience but to create conditions under which a disagreement can be rationally tested.

This changes how we should think about fallacies. A fallacy is not only an invalid inference in the abstract. It can also be a move that prevents a critical discussion from doing its job. A straw man attacks a position that was never advanced. Burden shifting allows someone to avoid defending their own claim. Ambiguous language prevents the parties from determining what is actually being asserted. Even a formally valid inference can be dialectically defective if it evades commitments or blocks appropriate criticism.

Rules of engagement

The full pragma-dialectical model contains ten rules, but for present purposes several are especially important:

Rule	What it requires
Freedom rule	No one should be prevented from advancing or challenging a standpoint through silencing, intimidation, or similar obstruction.
Burden of proof rule	A party who advances a standpoint must defend it when appropriately challenged rather than shifting the burden to the opponent.
Standpoint rule	Criticism must address the standpoint actually advanced rather than a distorted substitute.
Relevance rule	A defense should answer the criticism that was actually raised rather than diverting to an unrelated consideration.
Unexpressed premise rule	When an argument depends on an implicit premise, that premise must be open to clarification and challenge.
Starting point rule	Participants should not opportunistically deny commitments that were already accepted as common ground.
Validity rule	Arguments should employ reasoning that meets the appropriate standards of inferential validity.
Argument scheme rule	Appeals to analogy, authority, causal reasoning, and other schemes must be legitimate and answerable to the appropriate critical questions.
Closure rule	A successfully defended standpoint should be accepted for the purposes of the discussion; a refuted one should be withdrawn or revised.
Usage rule	Participants should avoid unclear, ambiguous, or manipulative language that prevents rational resolution.

Douglas Walton extends this contextual insight by distinguishing different types of dialogue. A persuasion dialogue aims to resolve a difference of opinion. An inquiry dialogue aims to establish what is true. Deliberation concerns what should be done. Negotiation seeks an acceptable settlement of conflicting interests. Information-seeking transfers knowledge from someone who has it to someone who needs it. Pedagogical dialogue aims at learning. Eristic dialogue, by contrast, is oriented toward quarrel, emotional release, or victory rather than rational resolution.

Dialogue type	Primary goal	Typical rationality constraints
Persuasion / critical discussion	Resolve a difference of opinion through reasoned argument.	Clear commitments, proper burdens of proof, accurate representation of opposing standpoints, and arguments open to critical questioning.
Inquiry	Establish what is the case.	Gather and weigh evidence impartially, test hypotheses, avoid suppression and cherry-picking, and keep contributions relevant to the inquiry question.
Deliberation	Choose a prudent course of action.	Consider consequences, trade-offs, alternatives, values, and shared goals rather than collapsing the choice into a false dilemma.
Negotiation	Reach a mutually acceptable settlement of conflicting interests.	State positions and concessions honestly, make genuine offers, avoid coercion and deception, and search for feasible compromise.
Information seeking	Transfer needed knowledge from one party to another.	Ask relevant and clear questions; answer accurately, completely, and without evasion.
Pedagogical	Facilitate learning and understanding.	Explanations should suit the learner, questions should be encouraged, misunderstandings corrected, and progress built cumulatively.
Eristic / quarrel	Vent, confront, or score points rather than resolve the underlying issue.	Even here, turn-taking, relevance, and avoidance of outright deception can distinguish disagreement from sheer incoherence.

These differences matter. If I treat an inquiry as though it were a quarrel, scoring points may replace truth-seeking. If I treat negotiation as though it were pure inquiry, I may ignore the legitimate role of interests and compromise. Rationality is therefore partly context-sensitive: a participant must understand the purpose of the dialogue and adopt norms that make that purpose possible.

What this adds to the model

Rationality is not exhausted by valid inference. It also requires a truth-enabling or resolution-enabling conversational structure: clear goals, appropriate burdens, relevant criticism, fair engagement, and a genuine possibility of revision.

2. What are we evaluating? Compare theories, not stockpiles of arguments

Once the purpose and rules of inquiry are clear, a second question appears: what is the proper unit of evaluation? It is tempting to think that rational disagreement proceeds by exchanging isolated arguments for P and not-P until one side accumulates enough victories. Graham Oppy’s work on derivations challenges that picture.

