The philosophical "problem of induction" has had such a powerful influence that it is worth spending a moment on what the problem means, and also what it does not imply about our examination of inductive arguments. For induction has posed more philosophical problems than deduction has. If David Hume bequeathed us a useful criterion for telling inductive arguments from deductive ones (see "Commonly Asked Questions"), he also stirred up trouble for induction by posing the question of its legitimacy.

The problem can be put simply: Inductive arguments are never valid. Suppose you want to argue that the sun will rise tomorrow. Hardly any conclusion could have more observations behind it: Think of the number of mornings on which you've known the sun to rise; then think of the uncountably greater number of mornings on which someone or other knew the sun to rise; consider the absence of any mornings without sun. Even this much evidence, Hume points out, produces no valid argument. To get a valid argument from the weight of past experience, you also need the premise that the future will resemble the past. Then you get:

The sun has always risen on past mornings.

The future will resemble the past.

Therefore, the sun will rise tomorrow morning.

A valid argument. And sound? That depends on the grounds for the second premise. How do you know that the future will resemble the past? Well, every former future has turned out to resemble its preceding past. Today was the future yesterday; anyone who predicted yesterday that the future would resemble the past would have turned out to be correct.

But these futures you're talking about are futures that already came to be; in other words they are past futures. How can we rely on them to show that the future from our present point of view will continue to resemble what we now call the past? We again need the premise that the future will resemble the past. So inductive arguments always presuppose a nondeductive jump from established evidence to as-yet-unknown events.

There is no denying that inductive arguments are not deductive ones. It would follow that no inductive arguments can be good ones either, if tests of validity were the only ways of telling good arguments from bad. Induction would come off as a poor relation to deduction.

But in fact, as Chapters 10 and 11 show, we possess many criteria for distinguishing good inductive arguments—that is, strong ones—from bad ones. Sample size and similarity to target are both reliable, from the point of view of guiding our judgment, and clearly applicable, from the point of view of practical assessments. Samples can be counted; people can reasonably discuss their resemblance to the target class. Arguing inductively does not leave us in the chaos of arbitrary experience just because it relies on other than deductive principles.

So keep the philosophical problem of induction separate from the practical problem of judging any particular inductive argument. Philosophers have continued to explore induction since Hume; some of them are still skeptical about induction. But no theoretical skepticism of this sort affects our response to inductive arguments.

Hence you may reply to an inductive argument in all sorts of ways, even challenge it in a number of ways, except by means of blanket doubt about induction. When you don't believe a new study about caffeine and pancreatic cancer, or about your mayor's unpopularity among the middle class, then by all means ask about the size of the sample studied, and its representativeness, and the questions the poll had asked, and the meaning of its error margin. Do not retort, "Studies don't mean anything," or "That's just a generalization."