It generally seems illegitimate to infer empirical conclusions from moral premises -- imagine, for example, a utilitarian inferring that killing Bob would fail to maximize happiness from their antecedent moral belief that killing is wrong. Or an incompatibilist arguing that determinism must be false because we're morally responsible. There seems something objectionable about the 'direction' of these inferences. (Consequentialists should start with empirical facts and infer moral conclusions; likewise for incompatibilists.)
But I wonder if we can legitimately draw some non-moral conclusions from moral premises. It doesn't seem so bad if the non-moral conclusions are still normative conclusions. I have in mind two examples in particular:
(1) Moral to Rational (all things considered practical reasons). The following seems platitudinous: If you had sufficient reason for acting as you did, then you didn't act wrongly. In other words: moral obligations are rational obligations. But if our views about moral and rational requirements initially diverge, it's an interesting question which way we should revise them: relaxing morality to match our lax view of rationality, or adopting more stringent rational requirements to match our moral convictions. The latter route, though open to question, is at least not obviously illegitimate. (See also the concluding sections of Michael Smith's 'Beyond the Error Theory'.)
(2) Moral to Epistemic. In light of the links between theoretical and practical reason, we might also find our moral beliefs giving rise to certain epistemic commitments. Consider the bridging principle that it's reasonable to act on a reasonable belief. If we think that it could never be reasonable to act so as to gratuitously increase the suffering of innocents, then it seems we should likewise think that it could never be reasonable to believe that the suffering of innocents is intrinsically good. (This is part of what motivates my resistance to Huemer's phenomenal conservatism, for example.)
What do you think? Is something fishy about these sorts of inferences?