We selected our differential equations or, what comes to the same thing, the principels they translate, because we wished to construct a mathematical representation of a group of phenomena; in seeking to represent these phenomena with the aide of a system of differential equations, we were presupposing from the very start that they were subject to a strict determinism; we were well aware, in fact, taht a phenomenon whose peculiarities did not in the least result from the initial data would rebel at any representation by such a system of equations. We were therefore certain in advance that no place was reserved for free actions in the classification we had arranged. When we note afterwards that a free action cannot be included in our clasification, we should be very naive to be astonished by it and very foolish to conclude that free will is impossible.
Imagine a collector who wishes to arrange sea shells. He takes seven drawers that he marks with seven colors of the spectrum, and you see him putting the red shells in the red drawer, the yellow shells in the yellow drawer, etc. But if a white shell appears, he will not know what to do with it, for he has no white drawer. You would, of course, feel very sorry for his reason if you heard him conclude in his embarrassment that no white shells exist in the world.
The physicist who thinks he can deduce from his theoretical principles the impossibility of free will deserves the same feeling. In manufacturing a classification for all phenomena produced in this world, he forgets the drawer for free actions!
Duhem, Physics of a Believer.
Quite true. By the way, "Physics of a Believer" is one of the most brilliant essays in the history of the philosophy of science. It has often been misunderstood, I think; the misunderstandings aren't all as utterly bizarre as V. I. Lenin's portrayal of Duhem as a Machian Kantian who was almost a dialectical materialist, but they are sometimes fairly serious. The argument toward the end, that physics as a whole, and thermodynamics in particular, suggests a (broadly) Aristotelian metaphysics, is especially misunderstood, despite the fact that Duhem explicitly denies the stronger claims that are sometimes attributed to him. It's the strange thing about Duhem; he's always saddled with interpretations that cut against what he explicitly says. "Physics of a Believer," in fact, was a response to just such an interpretation, the claim that "Duhem's scientific philosophy is that of a believer". As Duhem points out, the only senses in which this is true are (1) that Duhem himself is Catholic; and (2) that if Duhem's account is true, it would be impossible for science to disprove Christian doctrine (or prove it, for that matter).
In any case, the essay is a good source for understanding how it is that Duhem manages to be a positivist and yet also be a scientific realist in the modern sense (which, in fact, he is; for instance, he sharply rejects naive scientific realism about physical theories, but when he rejects the view that theories in physics are true he does so by arguing that they are approximate), or as close to a modern scientific realist as makes no real difference. For a good discussion of some of the issues surrounding the question of whether Duhem is a realist, see Karen Merikangas Darling's paper, Motivational Realism: The Natural Classification for Pierre Duhem (Word document). I don't think her discussion is quite right, because I think the basis for the anti-realist interpretation is somewhat weaker than she makes it out to be. I've already mentioned the crucial point about approximation (Duhem doesn't consider himself a realist because the view that physical theories are approximate rules out what he calls realism), but there are other issues. For example, by 'physical theory' Duhem means mathematics constructed for the purposes of physics; and it isn't clear that someone who tends instrumentalist about the role of mathematics in physics is automatically a scientific anti-realist, particularly when we have this other element involved, namely, the experimental laws that physical theory organizes. The case for Duhem's realism is also slightly stronger than she makes it out to be. To name just two points, Duhem quite explicitly says that science adds to the treasury of common sense, which cannot possibly be interpreted in an anti-realist way, and he insists that physical theory provides a "simplified picture" of what common sense knows to be true and certain, which also sounds realistic (in our sense).
And Darling is quite clearly wrong to say that, since the realist intuitions Duhem insists upon are not scientific or logical "they are unable to ground any pro-realist arguments" (p. 15); how she can regard this as true, given that Duhem is quite clear that the intuitions of common sense are what ultimately ground all arguments, is unclear. Since they are intuitions, we cannot justify them themselves; but since they must be believed by any rational person, this is not a problem. Just as basic intuitions about space don't need further justification, but are what justify our thought about space, so basic intuitions about natural classification and the unity of science don't need further justification but are what justify our (metaphysical) thought about science. This is Duhem's Pascalianism; in the Pascalian aphorism he likes to quote, with regard to the basic principles common sense provides, "We have an impotence to prove that is invincible to any dogmatism and an idea of truth that is invincible to any skepticism."
Darling's claim that Duhem would not consider these on par with scientific theories or logical conclusions will be simply incomprehensible to anyone who actually reads what Duhem says about the intuitions of the heart. And this is not surprising. One of the things we know by intuition of the heart (Duhem is, of course, following Pascal here) is geometry, which is founded on principles that are properly neither scientific nor logical. And perhaps, ultimately, this is where people go wrong with Duhem: they read as anti-realist (in our sense) claims that, in fact, merely mean that realism (in our sense) about physical theory is neither a physical theory nor an experimental law, despite the fact that Duhem's conclusion is undoubtedly right: physical realism is neither a physical theory nor the generalization of a physical experiment. Of course, Duhem thinks this is because nothing in the physicist's method actually justifies scientific realism. Physicists are scientific realists in the way they are realists about the external world: it's something metaphysical that overarches their work in physics, not a conclusion in physics at all. That's not anti-realist either, but perhaps that's where people are misinterpreting him.
But despite the severity of my criticisms, it's a great paper, and worth reading if you are interested in Duhem. You might also like this short paper, which gives a good introduction to Duhem's brilliant recognition that crucial experiments are not possible in physics, and the almost prophetic way in which he argued for this claim.
This post turned out completely different from what I expected to write when I started it; I expected to say something about free will, and ended up talking about scientific realism instead.