Now, moral philosophers generally prefer to talk about virtues, or about (specific) duties, rights, and so on, rather than about moral images of the world. There are obvious reasons for this; nevertheless, I think that it is a mistake, and that Kant is profoundly right. What we require in moral philosophy is, first and foremost, a moral image of the world, or rather--since, here again, I am more of a pluralist than Kant--a number of complementary moral images of the world.

The problem with all this--the problem I discussed in the first lecture--is that if the causes/background conditions distinction is fundamentally subjective, not descriptive of the world in itself, then current philosophical explanations of the metaphysical nature of reference are bankrupt.

To require that all of these must be reducible to a single version is to make the mistake of supposing that 'Which are the real objects?' is a question that makes sense independently of our choice of concepts.

If reason is both transcendent and immanent, then philosophy, as culture-bound reflection and argument about eternal questions, is both in time and eternity. We don't have an Archimedean point; we always speak the language of a time and place; but the rightness and wrongness of what we say is not just for a time and a place.

Even though the model referred to satisfies the theory, etc., it is 'unintended'; and we recognize that it is unintended from the description through which it is given (as in the intuitionist case). Models are not lost noumenal waifs looking for someone to name them; they are constructions within our theory itself. and they have names from birth.

To me, believing that some correspondence intrinsically just is reference (not as a result of our operational and theoretical constraints, or our intentions, but as an ultimate metaphysical fact) amounts to a magical theory of reference. Reference itself becomes what Locke called a 'substantial form' (an entity which intrinsically belongs with a certain name) on such a view. Even if one is willing to contemplate such unexplainable metaphysical facts, the epistemological problems that accompany such a metaphysical view seem insuperable. For, assuming a world of mind- independent, discourse-independent entities (this is the presupposition of the view we are discussing), there are, as we have seen, many different 'correspondences' which represent possible or candidate reference relations (infinitely many, in fact, if there are infinitely many things in the universe).

What we have is a device for producing sentences in response to sentences. But none of these sentences is at all connected to the real world. If one coupled two of these machines and let them play the Imitation Game with each other, then they would go on 'fooling' each other forever, even if the rest of the world disappeared! There is no more reason to regard the machine's talk of apples as referring to real world apples than there is to regard the ant's 'drawing' as referring to Winston Churchill.

Even if we consider not words by themselves but rules deciding what words may appropriately be produced in certain contexts - even if we consider, in computer jargon, programs for using words - unless those programs themselves refer to something extra-linguistic there is still no determinate reference that those words possess. This will be a crucial step in the process of reaching the conclusion that the Brain-in-a-Vat Worlders cannot refer to anything external at all (and hence cannot say that they are Brain-in-a-Vat Worlders).

On another possible world or another planet a word might be associated with much the same stereotype and much the same criteria as our term 'water', but it might designate XYZ and not H₂O. At least this could happen in a prescientific era. And it would not follow that XYZ was water; it would only follow that XYZ could look like water, taste like water, etc. What 'water' refers to depends on the actual nature of the paradigms, not just on what is in our heads.

In short, analytic statements are statements which we all accept and for which we do not give reasons. This is what we mean when we say that they are true by 'implicit convention'. The problem is then to distinguish them from other statements that we accept, and do not give reasons for, in particular from the statements that we unreasonably accept. To resolve this difficulty, we have to point out some of the crucial distinguishing features of analytic statements (e.g. the fact that the subject concept is not a law-cluster concept), and we have to connect these features with what, in the preceding section, was called the 'rationale' of the analytic-synthetic distinction. Having done this, we can see that the acceptance of analytic statements is rational, even though there are no reasons (in the sense of' evidence') in connection with them.