Let's begin with an old example. In De Rerum Naturua, Lucretius presented a thought experiment to show that space is infinite. We imagine ourselves near the alleged edge of space; we throw a spear; we see it either sail through the ‘edge’ or we see it bounce back. In the former case the ‘edge’ isn't the edge, after all. In the latter case, there must be something beyond the ‘edge’ that repelled the spear. Either way, the ‘edge’ isn't really an edge of space, after all. So space is infinite.

According to logical positivism, so the story goes, metaphysical questions are meaningless, since they do not admit of empirical confirmation or refutation. However, the logical positivists did not in fact reject as meaningless all questions about for example, the structure of space and time. Rather, key figures such as Reichenbach and Schlick believed that scientific theories often presupposed a conceptual framework that was not itself empirically testable, but which was required for the theory as a whole to be empirically testable. For example, the theory of Special Relativity relies upon the simultaneity convention introduced by Einstein that assumes that the one-way speed of light is the same in all directions of space. Hence, the logical positivists accepted an a priori component to physical theories. However, they denied that this a priori component is necessarily true. Whereas for Kant, metaphysics is the a priori science of the necessary structure of rational thought about reality (rather than about things in themselves), the logical positivists were forced by the history of science to accept that the a priori structure of theories could change. Hence, they defended a notion of what Michael Friedman (1999) calls the ‘relativised’ or the ‘constitutive’ a priori. Carnap and Reichenbach held that such an a priori framework was conventional, whereas Schlick seems to have been more of a realist and held that the overall relative simplicity of different theories could count as evidence for their truth, notwithstanding the fact that some parts of them are not directly testable. All this is part of the story of how the verification principle came to be abandoned, and how logical positivism transmuted into logical empiricism.

The rosy dawn of my title refers to that optimistic time when the logical concept of a natural kind originated in Victorian England. The scholastic twilight refers to the present state of affairs. I devote more space to dawn than twilight, because one basic problem was there from the start, and by now those origins have been forgotten. Philosophers have learned many things about classification from the tradition of natural kinds. But now it is in disarray and is unlikely to be put back together again. My argument is less founded on objections to the numerous theories now in circulation, than on the sheer proliferation of incompatible views. There no longer exists what Bertrand Russell called ‘the doctrine of natural kinds’—one doctrine. Instead we have a slew of distinct analyses directed at unrelated projects.

There is a well-worn example—well worn in some circles, anyway—of what you might think of as racist bridges. Robert Moses, the celebrated New York architect, designed unusually low bridges over the roads from Long Island to Jones Beach. The bridges were designed in this way so as to prevent the poor and predominantly black locals from travelling to the beach by bus—the affluent, white car owners can slip under them with ease. The bridges prevent certain members of the community from enjoying the beach as surely as a phalanx of clansmen. Perhaps the strongest moral drawn from the story is the claim that the bridges are political. The objects themselves are imbued with dubious values.

In a short 1997 book entitled Simplicity as Evidence of Truth, the Oxford philosopher Richard Swinburne has put forward the following thesis summarily: ‘… for theories (of equal scope) rendering equally probable our observational data (which, for brevity I shall call equally good at “predicting”), fitting equally well with background knowledge, the simplest is most probably true’.