In this paper Brian Weatherson argues that we can in principle make substantive discoveries in theoretical philosophy which correct mistakes in our pre-theoretic beliefs about some subject matter. The crux of the argument is that, according to the right (eg the Lewisian) theory about meaning, the referents of our theoretical terms are often stable over small variations in use. In some domain where there are few very natural candidate referents to which we might plausibly be interpreted as referring , even relatively systematic false beliefs can be tolerated before use is changed enough for reference to change. Thus it can be the case that the correct response to an intuitive counterexample is to reject certain kinds of intuitions in order to preserve overall theoretical unity and simplicity.
The example looked at in detail is the Gettier counterexamples to the JTB theory of knowledge. Weatherson isn’t committed to the JTB theory – he just thinks that it isn’t straight-away refuted by the Gettier examples. Maybe the theoretical benefits obtained from a simple and elegant epistemological theory outweighs the need to capture intuitions in certain types of case.
The part of the argument I found most puzzling was the part where Weatherson responds to the objection that the JTB theorist must mean something other than we do by ‘knows’. First he argues for the Lewisian theory of meaning – all well and good. But the Lewisian theory of meaning per se is quite compatible with the view that a tribe speaking a language in which the meaning of ‘knows’ was given by the JTB theory would be speaking a different language from ours. This would depend on the details of how use is balanced against simplicity. Indeed, this view seems intuitively plausible – surely such a wide difference in use would constitute a change in meaning?
In explaining away this intuition, Weatherson goes on to argue that the natural properties in the vicinity of our use of ‘knows’ are extremely scarce. But talk of the distribution of natural properties here presents a puzzle. What does it mean to say ‘there are just no reasonably natural properties in the vicinity of our disposition to use ‘knows’.’? It seems like to make sense of this sort of talk, we need some kind of measure over global-disposition-to-use space with which to compare distances between properties. But nobody has a clue how to explicate such a measure; until we do, talk about scarceness or abundance of natural properties in certain domains has to be taken as merely suggestive and metaphorical.
Luckily, the dialectic does not require talk of distribution of properties to be made precise. The resources needed to stave off the ‘meaning-change’ objection to the JTB theory are more straightforward. We require at the very least that there must be no property which is a) as natural or more natural than the property of having a JTB and b) as well-matched or better-matched than the property of having a JTB to our use of the term ‘knows’. But we also need that there be no property which, while doing less well on one criterion than JTB, does so much better on the other criterion that it is the best referent for ‘knows’.
Weatherson’s argument that there is no such property goes via ‘the failure of the ‘analysis of knowledge’ merry-go-round to stop.’ I take it the thought is that despite our best efforts, we have failed to find any really good candidate analysis. Maybe this does lend support to the idea that there is no extremely simple and natural property which corresponds exactly to our use of the term ‘knows’. But nonetheless there are several properties, like the one picked out by a causal theory of knowledge, which seem to do better at capturing our intuitions about Gettier cases. They go wrong elsewhere; but nothing Weatherson says gives us any reason to think that these post-Gettier ramified theories of knowledge don’t do better overall than the JTB account.
Weatherson’s defence of the JTB theory would be in more trouble still if we could appeal to the Lewisian thought that conjoining natural properties leads to no loss of naturalness. Then theories which consist of the JTB account plus some other necessary conditions given in terms of very natural properties would turn out at least as natural as the JTB account. Such theories also tend to match usage better than the JTB theory (that was the point of introducing the extra necessary conditions, after all) so it looks like they should be strictly preferable. Then we’d have to argue that the reduced simplicity of the more complex theory is such a cost that it counteracts these considerations. We’re then a long way from the ‘failure of the merry-go-round to stop’ line of argument.
However, the closing section of the paper is relevant here. For reasons which seem to me good ones, Weatherson argues against the idea that naturalness is always conferred on a conjunctive property by its conjuncts. If this is right, then we can take the ‘merry-go-round’ idea in a different direction. All of the proposed additions to the JTB theory involve conjoining extra properties onto the property of being a JTB; if conjoining extra properties can make a property less natural (for example in cases where there are multiple ways of satisfying the conjunctive property), then maybe all of the proposed additions to the JTB theory do score significantly lower on naturalness, enough to offset their better match with use. However, it’s unclear if Weatherson would want to go this way, as he doubts that the failure of naturalness to distribute across conjunction generalizes from the JTB property to all the proposed analyses of knowledge.
In this connection, I wasn’t convinced by the thought that the failure of naturalness to transferred to conjunctions from their conjuncts told in favour of the metaphysical thesis that naturalness is primitive. It seems quite compatible with the universals account that the properties of being F and being G are co-intensive with genuine universals, but the property of being-F-and-G is not co-intensive with a genuine universals. All we need to do is to reject conjunctive universals (or maybe just reject their genuineness.) And this is something we might want to do in any case. A similar response is available in the case of the primitive resemblance view. There can easily be cases in which the F’s resemble one another, and the G’s resemble one another, but the F-and-G’s do not all bear primitive resemblance relations to one another. All we need is that there is no single resemblance relation which holds among all the F-and-G’s, only two different resemblance relations, one holding among all the F’s and the other holding among all the G’s. Given that the resemblance relation is primitive, I can’t see any objection to stipulating that it works this way.