The justification of science appears at first glance to be the generalisation of experience. I heat metal x and see that it expands, I heat metal y and see that it expands, I heat metal z and see that it expands, and so on, such that it seems natural that I can claim that all metals expand when I heat them. When you hear then talk, then you would think that most scientists think this is what a scientific argument is, and most would also think this is what we might mean by scientific objectivity. There are, however, two questions we might ask of them. First of all, does the inductive method really produce knowledge in the way they think it does, and secondly even if it did is this how science itself operates in its own history? I actually think the second question is more important than the first. The first question is about scientific method, the second is more about what scientists actually do, and not what they say they do. It is a matter of pragmatics, rather than the logical definition of a method in the abstract.
Let us take the first question first, because it is the more traditional problem of induction, and has its canonical form in the argument of Hume. To understand his problem with induction we first of all need to understand, even if in the most basic way, his epistemology. For Hume, there are two kinds of propositions: relations of ideas, and matters of facts. In the first relation, the truth of our ideas is confined by our ideas alone. Thus if you understand the concept ‘bachelor’ you know the idea ‘unmarried man’ is contained within it. When it comes to matters of fact, however, we have to go beyond our concepts to experience. They tell us something new about the world and not just the ideas we already know. A matter of fact would be that Paris is the capital of France, or metals expand when heated. Of course when you know the idea then you know what is contained in it, but to obtain the idea you first of all have to get the knowledge. You only know that Paris is in France, if you have knowledge of basic geography. You only know that metals expand when heated, if you know metallurgy.
There can be false relations of ideas as there can be false matters of fact. Thus if you think that a whale is a fish, then you have made an error about a relation of ideas (you don’t know that a whale is a mammal), and if you think that Plato died in 399 BC, then you have made an error at the level of facts (Ladyman 2002, p.32). Relations of ideas can be proved true by deduction since the negation is a contraction. Basically relations of ideas are tautologies, you cannot assert that Peter is not a bachelor at the same time as asserting that he isn’t married as well, since being unmarried and being a bachelor are one the same thing. On the other hand, matters of fact cannot be proved by logic alone, but can only be derived from experience and their contradiction is not a fallacy. If I say that Everest is the tallest mountain on Earth, none of the terms have a logical relation to one another, so I could assume that there is taller mountain. I would have to experience the different tall mountains on Earth to know which one was tallest or not (Ladyman 2002, p.33). For this reason Hume was extremely sceptical about what one could claim to know deductively. All that one could claim are logical relations between concepts that we already known (whose origin anyway would be the senses). What we cannot claim is to produce new knowledge about the world simply through examining our concepts (as theology and metaphysics is wont to do in his opinion).
These distinctions seem very straight forward and at first glance appear to back up the ‘inductivist’ view of science. The problem for Hume, however, is whether the idea that matters of fact could have the same necessary conclusions as relations of ideas, as the idea of expanding metals as a universal law implies. The key to this problem for Hume is whether I can assert that what happens in the past is a necessary certainty for what will happen in the future. I have experienced the fact that the sun rises every morning. Does this give me the right to say it will rise again tomorrow, when I haven’t actually experience this dawn yet? If it does rise then I will be certain, and in terms of the past, I know that it did rise, but now can I know that I will rise again tomorrow? It is perfectly possible, even if it were unexpected, that the sun might not rise.
Induction for Hume is based upon causal arguments. Our only knowledge of cause and effect is through experience itself because there is no logical reason why any causal relation should hold or not hold. I know matches cause fires, because I know that from experience, not because matches logically contain fire. Just as we can only infer future behaviour of the world from the actual experience of the world, then we can only understand the category of causality from experience. In other words without experience we would not have the concept of causality as a generality. If I always experience the dawn as the rising of the sun then I conjoin this events. If A always follows B, then I will say that A causes B. This because I believe that the future always follows the same path as the past. So that if A happens, then B will happen. Linked to conjunction is contiguity and precedence. Contiguity means that B follows A in time and space, and precedence is that the effect is always after the cause. (the flame is after the lighted match and not before). It is because of conjunction, contiguity, and precedence, that we feel that we have good reason to say that A causes B, or that the sun will rise tomorrow. Hume assertion, however, is that this can never be a necessary reason, as is suggested by generalisation of a universal law, however compelling I feel this causality to be.