Consider a dispute between PRO and CON about some contested claim. If PRO derives the claim from premises that CON already accepts, the derivation has genuine dialectical force: it may reveal that CON's commitments jointly lead somewhere CON did not expect. A reductio can be especially powerful because it exposes an inconsistency internal to a position. But if PRO's derivation relies on premises CON simply rejects, then the argument may do little more than restate the disagreement in a longer form.

In the kinds of deep disputes Oppy has in view, this pushes us away from a fixation on stand-alone derivations and toward theory comparison. Instead of asking only whether one can derive a controversial conclusion, construct the best theory containing the claim and the best rival theory containing its negation. Test each for internal incoherence. If both survive, compare them using broader theoretical virtues: simplicity, explanatory power, fruitfulness, predictive success, fit with the data, and coherence with the rest of what we have reason to accept.

This is particularly useful for conspiracy thinking. Conspiracy claims rarely live in isolation. They are usually embedded within a broader interpretive framework about institutions, testimony, expertise, secrecy, incentives, and who can be trusted. Refuting one argument may therefore leave the worldview largely untouched; another argument is simply recruited from the stockpile. The deeper question is whether the conspiratorial package explains the total body of evidence better than its strongest non-conspiratorial rival.

This also helps explain why rational disagreement can persist. People may disagree not because one side has failed to perform a simple inference, but because the plausibility of premises and the interpretation of evidence depend on different background commitments. Rational persuasion then cannot always consist in producing one more argument. It may require comparing the rival frameworks that make those arguments appear plausible in the first place.

What this adds to the model

The possession of arguments is not enough. Under deep disagreement, rational assessment often requires comparing entire explanatory packages and asking which one best survives internal criticism while accounting for the total evidence.

3. What counts as evidence? Structure it before quantifying it

Theory comparison immediately creates another problem. We often say that the better theory is the one best supported by the evidence, but that formula hides an important question: what makes some datum evidence for a hypothesis in the first place?

Evidence is rarely a raw fact that arrives already labeled with its significance. A fingerprint does not directly state, “this person committed the crime.” It may support a proposition about presence at a scene; that may support a proposition about opportunity; and that may contribute, alongside many other considerations, to a conclusion about guilt. The evidential force comes from a network of intermediate inferences and defeasible generalizations.

The work of John Henry Wigmore, and later developments in evidence analysis by writers such as William Twining, Terence Anderson, and David Schum, is useful because it makes this structure explicit. A Wigmore-style chart maps an ultimate proposition through intermediate propositions to particular items of evidence. It makes visible which claims support others, where the inferential chain is weak, which sources need credibility assessments, and where an undercutter could break the connection.

That perspective highlights three features of evidential rationality that are easy to overlook:

Evidence requires interpretation. Raw observations must be translated into propositions whose relevance to a hypothesis can be argued. Interpretation is therefore epistemically upstream of belief updating.
Evidential support is often indirect and multi-layered. Evidence can support intermediate claims, which then support larger claims. The strength of the final conclusion depends on the integrity of the chain.
The collection process matters. Rational belief depends not only on what evidence is present, but on how it entered the evidential record: who gathered it, what was excluded, whether selection was biased, and whether the sources are trustworthy.

This matters enormously under deep disagreement because evidence is theory-laden. A scientific consensus, for example, may be interpreted by one person as evidence for a scientific hypothesis and by another as evidence that institutions are corrupt. The datum is the same; the evidential role assigned to it is different because the background theory differs. We therefore cannot always begin with “Here is the evidence.” Sometimes the dispute is precisely over why the datum should count as evidence for one hypothesis rather than another.

Component	Question for the inquirer
Evidential relevance	Why should this datum count for or against the claim?
Evidential structure	How do the pieces combine—do they corroborate, conflict, depend on one another, or attack one another?
Inferential chains	What intermediate propositions connect the observation to the final hypothesis?
Procedural access	How was the evidence gathered, filtered, measured, and selected?
Interpretive frames	What background assumptions make this item intelligible as evidence?

What this adds to the model

A rational inquirer does not treat evidence as a pile of inputs. Evidence must be identified, interpreted, connected to hypotheses, and audited for provenance before its force can be responsibly assessed.

4. How should belief change? Bayesian updating in its proper role

Once we have done the interpretive work of establishing what a piece of evidence is and how it relates to a hypothesis, we can ask a more familiar question: how should that evidence change our confidence?

