Handbook of the History of Logic, Vol. 10, Suppl (C), 2011

ISSN: 1874-5857

doi: 10.1016/B978-0-444-52936-7.50002-1

Hume and the Problem of Induction

Marc Lange

1 Introduction

David Hume first posed what is now commonly called "the problem of induction" (or simply "Hume's problem") in 1739 - in Book 1, Part iii, section 6 ("Of the inference from the impression to the idea") of A Treatise of Human Nature (hereafter T). In 1748, he gave a pithier formulation of the argument in Section iv ("Skeptical doubts concerning the operations of the understanding") of An Enquiry Concerning Human Understanding (E).1 Today Hume's simple but powerful argument has attained the status of a philosophical classic. It is a staple of introductory philosophy courses, annually persuading scores of students of either the enlightening or the corrosive effect of philosophical inquiry - since the argument appears to undermine the credentials of virtually everything that passes for knowledge in their other classes (mathematics notably excepted2).

According to the standard interpretation, Hume's argument purports to show that our opinions regarding what we have not observed have no justification. The obstacle is irremediable; no matter how many further observations we might make, we would still not be entitled to any opinions regarding what we have not observed. Hume's point is not the relatively tame conclusion that we are not warranted in making any predictions with total certainty. Hume's conclusion is more radical: that we are not entitled to any degree of confidence whatever, no matter how slight, in any predictions regarding what we have not observed. We are not justified in having 90% confidence that the sun will rise tomorrow, or in having 70% confidence, or even in being more confident that it will rise than that it will not. There is no opinion (i.e., no degree of confidence) that we are entitled to have regarding a claim concerning what we have not observed. This conclusion "leaves not the lowest degree of evidence in any proposition" that goes beyond our present observations and memory (T, p. 267). Our justified opinions must be "limited to the narrow sphere of our memory and senses" (E, p. 36).

Hume's problem has not gained its notoriety merely from Hume's boldness in denying the epistemic credentials of all of the proudest products of science (and many of the humblest products of common-sense). It takes nothing for someone simply to declare himself unpersuaded by the evidence offered for some prediction. Hume's problem derives its power from the strength of Hume's argument that it is impossible to justify reposing even a modest degree of confidence in any of our predictions. Again, it would be relatively unimpressive to argue that since a variety of past attempts to justify inductive reasoning have failed, there is presumably no way to justify induction and hence, it seems, no warrant for the conclusions that we have called upon induction to support. But Hume's argument is much more ambitious. Hume purports not merely to show that various, apparently promising routes to justifying induction all turn out to fail, but also to exclude every possible route to justifying induction.

Naturally, many philosophers have tried to find a way around Hume's argument - to show that science and common-sense are justified in making predictions inductively. Despite these massive efforts, no response to date has received widespread acceptance. Inductive reasoning remains (in C.D. Broad's famous apothegm) "the glory of Science" and "the scandal of Philosophy" [Broad, 1952, p. 143].

Some philosophers have instead embraced Hume's conclusion but tried to characterize science so that it does not involve our placing various degrees of confidence in various predictions. For example, Karl Popper has suggested that although science refutes general hypotheses by finding them to be logically inconsistent with our observations, science never confirms (even to the smallest degree) the predictive accuracy of a general hypothesis. Science has us make guesses regarding what we have not observed by using those general hypotheses that have survived the most potential refutations despite sticking their necks out furthest, and we make these guesses even though we have no good reason to repose any confidence in their truth:

I think that we shall have to get accustomed to the idea that we must not look upon science as a 'body of knowledge,' but rather as a system of hypotheses; that is to say, a system of guesses or anticipations which in principle cannot be justified, but with which we work as long as they stand up to tests, and of which we are never justified in saying that we know that they are 'true' or 'more or less certain' or even 'probable'. [Popper, 1959, p. 317; cf. Popper, 1972]

However, if we are not justified in having any confidence in a prediction's truth, then it is difficult to see how it could be rational for us to rely upon that prediction [Salmon, 1981]. Admittedly, "that we cannot give a justification . for our guesses does not mean that we may not have guessed the truth." [Popper, 1972, p. 30] But if we have no good reason to be confident that we have guessed the truth, then we would seem no better justified in being guided by the predictions of theories that have passed their tests than in the predictions of theories that have failed their tests. There would seem to be no grounds for calling our guesswork "rational", as Popper does.

Furthermore, Popper's interpretation of science seems inadequate. Some philosophers, such as van Fraassen [1981; 1989], have denied that science confirms the truth of theories about unobservable entities (such as electrons and electric fields), the truth of hypotheses about the laws of nature, or the truth of counterfactual conditionals (which concern what would have happened under circumstances that actually never came to pass - for example, "Had I struck the match, it would have lit"). But these philosophers have argued that these pursuits fall outside of science because we need none of them in order to confirm the empirical adequacy of various theories, a pursuit that is essential to science. So even these interpretations of science are not nearly as austere as Popper's, according to which science fails to accumulate evidence for empirical predictions.

In this essay, I will devote sections 2, 3, and 4 to explaining Hume's argument and offering some criticism of it. In section 6, I will look at the conclusion that Hume himself draws from it. In sections 5 and 7-11, I will review critically a few of the philosophical responses to Hume that are most lively today.3

2 Two Problems of Induction

Although Hume never uses the term "induction" to characterize his topic, today Hume's argument is generally presented as targeting inductive reasoning: any of the kinds of reasoning that we ordinarily take as justifying our opinions regarding what we have not observed. Since Hume's argument exploits the differences between induction and deduction, let's review them.

For the premises of a good deductive argument to be true, but its conclusion to be false, would involve a contradiction. (In philosophical jargon, a good deductive argument is "valid".) For example, a geometric proof is deductive since the truth of its premises ensures the truth of its conclusion by a maximally strong (i.e., "logical") guarantee: on pain of contradiction! That deduction reflects the demands of non-contradiction (a semantic point) has a metaphysical consequence - in particular, a consequence having to do with necessity and possibility. A contradiction could not come to pass; it is impossible. So it is impossible for the premises of a good deductive argument to be true but its conclusion to be false. (That is why deduction's "guarantee" is maximally strong.) It is impossible for a good deductive argument to take us from a truth to a falsehood (i.e., to fail to be "truth-preserving") because such failure would involve a contradiction and contradictions are impossible. A good deductive argument is necessarily truthpreserving.

In contrast, no contradiction is involved in the premises of a good inductive argument being true and its conclusion being false. (Indeed, as we all know, this sort of thing is a familiar fact of life; our expectations, though justly arrived at by reasoning inductively from our observations, sometimes fail to be met.) For example, no matter how many human cells we have examined and found to contain proteins, there would be no contradiction between our evidence and a given as yet unobserved human cell containing no proteins. No contradiction is involved in a good inductive argument's failure to be truth-preserving. Once again, this semantic point has a metaphysical consequence if every necessary truth is such that its falsehood involves a...