Epistemology

Problem of Induction

How can past observations justify future predictions? — Hume's puzzle

The problem of induction asks: how can we justify generalizing from past observations to future predictions? David Hume (1748): induction relies on assumption that nature is uniform — but this assumption itself can only be justified by induction (circular). Sun has risen every day; will it tomorrow? No deductive guarantee. Just past pattern. Yet science depends on this. Major problem in philosophy. Responses: pragmatic (Hume's own — habit of mind), Popper's falsificationism, Bayesian probability, naturalism. No fully satisfying solution. Foundational question for science and reasoning.

  • Posed byDavid Hume (1748)
  • QuestionHow justify generalizing from past to future?
  • Hume's answerHabit/custom; no rational justification
  • Popper's responseFalsificationism (test theories by trying to disprove)
  • Bayesian responseProbabilistic reasoning
  • StatusUnsolved foundational problem

Interactive visualization

Press play, or step through manually. The visualization is yours to drive — try it before reading on.

Open visualization fullscreen ↗

Watch the 60-second explainer

A condensed visual walkthrough — narrated, captioned, under a minute.

Why induction problem matters

  • Science. Foundation of scientific method.
  • Epistemology. Limits of empirical knowledge.
  • Machine learning. Generalization problem.
  • Statistics. Inference foundations.
  • Daily life. Every prediction.
  • Decision theory. Choosing under uncertainty.
  • Critical thinking. Logical foundations.

Common misconceptions

  • Solved by science. Active philosophical issue.
  • Statistics solves it. Statistics uses induction; doesn't justify it.
  • Practical issue only. Foundational philosophical problem.
  • Hume undermined science. Showed practical, not theoretical issue.
  • Falsificationism bypasses. Doesn't fully escape.
  • Easy to solve. Centuries of debate.

Frequently asked questions

What's the problem of induction?

Hume's puzzle. We use induction (past observations → general law → future predictions). Sun has risen every observed day; therefore will rise tomorrow. But: what justifies this inference? Not deduction (no contradiction in sun not rising). Empirical generalization itself uses induction. Circular. Hume: no rational justification for induction.

Why is it a problem?

Science depends on induction. Theories generalize from observations. Predictions made for future. If induction unjustified: science's foundations shaky. We've never directly observed laws of physics; just instances. Why should they hold tomorrow? No logical proof. Yet we can't function without inductive reasoning.

What's Hume's response?

Pragmatic, not rational. We do induction by psychological habit/custom. Constant conjunction creates expectation. Not rationally justified — just how we're built. Skeptical conclusion: no rational basis for induction; we shouldn't claim to know future. But: we can't help expecting based on past. Practical necessity overrides theoretical doubt.

What's Popper's falsificationism?

Karl Popper (20th c.). Science doesn't actually use induction. Scientists propose theories; test by trying to falsify. Theories that survive tests are tentatively accepted. Never proven; only disproven. Avoids induction problem: doesn't claim to confirm theories. Critic: still uses induction (past test results → future expectations); just not directly.

What's the Bayesian response?

Probabilistic reasoning. Prior belief × evidence → posterior belief (probability). Allows non-zero probability without certainty. Past observations increase posterior probability of generalization. Doesn't fully solve: priors arbitrary; learning rules use induction. But: useful framework. Modern science increasingly Bayesian.

What's the new riddle of induction?

Nelson Goodman (1955). "Grue paradox." Define "grue" = green if observed before time t; blue otherwise. All observed emeralds are grue. Yet we project "green" not "grue" to future. Why? Both equally consistent with evidence. Suggests: induction depends on which predicates we use. No purely logical solution.

How is this practical?

Bears on: (1) Scientific reliability — theories tested can fail. (2) Decision-making under uncertainty. (3) Machine learning — generalizing from training data. (4) Epistemology — what can we know? (5) Daily life — every prediction is inductive. Practically: we accept some inductive risk; refine via evidence; revise when wrong.