Take the example of billiard balls, which seems the most basic relation of causality. The ball X hits the ball Y and causes it to move. But what do we mean by that? Do we mean that the ball X makes the ball Y move or that it produces its movement? We think there is a necessary connection between the two events. X moving and Y moving. What we experience is conjunction, contiguity and precedence, what we do not experience is some mysterious ‘necessary connection’. What we see is ball X and ball Y, what we do not see is some other third thing (like an invisible connection, indeed what we do not see is causality). What does it add to our explanation of the events, even if we were to add this mysterious cause. Wouldn’t the ball X and the ball Y just move in exactly the same way?
The point for Hume is just because two events have always in the past be conjoined, does not mean that we can be universally certain that they will always do so. The conclusion of inductive argument could be false but that would never make it invalid (indeed it might make it more interesting, if the sun did not rise the next day), but this is never the case with a deductive argument if the premises are true, then the conclusion is necessarily true. What underpins the inductive generalisation is the belief that nature is well ordered spatially and temporally, that what happens many times will happen again in the same way. But that is just an assumption. Why must the future always be the same as the past and it certainly is not a logical contradiction if it were not.
Now of course we make these kind of inferences all the time, and Hume accepts that. I probably would not be able to live if I really though the sun would not rise tomorrow every time I went to bed. But this uniformity is a result of our psychology (perhaps it is an evolutionary trait) rather than reason or logic. We find regularity in nature because our habitual associations of events, and not because these events are necessarily connected.
There is no doubt that Hume’s problem is very profound and does make us look at induction more critically, but we might think that the idea that science itself is inductive in the simple way that ‘inductivism’ implies is too simplistic. So the problem is not with induction as such, but how we are using it. It is important to note that this is a very different critique from the methodological one. In the first case, we investigate the method of induction, and like Hume say that is flawed, or might even argue that Hume’s own account of induction is not a correct description of induction. Whereas in the historical account of science, we are arguing whether the description of method is actually how scientists themselves work. One is a description of the content of scientific knowledge, the other is a description of the activity of scientists themselves. Do scientists really act the way that Hume’s example suggests they do? This is a completely different way of doing philosophy of science. For it does not first of all describe a method of doing science and then apply it to scientists, rather it examines what scientists do and from that derives the method. We shall see that this way of understanding science is going to be very important to Kuhn.
Why might we think that scientists do not use the inductive method in the way that induction has been described so far? Take the example of Newton’s Principia (Ladyman 2002, pp.55–6). Newton presents in this work the three laws of motion and the law of gravity. From these laws in explains natural phenomena like planetary motion. He says that he has inferred these laws through induction from observation. Now it is French philosopher of science Duhem that points out that there is a problem with Newton’s explanation. The data he is using is Kepler’s. His data proves that the planet will move in circles, whereas Newton’s in ellipses. This means that he could not have inferred gravity from Kepler’s data, rather he already the hypothesis of the law of gravity to interpret Kepler’s data. Even Kepler’s theory could not have be derived from observation, because he took his data from Brahe, but could only organise it by already assuming that planets moved in circles, a hypothesis he didn’t receive from data, but from the mystical Pythagorean tradition.
So there are two reasons why we might be sceptical of the simple inductive explanation of science. One is methodological through the problem of induction (though we might come up with a better inductive method to solve this), and the other is historical, that science does not work in the way that theory of induction describes. I think the latter is the more serious issue than the former. For in the end science is what scientists do, and not what philosophers might idealise that they do. If you like, the problem of induction is a problem for philosophers. It isn’t one for scientists. They work in a very different way indeed.
Ladyman, J., 2002. Understanding Philosophy of Science, London; New York: Routledge.
 A group of philosophers from the 20th century called logical positivists also liked this distinction, and differentiated mathematical and logical truths, on the one hand, and science on the other. Anything that didn’t fit this schema was said to be nonsense or meaningless. I am not sure that Hume would have gone that far.
 Kant’s argument against Hume is that causality is not merely a habit of the mind but a necessary part of our representation of the world. It would not make sense without it.
 This is what Ladyman does when he lists all the different ways in which we might counter Hume, the most telling being induction as the ‘best explanation’ (Ladyman 2002, pp.46–7).
Typo: “Relations of ideas can be proved true by deduction since the negation is a contraction” (contradiction).
When investigating the degree to which science is inductive, it probably makes sense to differentiate induction in the narrow sense from what many writers mean by induction, including any explanatory inference. Gilbert Harman has interesting thoughts on whether inference to the best explanation is more fundamental than induction in the narrow sense.