Test the evidence

Hume Made Induction a Habit Before Bias Research Measured It

Hume's problem of induction and the heuristics programme meet in the mind's tendency to move beyond what the evidence strictly warrants. The experiment does not dissolve Hume's problem; it gives it a nervous system.

One system

philosophers

The other

heuristics and biases

Hume's account of custom and the availability-style substitution documented in heuristics research touch at inference being guided by learned expectation rather than deductive necessity.

The ordinary leap from past regularity to future expectation has both a philosophical gap and a psychological history.

David Hume begins with a piece of ordinary furniture: tomorrow.

He asks what licenses the move from what has happened to what will. The sun has risen so far. Therefore it will rise tomorrow.

Nobody makes this argument while looking for the kettle, which is perhaps why it gets away with so much.

Hume divides the intellectual house into relations of ideas and matters of fact. The former can be demonstrated. The latter concern a world that might have been otherwise.

Causal claims live in the second room, where certainty has been asked to leave its coat at the door.

Induction is the move from observed instances to a general claim or an unobserved one.

Every raven examined so far is black, therefore all ravens are black. Past sunrises, therefore tomorrow’s sunrise.

The premises may be true and the conclusion still false. That is not a defect in the sample size. It is the kind of inference this is.

Hume’s problem is cruelly compact.

Induction presupposes the uniformity principle: that unobserved cases resemble observed ones, that the future resembles the past.

Demonstrative reasoning cannot establish it, because nature changing course is no contradiction. Probable reasoning cannot establish it either, because an inductive proof of induction has already helped itself to the principle under examination.

The future is not logically obliged to continue its subscription.

Hume’s own explanation is not that people therefore stop expecting anything. They do not.

Repetition produces custom, or habit. Having seen one event regularly follow another, a mind comes to anticipate the second.

Causation, in the relevant human sense, is not a little chain visible between events. It is an expectation with a history.

Two centuries later, Tversky and Kahneman put that history into a laboratory with consonants.

Their letter-frequency study asked whether a letter such as K occurs more often as the first letter of English words or as the third.

Participants tended to choose first position. Third-letter occurrences are roughly twice as frequent. First-letter examples are easier to retrieve.

Ease of recall quietly becomes a judgment of frequency.

That is the availability heuristic: judging frequency or probability by the ease with which examples arrive, rather than by actual frequency.

The mind has not consulted a census of English words. It has checked what was easiest to fetch from its own shelves and called the result evidence.

There are worse librarians, but few faster ones.

The same programme supplied representativeness.

In the Tom W. study, one group estimated base rates for graduate students across nine fields, another rated a personality sketch’s resemblance to a typical student in each field, and a third predicted Tom’s field.

The predictions followed resemblance more closely than the actual population figures. A vivid fit to a prototype crowded out the quieter arithmetic of a base rate.

There is a genuine Humean atmosphere here.

Neither availability nor representativeness yields a deductive guarantee. Each shows a mind passing from what is present, whether retrieved instances or a familiar-looking sketch, to a wider judgment.

The evidence comes with a little psychological weather attached.

So the laboratory looks as though it has finished the philosopher’s sentence.

Hume says expectation is habit. Heuristics research measures habits of inference. The ghost has received a clipboard, a dependent variable, and a rather unflattering acronym.

It has not.

The experiment can show that a person substitutes retrieval ease for frequency, or resemblance for a base rate. Hume asks whether any move from observed regularity to an unobserved case is justified.

A causal history explains an expectation. It does not provide the missing premise.

The description of how an inference happens and the justification of whether it should happen are different jobs, however often they share an office.

It would be speculation to say that availability is the psychological mechanism Hume had in mind.

His custom is an account of expectation formed from observed regularity. Availability is a named effect with a specific task and a specific failure mode. The resemblance is structural, not a time-travelling diagnosis.

The catalogue itself insists on this sort of restraint.

Anchoring is relatively robust when an experimenter provides an anchor and participants make a self-generated adjustment. In the wheel-of-fortune study, a rigged wheel stopping at a low or high number moved estimates of an unrelated quantity by twenty points.

But genuinely random anchors, from a die roll or a spreadsheet function, have failed in some replications. A wheel with suspicious stage lighting is not every wheel.

Other effects are sturdier in one formulation than another.

In the Linda problem, a large majority ranked a conjunction as more probable than one of its constituents. Asked in a frequency format, the error falls sharply.

The effect has not evaporated. Neither does its first famous number get to represent the whole family.

The lesson generalizes past this catalogue.

A named effect is a claim with a design, a scope, and boundary conditions attached. It is not a portable law, and its most-cited version is usually its most flattering one.

What holds across the two accounts is not that Hume anticipated cognitive science, or that a laboratory has finally repaired a crack in eighteenth-century epistemology.

It is that the ordinary leap from past regularity to future expectation has two descriptions that need each other without becoming one another.

Philosophy finds the gap. No deduction and no non-circular induction makes the future answerable to the past.

Psychology finds the history. Retrieval, resemblance, adjustment, and custom move a finite creature through a world that has not offered guarantees.

The world repeats itself often enough to teach expectation. The mind learns the rhythm.

Logic still asks why it gets to call that rhythm a promise.

The tempting reading, refused

Cognitive-bias research has experimentally solved the problem of induction.

Showing how people infer cannot justify the inference they make.