Bayesian epistemology provides a formal model of belief updating. In its most familiar form, Bayes' theorem is:

P(H \mid E) = \frac{P(E \mid H) \cdot P(H)}{P(E)}

Here, H is a hypothesis, E is evidence, P(H) represents prior confidence in the hypothesis, and P(H | E) represents confidence after taking the evidence into account. The framework captures something genuinely important about rationality: beliefs should be responsive to new information, and the impact of that information depends both on what we already believe and on how likely the evidence would be under competing hypotheses.

But Bayesian updating does not by itself solve the problems raised in the previous section. The formalism needs an E, a prior, and likelihoods. Those inputs already reflect judgments about relevance, background knowledge, model structure, and interpretation. If two people assign radically different meanings to the same observation, merely telling them to “update on the evidence” does not resolve the dispute.

Priors create a related difficulty. They encode an agent's starting commitments, and deep disagreement may partly consist in radically different starting points. A person can also manipulate priors or likelihoods in ways that make a favored conclusion resilient while remaining formally coherent. Formal consistency is therefore a constraint on rationality, but it cannot be the whole of rationality.

The most useful role for Bayes in the present framework is therefore local rather than total. Where a sub-inference is sufficiently well specified—for example, how a test result should change our confidence given competing assumptions about reliability or contamination—Bayesian reasoning can discipline the update. At the worldview level, however, we still need the broader work of evidential interpretation and theory comparison.

What this adds to the model

Rationality requires evidence-responsiveness. Bayesian reasoning is a powerful tool for disciplined updating within a sufficiently specified model, but it cannot by itself determine what counts as evidence, which background model is correct, or whether the starting assumptions were responsibly formed.

5. What must the inquirer be like? Epistemic virtues as quality control

At this point the emerging picture still leaves something out. A method can be perfectly respectable on paper and badly abused in practice. Someone can build an impressive evidence map from cherry-picked sources. Someone can choose priors that insulate a preferred conclusion. Someone can compare rival worldviews while being systematically unfair to every alternative. The quality of inquiry therefore depends partly on the character and habits of the person conducting it.

This is where virtue epistemology becomes useful. Reliabilist versions focus on stable faculties or capacities that reliably produce true beliefs. Responsibilist versions emphasize intellectual character traits such as humility, open-mindedness, fairness, courage, attentiveness, and conscientiousness. For my purposes, the important point is shared: rationality is partly process-sensitive. How a belief was formed matters, not only whether a person can later construct a coherent defense of it.

This insight connects directly with the earlier layers of the model. Oppy's worldview comparison presupposes enough intellectual honesty to acknowledge weaknesses in one's own view and enough fairness to represent rivals at their best. Bayesian reasoning presupposes that priors and likelihoods are not being opportunistically tuned to reach a desired conclusion. Evidence analysis presupposes care in collection, patience in tracing inferential chains, and integrity in refusing to suppress inconvenient data.

If those virtues are absent, formally respectable techniques can become tools of motivated reasoning. A biased agent can be probabilistically coherent. A selective evidential chart can look rigorous. A theory comparison can be arranged so that one side always receives charitable assumptions while the other is held to impossible standards. The virtues function as a kind of quality-control system for the procedures themselves.

The following intellectual standards are useful as a reminder of what this process sensitivity looks like in practice:

Virtues especially relevant to inquiry

Intellectual humility: awareness of one's cognitive limits and the revisability of one's conclusions.
Reflective awareness: monitoring how one's own reasoning, emotions, and incentives may be shaping judgment.
Intellectual courage: willingness to follow evidence or argument where it threatens a preferred belief.
Open-mindedness: serious engagement with plausible alternatives rather than ritual acknowledgment of them.
Conscientiousness: care in searching, checking, collecting, and documenting evidence.
Fairness: applying comparable standards to one's own view and to competing views.

What this adds to the model

Rational procedures need virtuous operators. The inquirer's habits determine whether evidence is gathered honestly, alternatives are treated fairly, and formal tools are used to discover rather than rationalize.

6. What if the environment is adversarial? Rationality is socially situated

Virtue epistemology can make it sound as though epistemic failure is primarily an individual defect: if people were simply more careful, humble, or open-minded, better beliefs would follow. C. Thi Nguyen's work on hostile epistemology complicates that picture. Human beings are finite. We cannot investigate everything ourselves, so we must use heuristics, rely on trust, take cognitive shortcuts, and defer to other people and institutions. These are not optional flaws that can simply be removed. They are conditions of being limited knowers.

