The one thing people can be counted on to know about Carnap is that he was against metaphysics.  But what is metaphysics?  According to the Cambridge Dictionary of Philosophy, it is “the philosophical investigation of the nature, constitution, and structure of reality” and is “broader in scope than science, e.g. physics and even cosmology. . ., since one of its traditional concerns is the existence of non-physical entities. . . It is also more fundamental, since it investigates questions science does not address but the answers to which it presupposes” (2nd ed., p. 563).  Ladyman and Ross (LR) in their book Every Thing Must Go contrast their “naturalized” metaphysics to this Cambridge-Dictionary (CD) type, which they refer to as the “metaphysics of domestication,” since it tries to make counter-intuitive scientific knowledge accessible to the crude categories of our inherited vernacular, the ways of thinking that have evolved from the accumulated experience of the species since the origins of language. I agree with them that CD-metaphysics clearly falls afoul of Carnap’s strictures, and am just as baffled as they are that Chalmers and other analytic metaphysicians insist on claiming Carnapian descent.

Their own “scientific,” LR-metaphysics, however, they think immune to Carnap’s critique. In fact, they see LR-metaphysics as Carnapian in spirit.  They are right about this, actually, but how can something be in Carnap’s spirit that ignores his well-known exclusion of metaphysics?

It’s mostly a matter of terminology; what LR call metaphysics, Carnap called a choice among languages.  They are pretty careful not to legitimize what Carnap called “external questions” (in ESO), but on the other hand they — like Quine, Mark Wilson, and many others — refuse to reinterpret metaphysical questions as choices among languages.  I think this unnecessarily weakens their case.

They motivate their own metaphysics by the need for unification.  This goal was of course shared by Carnap and the logical empiricists with their well-known program of the “unity of science” — which they did not consider metaphysical. For Carnap it was a matter of pragmatic language choice. In his view, a universal language that can serve as a single object language for all of science may “be understood as the hypothesis that in future it will become possible to an ever greater extent to derive known extra-physical laws from known physical laws.” (Schilpp vol. p. 883), i.e. as the distant goal of eventually unifying the concepts of the currently disparate scientific object languages.  It obviously makes simpler and more straightforward the task of constructing a single metalanguage for all of knowledge, and was thus pragmatically preferable to a language without these capabilities. In many practical social and engineering contexts, he argued, we need to be able to bring different sciences to bear on the same problem, and we can’t be guided by these various forms of knowledge unless we know how to weigh them against each other and understand their interrelations (e.g. this 1936 lecture [first item], esp. pp. 21-2).

LR’s Carnapian avoidance of external questions (in the ESO sense) becomes particularly clear in their chapter “The World in the Data” in the Scientific Metaphysics book (2012).  Here they train their sights on precisely the kind of residual scientific realism that underlies a good deal of the amateur metaphysics popular among scientists and is articulated more explicitly by Popper (they leave aside the more refined versions of Quine or Mark Wilson).  Their own “structural” realism is not of that kind at all, and in fact would have been classified by Carnap as a proposal for a language of unified science.  “Things” must go, in their view, precisely in the sense that they are grain-level relative, i.e. discourse-relative.  There is no single, stable “ontology” for the whole of knowledge, there are no “things” in any absolute sense, fixed across different levels of discourse.  But once you’ve conceded that there are no fixed objects for words to attach to, and that all talk of “objects” is discourse- (i.e. word-) relative, what objection remains to the transfer of our attention from objects to words?

The most obvious advantage of this transfer is to shift attention from “what is” to “what we want there to be” — i.e. a shift in the direction of regarding a language choice as a matter of engineering, as a practical question (perhaps ultimately an ethical question).  To authors who claim (e.g. Every Thing, pp. 27-8) that their justification of metaphysics is pragmatic in character, rather than “positivistic,” this should be entirely welcome.  Also, to authors who are fully on board with the Enlightenment program, it should be of some value to avoid mortgaging that program to any substantive knowledge claim, and thus to lift the Enlightenment out of its ancient (and, as LR themselves are at pains to show, untenable) rut of claiming to replace traditional authority by the new dogma of authoritative science.  About this, I think we pretty much all agree nowadays — d’Alembert was wrong (and Diderot was on the right track).  What Carnap achieved (as I tried to argue in my book) was precisely to liberate the Enlightenment from this clunky old ball and chain, or at least to sketch out a program of liberation from it.

But he didn’t relapse into any sort of anti-realism, either.  This is a frequent misunderstanding. (I suspect there will be more to say about this.)  But never mind what Carnap says about it (not much); LR themselves deploy a formidable battery of arguments to show precisely how it’s possible to do without objects of any kind as fixed, language-independent furniture of the universe without embracing anti-realism.  They thereby expunge the last vestiges of the metaphysics of substance, and incidentally refute — without mentioning! — Quine’s “Speaking of Objects” (and the central argument of Word and Object), at least insofar as it can be used against Carnap’s replacement of “what is?” by “how should we best articulate what we know?”  So they leave no excuse whatever for realism of any sort at all.  And yet they choose to call what they argue for so effectively a form of realism, indeed a form of metaphysics.  But they never explain what advantages they discern in this curiously reactionary way of putting things.

UPDATE:  I develop these thoughts on Ladyman and Ross further here and here.