Non-cognitivism and rational inference

Abstract

Non-cognitivism might seem to offer a plausible account of evaluative judgments, at least on the assumption that there is a satisfactory solution to the Frege–Geach problem. However, Cian Dorr has argued that non-cognitivism remains implausible even assuming that the Frege–Geach problem can be solved, on the grounds that non-cognitivism still has to classify some paradigmatically rational inferences as irrational. Dorr’s argument is ingenious and at first glance seems decisive. However, in this paper I will show that Dorr’s argument equivocates between two different notions of evidence, and that once this equivocation is noted there is no reason to doubt that non-cognitivism is consistent with the rationality of such inferences, at least if it is assumed that the Frege–Geach problem can be solved. In particular, I will show that non-cognitivists can endorse the same explanation of the rationality of such inferences that cognitivists should endorse, and that there is thus no need for non-cognitivists to offer any sort of idiosyncratic account of the epistemology of such cases, in contrast to what other commentators on Dorr’s argument have thought.

Appendix: two possible objections

Even with the preceding discussion in hand, there might still be residual worries about the account I have given. In what follows, I try to anticipate and respond to two such worries.

The first worry is that my epistemological account of the relevant cases is insufficiently detailed and thus leaves something to be desired. My main response to this objection is that it is a virtue of my account that it lacks detail, because this means that it does not rely on the idiosyncratic and controversial details of any particular view. However, at the same time, I share the desire for a more detailed explanation of the cases above, which rely on testimony and a priori reasoning in interesting ways. In what follows I will try to bring out some interesting and complicated features of such cases, with the goal of reinforcing the idea that non-cognitivism is consistent with whatever the correct explanation of such cases turns out to be.

To begin, it may be useful to note the existence of an interesting kind of case, not discussed above, in which a person rationally accepts P if and only if Q on the basis of testimony, then comes to rationally accept P, and then rationally infers ~Q. Although such a case can sound absurd in the abstract, it is easy to understand when considered in detail. For example, consider a variant on a case discussed above: Steve asks his father whether he is descended from Icelandic royalty, and his father answers in the form of a riddle by telling him that he is descended from Icelandic royalty if and only if lying is wrong. Steve carefully reflects on the morality of lying, and after recognizing some new considerations relevant to the morality of lying, he comes to accept in a rational way that lying is wrong; however, at the same time and through the same process of reasoning, Steve realizes that his father expected him to reach the opposite conclusion, and thus expected him to infer that he is not descended from Icelandic royalty. In such a case, if Steve knows that his father would never try to mislead him, it might well be rational for Steve to infer that he is not descended from Icelandic royalty. In this case, Steve rationally accepts something of the form P if and only if Q on the basis of testimony, then rationally accepts P, and then rationally infers ~Q. Of course, Steve rejects P if and only if Q as soon as he accepts P, which is how his view remains consistent throughout the story.

This modified version of Steve’s case shows two important things. First, it shows that a naïve view of rational inference and testimony is mistaken according to which one is to simply take the contents of testimony as given, and then uncritically use them “as coherence demands” whenever one adds new beliefs. A more plausible view might appeal to something like evidential probabilities, allowing for cases in which a biconditional of the form P if and only if Q has a high probability on the basis of testimony, but a low conditional probability given some further evidence E, where it is also true that the probability of P and ~Q given E is high. This kind of probabilistic explanation could explain how Steve can be rational throughout the story just considered.¹³ A more detailed account would focus on Steve’s evidence before receiving his father’s testimony, because that background evidence would explain the rationality of the particular way that Steve uses his father’s testimony in subsequent reasoning.

With that in mind, the second thing to note is that Edgar’s case must be understood in a slightly different way from the case involving Steve just considered, since Dorr stipulates that on the cognitivist view Edgar is rational in continuing to accept the relevant conditional even after coming to accept its antecedent. In probabilistic terms, this means that Edgar’s case must be understood as a case in which there is a high probability of his ultimate conclusion C given E*, where E* is whatever total evidence Edgar has after reflecting on the morality of lying. Of course, as noted above, even if non-cognitivism is true Edgar will also have total evidence E* after reflecting on the morality of lying, so no problem arises for non-cognitivism regarding the evidential probabilities in such a case. This illustrates how non-cognitivism is consistent with any plausible probabilistic explanation of such cases.

