What Is the Problem of Induction?

Direct Answer

The problem of induction is one of the most famous and enduring puzzles in philosophy. It was formulated by the Scottish philosopher David Hume in the 1730s, and it asks a deceptively simple question: what justifies our practice of reasoning from what we have observed to what we have not?

Consider how we reason in everyday life. You've seen the sun rise every morning of your life. You confidently expect it to rise tomorrow. Or: every piece of bread you've eaten has nourished you, so you expect the next piece to nourish you too. Or: every time you've dropped an object, it has fallen downward, so you expect the next dropped object to fall. These are all instances of inductive reasoning — reasoning from observed cases to unobserved ones, from the past to the future, from samples to populations.

Hume's disturbing insight was that this kind of reasoning, which we rely on constantly, cannot be rationally justified. The justification would have to be either deductive or inductive. A deductive justification would show that the conclusion follows necessarily from the premises — that is, that it is logically impossible for the future to differ from the past. But this is clearly not the case: there is no logical contradiction in the sun not rising tomorrow, or in bread failing to nourish. The premises of our inductive arguments — the observations we've made so far — do not logically entail the conclusion. So a deductive justification is unavailable.

An inductive justification would appeal to the past success of induction: induction has worked well in the past, so it will probably work well in the future. But this is circular. It uses induction to justify induction. It assumes, in the premise, the very principle it is trying to establish in the conclusion. As Hume pointed out, we cannot justify the principle that the future will resemble the past by appealing to the fact that it has resembled the past in the past — that argument itself depends on the principle we are trying to prove.

This is the logical core of the problem: induction appears to be rationally unjustifiable. And yet we cannot help relying on it. Science, in particular, depends on inductive reasoning — generalizing from experiments to laws, from samples to populations, from past observations to future predictions. If induction cannot be justified, what does that say about the rational status of science?

Historical Context

Hume first presented the problem of induction in A Treatise of Human Nature (1739), Book I, Part III, and later refined it in An Enquiry Concerning Human Understanding (1748), Section IV. The Stanford Encyclopedia of Philosophy notes that Hume's argument emerged from his broader investigation into the origins and limits of human knowledge. Hume was working within the empiricist tradition, which held that all ideas derive from experience. He accepted this framework but drew from it conclusions that shocked his contemporaries.

The intellectual context matters. In the seventeenth and eighteenth centuries, the Scientific Revolution had transformed European thought. Newton's physics, in particular, seemed to demonstrate that careful observation and mathematical reasoning could unlock the secrets of nature. There was enormous confidence in the power of empirical method. Hume's argument struck at the foundations of this confidence. If the very method that science relies on — generalizing from observations — cannot be rationally justified, then the scientific enterprise rests on a foundation that reason cannot secure.

Hume was not the first to notice that induction poses difficulties. Ancient skeptics like Sextus Empiricus had pointed out that past regularities do not guarantee future ones. But Hume gave the argument its sharpest and most influential formulation. His version was distinguished by its precision: he didn't just note that induction might sometimes fail, but argued that there is no rational basis at all for expecting the future to resemble the past. The expectation is not grounded in reason but in habit — a psychological propensity, not a logical entitlement.

Hume's own response to the problem was not despair but a kind of naturalistic resignation. He acknowledged that we cannot help reasoning inductively — it is part of our nature. We are creatures of habit, and the mind naturally moves from the constant conjunction of past events to the expectation of their future conjunction. But this is a fact about our psychology, not a rational justification. Hume's position is sometimes called "mitigated skepticism": we should be skeptical about the power of reason to justify our most fundamental cognitive practices, but we should also recognize that we cannot abandon those practices because they are built into our nature.

Philosophical Perspectives

The problem of induction has generated an enormous literature, and philosophers have proposed a wide range of responses. None has achieved universal acceptance, and the problem remains a live issue in epistemology.

One influential response is Karl Popper's falsificationism. Popper accepted Hume's argument that induction cannot be justified, but he drew a radical conclusion: science doesn't need induction at all. Scientific theories, Popper argued, are not arrived at by generalizing from observations. They are bold conjectures, and the role of observation is not to confirm them but to test them — to try to falsify them. A theory that survives severe testing is not thereby shown to be true or probable, but it has proven its mettle. On this view, science progresses not by inductive confirmation but by conjecture and refutation. Popper's approach sidesteps the problem of induction by denying that science relies on induction in the first place.

Critics of Popper have pointed out that this seems to leave out something important. Even if scientific theories are conjectures, we do seem to regard well-tested theories as more credible than untested ones. And when we apply scientific knowledge — building bridges, designing medicines, launching spacecraft — we are implicitly assuming that the theories will continue to hold, which is an inductive assumption. Popper's account of science may avoid the problem of induction at the level of theory testing, but it doesn't eliminate the need for induction in the practical application of science.

Another response is the pragmatic vindication, associated with Hans Reichenbach. Reichenbach argued that even if induction cannot be justified in the sense Hume demanded, it can be vindicated pragmatically. If there are regularities in nature — if the future does resemble the past — then induction will eventually discover them. If there are no regularities, then no method will work, and induction is no worse than any alternative. So induction is the rational strategy: it is the best bet available, even if its success is not guaranteed. This approach doesn't provide the kind of deductive certainty that Hume demanded, but it offers a practical reason for relying on induction.

