8.4 Problems with Induction

Contemporary science is largely an inductive exercise, meaning that general hypotheses are derived from observation and experience. Therefore, the findings of science give us probable, never certain, conclusions, “because our experience is always limited, our inductive generalizations are inevitably fallible…in most cases, there are so many objects within a class that we simply cannot observe all of them. In those numerous cases, our generalizations remain fallible because of the problem of induction” (Barseghyan, op. cit., Ch. 2).

However, induction carries with it some philosophical problems.

For one, induction assumes that observation of a few instances of a class of objects can tell us about the entire class of objects. For example, up until the late 17th century, Europeans assumed that all swans were white, based on their observations in Europe. However, in 1697, Dutch explorers led by Willem de Vlamingh became the first Europeans to see black swans, in Western Australia, thus disproving the original thesis and raising doubts about induction.

For another, as the empiricist David Hume pointed out, the early scientific assumption of the uniformity of nature—that our observations of a small part of the universe could be turned into scientific laws about the entire universe—was philosophically problematic for it was based on a fallacy of composition, assuming something about the whole based upon our experience of only a part.

One classic attempt to solve the problem was to invoke the so-called principle of the uniformity of nature – the idea that in similar circumstances objects of the same class behave similarly. According to this principle, there are certain regularities in nature and identical causes always lead to identical effects. But since nature is uniform, so the argument goes, we don’t need to observe millions of objects of the same class. It would suffice to observe a few objects of the same class, to note what they have in common, and then safely generalize that to all objects of that class. This generalization would hold because nature is uniform, i.e. because identical initial conditions always produce identical effects. According to this line of reasoning, we don’t need to observe all swans in the world to prove that all of them are white. They are all white because the ones we have observed so far have all been white and because the principle of uniformity tells us that all swans are similar as they are all products of the same biological mechanism. By the same token, there is no need to test the law of gravity for every pair of objects in the universe, since according to the principle of uniformity all objects with mass must act similarly. In short, the principle allows to safely extrapolate from past experience to all cases of the same class and to ensure that there won’t be any surprises in the future… (Barseghyan, op. cit. Ch. 2).

Yet another problem with induction is the assumption that the past is a reliable indicator of future events. Bertrand Russell would echo this concern with his analogy of the “Inductivist Turkey.”

This turkey found that, on his first morning at the turkey farm, he was fed at 9 a.m. However, being a good inductivist, he did not jump to conclusions. He waited until he had collected a large number of observations of the fact that he was fed at 9 a.m., and he made these observations under a wide variety of circumstances, on Wednesdays and Thursdays, on warm days and cold days, on rainy days and dry days. Each day, he added another observation statement to his list. Finally, his inductivist conscience was satisfied and he carried out an inductive inference to conclude, “I am always fed at 9 a.m.”. Alas, this conclusion was shown to be false in no uncertain manner when, on Christmas eve, instead of being fed, he had his throat cut. An inductive inference with true premises has led to a false conclusion.  (Chalmers, What is this thing called Science, 2nd edition, University of Queensland Press, St. Lucia, 1982).

Induction, as we saw in the chapter on logic, must always be employed with the caveat that inductive hypotheses will never yield certainty!