But unavoidable shortcuts create exploitable surfaces. An information environment can be designed—or can evolve—in ways that reward exactly the cues we normally use to navigate complexity. Propagandists, trolls, institutions, media systems, social structures, cultural practices, algorithms, recommendation engines, and metrics can all shape what becomes salient, whom we trust, and when inquiry feels complete.

This matters for conspiracy thinking because conspiratorial beliefs often develop inside information environments that do more than provide false propositions. They can organize trust and distrust, establish which sources are permitted, provide emotionally compelling interpretive templates, and immunize the worldview against outside correction. The problem is therefore partly structural, not merely a collection of individual mistakes.

Nguyen's broader category of the epistemic environment therefore includes not only individual speakers but social and technical structures: other people such as propagandists or trolls; echo chambers and polarized communities; cultural practices that gamify attention; institutions such as media, political organizations, and platforms; and technologies such as algorithms, recommendation systems, and metrics. Each can alter what information reaches us and how cognitively expensive it is to resist the default interpretation.

Exploit pattern	Why it matters
Clarity as seduction	Simple, vivid explanations can create a feeling of understanding that outruns the underlying evidence and prematurely terminates inquiry.
Trust as vulnerability	Because we must depend on experts and institutions, credibility can be spoofed through fake expertise, manufactured consensus, or captured trust networks.
Transparency traps	More information is not always more understanding. Radical transparency can produce overwhelming data dumps or force specialists into misleading simplifications for lay audiences.
Echo chambers	Some communities do not merely omit outside information; they actively restructure trust so that external criticism is discounted before it is heard.
Quantification and gamification	Scores, rankings, likes, and KPIs provide hyper-clear signals whose ease of interpretation can exceed their real epistemic value.

The seductions of clarity

Nguyen's discussion of clarity provides a particularly useful example. Clarity is normally an epistemic virtue: a clear explanation can illuminate structure, expose ambiguity, and make reasoning easier to inspect. But clarity can also be seductive. Some subjects are intrinsically complicated, uncertain, or messy. An explanation can feel satisfying precisely because it removes that complexity rather than because it represents the world accurately.

Conspiracy narratives are often powerful in this way. They can turn a confusing collection of events into a single coherent story with identifiable agents, motives, and causes. That feeling of suddenly “getting it” can act as a thought-terminating signal. The same vulnerability appears in other forms: scores, rankings, dashboards, metrics, and institutional categories can offer hyper-clear representations whose cognitive appeal exceeds their epistemic value.

The lesson is not to reject clarity. It is to treat unusually effortless clarity as something that sometimes deserves more scrutiny rather than less. Rational inquiry needs counter-heuristics for moments when a story is too comprehensive, too smooth, or too emotionally satisfying.

Trust is another unavoidable vulnerability

We cannot personally verify every claim on which our beliefs depend. We rely on experts, institutions, journalists, technical systems, witnesses, and communities. That dependence makes trust an essential part of rationality, but it also creates an opportunity for manipulation. Fake expertise, manufactured consensus, selective transparency, and echo chambers can all exploit the fact that finite agents must defer somewhere.

An especially important warning sign is pre-emption: a worldview that explains in advance why every outside source is corrupt, deceived, or hostile. Such a system can convert counter-evidence into confirmation. The more outsiders disagree, the more successfully the conspiracy is supposedly concealing itself. At that point, the worldview is not merely interpreting evidence; it is restructuring the conditions under which anything could ever count against it.

Designing trust more defensively

If dependence on others is unavoidable, the answer cannot be radical self-reliance. The better response is to make trust more discriminating and auditable. Several practices follow from the hostile-epistemology perspective:

Independence triangulation: look for convergence across sources or institutions with genuinely different incentives and information pathways, rather than counting repetition inside one ecosystem as independent confirmation.
Pre-emption detection: notice when a worldview dismisses all external sources in advance and thereby converts possible counter-evidence into further confirmation.
Bounded transparency: prefer auditable methods, data access, documented procedures, and other inspectable artifacts over public-relations simplifications that merely feel transparent.
Communicative-facility caution: remember that people who speak fluently in an institution's preferred simplified language may appear clearer or more expert partly because the system is optimized to recognize that form of communication.

