Abstract

Van Fraassen's constructive empiricism is presently the most influential and well-developed alternative to scientific realism. In this paper I argue that a reasonable condition on the distinction between belief and agnosticism prevents van Fraassen from claiming that we can be agnostic about what a theory says about unobservable entities while simultaneously accepting that theory. The upshot is that we must find some other way to do justice both to the argument for constructive empiricism and to van Fraassen's cogent criticisms of scientific realism. I suggest that a highly attractive alternative is founded upon semantic anti-realism, and that empiricists should develop this alternative.