With this sort of probabilistic structure in mind, we can consider other interesting cases involving testimony. For example, suppose that my religious leader tells me that if lying is wrong, then liars will be punished in the afterlife; then, later, my wife tells me that lying is wrong. Am I justified in concluding that liars will be punished in the afterlife?¹⁴ This is a difficult question to answer. However, a lot will turn on what my total evidence is after hearing my wife’s testimony. If that evidence sufficiently supports the conclusion that liars will be punished in the afterlife, then it would be rational for me to reach that conclusion if I based my conclusion on that evidence in the right kind of way. Importantly, nothing here seems to turn on the question of whether cognitivism or non-cognitivism is true; instead, everything seems to turn on the quality of my evidence, which will be the same regardless of whether cognitivism or non-cognitivism is assumed.

Although probability theory provides an attractive account of many aspects of these cases, it is unclear whether it can account for all aspects of such cases. For example, there are some ‘a priori cases’ in which a subject is not in a position to infer an empirical conclusion until engaging in some a priori reasoning, despite the fact that the a priori reasoning does not seem to provide them with new evidence in any sense. For example, recall the case discussed above in which a detective has interviewed all the witnesses and gathered all the available evidence, but has not yet formed an opinion about who the murderer is. The detective retires to her armchair, closes her eyes, and carefully reflects on her evidence, before finally realizing that the murderer must have been Jones. In this case, it could be argued that the evidential probability of Jones being the murderer is the same both before and after the detective’s reflection, but her reasoning allows her to base her conclusion on her evidence in the right kind of way to ensure a rational inference. Despite such complications regarding ‘the basing relation’, there is no reason to doubt that non-cognitivism is consistent with whatever the correct account of that relation turns out to be.

Even setting those issues aside, there are other cases in which a simple appeal to evidential probability might seem inadequate to explain the rationality of inference. For example, suppose that I accept the conditional if P then Q on the basis of testimony, but then later forget why I initially accepted the conditional; subsequently, I come to believe P in a rational way. In such a case, an inference that Q might arguably be rational, even if I can’t recall why I believe the conditional (Harman 1986, Chapter 4).¹⁵ On the assumption that this kind of case is genuinely possible and that there is a solution to the Frege–Geach problem, non-cognitivism is perfectly consistent with whatever background facts might be claimed to explain the rationality of my inference—for example, current vs. forgotten evidence, coherence conditions, etc.

The upshot of all of this is that although complicated factors might often be at play in the explanation of rational inferences, non-cognitivism is consistent with all of those factors if there is a solution to the Frege–Geach problem, and so there is no special problem for the view on any plausible account of the rationality of inference.

A second worry with my general response is that even if Dorr’s case against non-cognitivism fails, there might be a different problem for non-cognitivism that arises from cases like Edgar’s. For example, it could be objected that if non-cognitivism is true, then there are no rational constraints on the adoption of moral views, and thus according to non-cognitivism one has rational permission for adopting any moral view at any time, and if that’s right, then whenever one rationally accepts a conditional like P1, non-cognitivism implies that it is absurdly easy to put oneself in a position to rationally infer the corresponding empirical consequent (C) by simply changing one’s view so that one accepts the moral antecedent of the conditional (P2). But (such a worry concludes), this is absurd, since it is obviously not so easy to come to be in a position to infer empirical claims, and so non-cognitivism is false.

One problem with this objection is that it assumes that there are no rational constraints on the adoption of moral views if non-cognitivism is true; however, that assumption is false if there is a solution to the Frege–Geach problem, because then it is logically inconsistent and thus rationally impermissible to accept, for example, P2 while continuing to accept P1 and ~C.

Even setting that issue aside, any plausible non-cognitivist view will reject the claim that there are no rational constraints on the adoption of moral views for more general reasons. For example, on Gibbard’s view the adoption of moral views is constrained by norms of rationality that are indistinguishable in practice from the norms of rationality that cognitivists would endorse (Gibbard 1990, pp. 169–171; Gibbard 1993, pp. 324–326). As far as I can tell other non-cognitivists would say approximately the same thing. Thus, the lack of rational constraints that the objection assumes is neither an essential nor popular part of non-cognitivist views. As a result, non-cognitivists can legitimately respond to this objection by denying that their view permits the adoption of moral views in the arbitrary way that the objection assumes.