Error metabolism

Because limited agents will inevitably make mistakes, a realistic epistemology should not demand errorlessness. It should value systems that notice and repair errors. Nguyen, drawing on ideas about error management, emphasizes the importance of environments and practices that make correction possible. For an individual or institution, that means creating mechanisms for reopening settled conclusions, recording failed predictions, seeking disconfirmation, running adversarial tests, and distinguishing the ability to detect error from the ability to avoid all error in advance.

What this adds to the model

Rationality is situated. A good inquirer must understand not only their own biases but also the architecture of the environment supplying their evidence, shaping their trust, rewarding certain narratives, and determining which mistakes are easy or difficult to correct.

7. The ideal inquirer: a unified procedure

We can now put the pieces together. The goal is not a universal “rationality filter” that mechanically labels beliefs rational or irrational. It is a procedure for conducting inquiry in a way that remains responsive to evidence, alternatives, criticism, and the limitations of real human knowers.

Stage	Core question	Failure mode to watch for
1. Identify the dialogue	What are we trying to accomplish, and what rules make that goal possible?	Treating inquiry as performance, quarrel, persuasion-at-all-costs, or identity defense.
2. State rival models	What are the strongest competing explanatory packages, not merely isolated claims?	Stockpiling arguments for one view while neglecting comparative performance.
3. Map the evidence	How does each datum connect through intermediate propositions to the hypotheses?	Treating data as self-interpreting or ignoring weak inferential links.
4. Audit provenance	Who produced the evidence, how was it collected, what is missing, and what selection pressures were present?	Cherry-picking, source laundering, filtered datasets, unreliable collection, or selective transparency.
5. Update locally	Given a sufficiently specified sub-inference, how should this evidence change confidence?	Arbitrary priors, manipulated likelihoods, or using formal coherence as a substitute for substantive justification.
6. Exercise epistemic virtues	Am I applying the same standards to my view and its rivals, and am I genuinely willing to revise?	Dogmatism, motivated reasoning, overconfidence, negligence, or asymmetrical skepticism.
7. Audit the environment	How are platforms, institutions, communities, incentives, and trust networks shaping what I see and believe?	Echo chambers, manufactured credibility, clarity traps, gamified attention, or pre-emptive distrust of outsiders.
8. Compare and reopen	Which model best handles the total evidence, and what would cause me to revisit this conclusion?	Premature closure or a worldview with no imaginable condition for revision.

A few operational guardrails

Clarity tripwire: when an explanation suddenly makes everything easy to categorize, increase scrutiny rather than declaring the inquiry finished.
Disconfirmation rule: state in advance what observations would reduce confidence in the favored view.
Pre-emption audit: ask whether the theory contains a built-in reason to dismiss every possible outside source.
Independence triangulation: prefer convergence among sources with genuinely different incentives and information pathways over repetition within one ecosystem.
Provenance log: record what was searched, what was excluded, what was not found, and how important evidence entered the inquiry.
Rival-model test: periodically steel-man the strongest alternative and ask which facts it explains more naturally.
Reopening mechanism: treat major conclusions as revisable and establish conditions under which they should be reconsidered.

The contrast with conspiracy thinking

This framework helps clarify what can go wrong in conspiratorial inquiry without reducing the problem to “bad logic.” A conspiracy theorist may reason validly from their premises. They may update their confidence when new information appears. They may even produce a large body of evidence. The failure may instead occur earlier or elsewhere: the evidential base may be selectively constructed; outside testimony may be pre-emptively discredited; rival explanations may not receive comparable treatment; the theory may convert anomalies into confirmation; a seductive narrative may produce premature closure; or the surrounding community may punish revision while rewarding increasingly elaborate defenses.

The ideal inquirer is distinguished less by possession of a particular formal technique than by the way these layers interact. They seek a dialogue capable of resolution. They compare whole theories rather than counting arguments. They do not treat evidence as given, but investigate how it was produced and why it is relevant. They update confidence when the inferential structure is sufficiently clear. They cultivate habits that make self-correction possible. And they remain aware that the environment itself can manipulate attention, trust, and the feeling of understanding.

So this is the theory of rationality I find most useful: not rationality as mere formal validity, and not rationality as the absence of named fallacies, but rationality as a disciplined, socially situated, self-correcting practice of inquiry. Its central demand is not that we become perfectly unbiased or computationally ideal agents. It is that we build procedures, habits, and environments that make error easier to discover, alternatives harder to ignore, and belief revision genuinely possible.

Search This Blog

CriticalThinkingAcademy