Michael Smith takes it as a conceptual truth about fully rational agents that they have no false beliefs. This is a view he picks up from Bernard Williams (Ch. 5 of the Moral Problem; p. 156 actually). This strikes me as too strong and hard to motivate. The reason Smith gives in its defense comes from Williams. Suppose an agent, Wilfrid, has a cup of tonic in front of him and what he thinks is a cup of gin but in reality it is a cup of petrol. He wants to mix the petrol with the tonic and drink it cause he thinks it is gin. According to Smith, it is odd to say he has a reason to do this. In fact, Wilfrid has no reason to do so even though he might think to the contrary. The conclusion that Smith draws from this line of thought is that a fully rational Wilfrid would not have reason to mix the liquids because he would have no false beliefs. The argument he gave doesn’t really support that conclusion. It would support the conclusion that were he fully informed (and rational) he would give up his reason to mix the liquids. Maybe the concept of full rationality is supposed to support the idea that one can, given a set of beliefs, rationate them in such a way that the true ones are preserved and the false ones are weeded out by the powers of logic alone. It isn’t insane to think that a fully rational agent could weed out most false beliefs given some time and access to a fair amount of resources. But, it doesn’t seem supportable to hold that the machinations of pure reason alone rule out the very possibility of false belief. I feel like there are some collateral premises Smith is relying on that I can’t quite get.

In any case, the argument he gives is rather poor. There is a difference between a fully rational agent and a fully informed one. Since we’re stipulating the situation, we are fully informed. Even though I’m not fully rational I can say that Wilfrid shouldn’t mix the liquids, that he has no reason. But, this is just because I know what Wilfrid does not. The concept of rationality does not seem to be pulling any weight in the example. Because of this, it really can’t pull any dialectical weight in the argument. That being said, this particular requirement of rationality doesn’t seem to figure in the rest of Smith’s positive account of non-Humean normative reasons in any big way.

[Addition: Nate continues this discussion here.]