What you can take forward

Hume distinguishes relations of ideas from matters of fact, and induction concerns claims that move beyond what has been observed.
Hume's problem of induction argues that the uniformity principle is neither deductively demonstrable nor inductively justifiable without circularity.
Hume treats causal expectation as custom or habit formed through observed regularity rather than as knowledge of necessary connection.
The availability heuristic substitutes ease of retrieval for actual frequency, as in the 1973 letter-frequency study.
The representativeness heuristic substitutes resemblance to a prototype for attention to base rates, as in Tversky and Kahneman's 1973 Tom W. study.
Replication status and effect size are effect-specific: anchoring has robust boundary-conditioned support, while some famous effects shrink or fail under later tests.
The Open Science Collaboration's 2015 replication estimate is contested in meaning and is not a universal rate for psychology.

Go deeper

What is Hume's problem of induction?: Hume's problem of induction asks what justifies moving from observed cases to an unobserved case or a general rule. Past sunrises support an expectation of another sunrise, but they do not deductively guarantee it. The needed assumption is that unobserved cases will resemble observed ones. Hume argues that deduction cannot establish that assumption and induction cannot establish it without assuming the same point in advance.
What are relations of ideas and matters of fact in Hume?: Hume distinguishes relations of ideas from matters of fact. Relations of ideas concern claims whose denial produces a contradiction, such as the relations established by mathematics or definitions. Matters of fact concern the world and could have been otherwise. Expectations about causes, future events, and unobserved cases belong to matters of fact, so they cannot receive the same necessity as a deductive proof.
Why does Hume say induction cannot prove itself?: Induction assumes that patterns observed so far are a guide to cases not yet observed. An inductive argument for that assumption would say that induction worked in the past and will therefore work in the future. That reasoning already relies on the future resembling the past. The argument is circular, so past success can explain confidence in induction without deductively or non-circularly justifying it.
What does Hume mean by custom or habit?: For Hume, repeated observation of one kind of event following another forms an expectation that the sequence will continue. Custom or habit is the psychological disposition produced by that repetition. It explains why a person expects an effect after its familiar cause even though no necessary connection is perceived. The account explains expectation as a fact about human minds rather than a logical proof that nature must continue uniformly.
What is the availability heuristic?: The availability heuristic is judging an event's frequency or probability by how easily examples come to mind rather than by its actual frequency. In Tversky and Kahneman's 1973 letter-frequency study, participants judged a consonant such as K to occur more often as the first letter of English words than as the third. Third-letter occurrences are roughly twice as frequent, but first-letter examples are easier to retrieve.
What is the representativeness heuristic?: The representativeness heuristic judges the probability that an instance belongs to a category by how closely it resembles a mental prototype rather than by the category's actual base rate. In Tversky and Kahneman's 1973 Tom W. study, predictions about a student's field tracked similarity ratings more closely than the base-rate information supplied to another group. The effect is classified as a cold bias because the founding task gave participants no personal stake.
Does availability prove that Hume was describing a cognitive bias?: No. Hume's account says that custom formed through observed regularity guides causal expectation. Availability research shows a particular experimental pattern in which ease of retrieval substitutes for actual frequency. Both concern inference influenced by psychological history rather than deductive necessity, but identifying availability as Hume's direct psychological mechanism is speculation. The philosophical problem remains whether an inference is justified, not merely how it occurs.
Did cognitive-bias research solve the problem of induction?: No. Cognitive-bias research can measure how people make probability judgments, retrieve examples, or use resemblance and base rates. Hume's problem asks what justifies projecting observed regularities into unobserved cases. A descriptive finding about the mind cannot by itself supply that justification. It may explain why an expectation occurs, but explanation of an inference and justification of an inference answer different questions.
How well does anchoring replicate?: Anchoring has comparatively strong replication support when an experimenter provides an anchor and participants adjust away from it themselves. Tversky and Kahneman's 1974 wheel study used values of 10 and 65 and produced mean estimates of 25% and 45% for African countries in the United Nations. However, truly random anchors from an Excel generator or die roll have failed in some replications, so robustness has a documented boundary.
Why does the Linda problem change when it is asked differently?: The conjunction fallacy occurs when a conjunction is judged more probable than one of its parts. In the original 1983 Linda probability-ranking format, roughly 85% of participants made that error. When the same structure is expressed as a frequency estimate, the error rate falls to roughly 20%. The change shows that an effect's size and interpretation depend on the task format, not only on its famous name.
What did the 2015 psychology replication project actually find?: The Open Science Collaboration's 2015 Reproducibility Project attempted to replicate 100 studies from three high-ranking psychology journals. Of the 97 originally significant studies, 36% replicated at p < .05, and successful replications averaged roughly half the original effect size. The headline estimate is disputed: Gilbert, King, Pettigrew, and Wilson challenged its interpretation in 2016, and Anderson and colleagues replied that multiple readings remain possible.
Why is a bias's fame not enough to establish it?: A famous effect can have a robust replication record, a smaller effect under revised formats, or a failed later replication. The Many Labs Replication Project tested 13 classic and contemporary effects across 36 independent samples in 2014, and found that replicability tracked the effect being tested rather than the laboratory or setting testing it: ten replicated consistently, one was weak and inconsistent, and two failed. Effect size, task design, and the exact replication project therefore matter more than textbook familiarity when evaluating a named bias.