A third approach treats the principle of induction — the principle that the future will resemble the past — as a fundamental assumption that cannot be justified but must be accepted. Some philosophers have argued that this principle is a constitutive norm of rational inquiry: to engage in inquiry at all is to presuppose that nature is uniform, that patterns observed will continue. On this view, asking for a justification of induction is like asking for a justification of the laws of logic — it misunderstands the role that the principle plays in our cognitive economy. The principle is not a hypothesis to be tested but a framework within which testing takes place.

Nelson Goodman proposed a refinement of the problem that has become known as the "new riddle of induction." Goodman pointed out that even if we grant that the future will resemble the past, we still face the question of which resemblances matter. He introduced the predicate "grue," which applies to things observed before a certain time if they are green and to things not yet observed if they are blue. Every observed emerald has been green, but it has also been grue (since all observations so far have been before the time in question). So the same evidence supports the prediction that the next emerald will be green and the prediction that it will be blue. The problem is not just that induction can't be justified but that we can't even say which inductive projections are the right ones without smuggling in our prior commitments about which properties are "projectible." Goodman's riddle shows that the problem of induction is even deeper than Hume suggested.

Modern Reflection

The problem of induction continues to resonate in contemporary philosophy, and its implications extend beyond academic epistemology. In the philosophy of science, the problem intersects with debates about scientific realism — the view that scientific theories describe a mind-independent reality. Scientific realists argue that the success of science is best explained by the truth (or approximate truth) of its theories. But this argument relies on induction: we infer from the past success of science to the likely truth of current theories. If induction is unjustifiable, the argument for scientific realism is weakened.

In statistics and probability theory, the problem of induction takes a more formal shape. Bayesian approaches to inference assign probabilities to hypotheses and update them in light of evidence. This framework provides a rigorous method for reasoning under uncertainty, but it does not solve the problem of induction. The Bayesian approach requires prior probabilities, and the choice of priors involves assumptions about the structure of the world — assumptions that are themselves inductive. Bayes's theorem tells you how to update your beliefs given evidence, but it doesn't tell you why you should expect the future to resemble the past.

Machine learning raises the problem in a new form. A machine learning model trained on past data makes predictions about future data. The assumption that the future data will resemble the past data — that the training distribution will match the test distribution — is an inductive assumption. When this assumption fails, the model suffers from "distribution shift," and its predictions become unreliable. The problem of induction is not just a philosophical curiosity but a practical challenge that confronts anyone building predictive systems.

Some contemporary philosophers have argued that the problem of induction, as Hume formulated it, rests on questionable assumptions. It assumes that justification must be either deductive or inductive, and that a circular justification is worthless. But perhaps there are forms of justification that don't fit neatly into these categories. Perhaps a limited, reflective equilibrium — in which our inductive practices and our philosophical principles are adjusted to fit each other — provides a kind of justification that, while not the certainty Hume demanded, is the best we can reasonably expect. This approach doesn't dissolve the problem but suggests that the standard set by Hume — a non-circular, deductive or inductive proof — may be too demanding.

Related Thinkers

The problem of induction is inseparable from the work of David Hume, who first formulated it with full force in the eighteenth century. Hume's skeptical arguments about induction, causation, and the external world represent one of the most thoroughgoing applications of empiricist principles, and they have shaped epistemology ever since.

Karl Popper's falsificationism was developed in direct response to Hume's problem. Popper's The Logic of Scientific Discovery (1934) argued that science progresses through conjecture and refutation, not through inductive generalization, and he credited Hume with showing that induction is unjustifiable. Nelson Goodman's Fact, Fiction, and Forecast (1954) introduced the "new riddle of induction," which has become a central topic in contemporary epistemology. Hans Reichenbach's The Theory of Probability (1949) offered the pragmatic vindication of induction, while W.V.O. Quine's "Two Dogmas of Empiricism" (1951) challenged the analytic-synthetic distinction that underlies the traditional formulation of the problem.

Related Quotes

Hume's own words capture the problem with characteristic clarity. In the Treatise, he writes that "even after we have experience of the operations of cause and effect, our conclusions from that experience are not founded on reasoning, or any process of the understanding." This statement encapsulates the core of the problem: our inductive inferences are not rationally grounded but psychologically driven.

The Stanford Encyclopedia of Philosophy notes that the problem of induction has been called "Hume's problem" and that it remains, nearly three centuries after its formulation, one of the most discussed topics in epistemology — a testament to the depth and durability of Hume's insight.

Further Learning

1. Stanford Encyclopedia of Philosophy, The Problem of Induction.
2. David Hume, A Treatise of Human Nature (Oxford University Press, 2000 [1739]), Book I, Part III.
3. David Hume, An Enquiry Concerning Human Understanding (Oxford University Press, 2007 [1748]), Section IV.
4. Karl Popper, The Logic of Scientific Discovery (Routledge, 2002 [1934]).
5. Nelson Goodman, Fact, Fiction, and Forecast (Harvard University Press, 4th ed., 1983).
6. Hilary Putnam, "The 'Corroboration' of Theories," in Mathematics, Matter and Method (Cambridge University Press, 1975).

To explore related concepts, consider reading about epistemology, skepticism, and the nature of knowledge.