Experiments Can Correct Philosophy Without Replacing Philosophical Method

Edmund Gettier had a definition, two small cases, and the manners to fit the whole disturbance into three pages.

He put justified true belief in a room with luck and watched knowledge leave by the fire exit.

The old analysis was clean. Knowledge is belief. It is true. It is justified.

Gettier’s subjects possess all three and arrive at the truth by accident. The definition has collected every required ticket and still boarded the wrong train.

So conceptual analysis begins its little administrative war. Propose necessary and sufficient conditions, find a counterexample, revise, repeat.

Post-Gettier epistemology spent decades adding no-false-lemmas clauses, causal connections, and reliability conditions. A definition does not receive a verdict. It acquires enemies.

Thought experiments have the same temper, with more railways.

Foot’s trolley case asks whether five deaths may be prevented by diverting a runaway train toward one. Thomson’s footbridge case holds the numbers steady and changes the physical relation to harm.

Footbridge, loop track, transplant surgeon. Each variation changes a variable to find out what the original verdict was secretly using.

Reflective equilibrium is more polite about the furniture, but it moves just as much of it.

A theorist starts with considered judgments, builds a principle to explain them, applies it to old and new cases, and revises whichever side now seems less defensible.

Daniels’s wide version adds background theories of persons, society, and rationality, so a theory cannot simply flatter its own first impressions and call the result independence.

This is philosophy’s discipline of revision.

It asks what follows from a concept, a reason, or a judgment that has survived reflection. It does not put its intuitions in a white coat and pretend that settled the matter.

Experimental research arrives at the adjacent office with a random-number generator, a control condition, and a justified suspicion that the room itself is making faces at the participants.

Its question is different. What do people report, and what changes that report?

A true experiment randomly assigns participants to two or more conditions and directly manipulates the variable of interest.

Random assignment does not make groups magically identical. It makes membership independent, in expectation, of what was true before assignment.

That is how it neutralizes confounding, including the confound no one had the foresight to put on the spreadsheet.

The controls have jobs.

Allocation concealment keeps an enroller from steering a participant toward a predictable condition. Differential attrition can undo the balance after the fact.

Single blinding limits placebo and demand-characteristic effects. Double blinding also limits an experimenter’s unconscious ability to deliver a treatment differently, or to score it with a thumb on the scale.

The apparatus is not a truth machine. It is a machine for making certain excuses harder to use.

A survey of responses to a trolley case can therefore reveal something armchair confidence has neglected. Perhaps the verdict moves with wording, order, familiarity, or who has been asked.

That such a response reveals philosophical competence rather than framing sensitivity is speculation until a design earns more. A survey without an assigned exposure remains observational, no matter how many clipboards it has eaten.

This matters because psychology has acquired a useful distrust of its own mirrors.

The WEIRD critique separates a sampling problem from a universal human nature problem. Western, educated, industrialized, rich, democratic participants may be statistical outliers rather than humanity with better campus parking.

The next problem sits at analysis, and it is the one a design can actually be built against.

Decide the covariate after seeing the result. Check the data every ten participants and stop when it looks good. Run several outcome measures and report the one that worked.

Each choice is defensible in isolation. Taken together they can push a nominal five percent false-positive rate past sixty.

The moral is not that every researcher has a villain moustache. It is that a method can give curiosity enough loose screws to assemble a result that should not exist.

Preregistration tightens those screws before the data start flattering anyone.

It records hypotheses, planned sample size, and analysis plan before collection.

It cannot repair a poor measure, restore lost participants, or make a narrow sample universal. It can distinguish a planned test from one selected after inspection, which is the distinction the conventional error rate depends on.

Medical journals made advance trial registration a publication condition two decades ago. Registered Reports move the same commitment into journal review, before results are known.

Notice what none of that machinery does.

An experiment measures responses and their conditions. It does not, by that fact alone, identify the norm that governs a concept.

A randomized wording effect can show that an intuitive verdict is fragile. It cannot decide whether fragility disqualifies that verdict, whether the case has isolated the right feature, or what revision of the principle is justified.

Those are philosophical questions, not leftovers from a study protocol.

The easy reconciliation now presents itself. Randomize the vignettes, preregister the analysis, and armchair philosophy has finally been sent downstairs to collect its things.

The correction actually runs in both directions.

A philosopher who treats an initial intuition as a transparent report from reason has ignored the possibility that it is a product of framing, sampling, demand, or a word doing unannounced work in the sentence.

The experiment does not replace the judgment. It tells the judgment that it has been observed behaving strangely in public.

The continuity worth keeping is elsewhere.

Philosophy is not secretly an unfinished laboratory science, and experimental psychology is not philosophy after it has misplaced the armchair.

Both are disciplines of revision. Gettier’s counterexample, Rawls’s equilibrium, random assignment, blinding, and preregistration each put an initial answer under systematic pressure.

They differ over what may count as a reason to move.

The experiment can show which response occurs and what changes it. Philosophy asks what that response commits anyone to, and whether the commitment survives considered revision.

The laboratory gives a judgment conditions. Philosophy gives it consequences.

Between them, a first answer loses the privilege of arriving first.