symposium.gif - 4.0 K
Michael A. Smith
Australian National University
"Evaluation, Uncertainty, and Motivation"
Ethical Theory and Moral Practice, Vol. 5 2002, pp. 305-320.
Bear4.gif - 2.6 K

John Tresan, University of Florida

Posted 01/15/03

"Commentary on Michael Smith’s “Evaluation, Uncertainty, and Motivation”"

Michael Smith makes a strong case that noncognitivists have trouble accounting for various features of evaluative judgement. I don’t think, however, he has quite shown it. Here I’ll explain some resources noncognitivists might have which Smith doesn’t take seriously enough. I’m not at sure that, in the end, noncognitivists can succeed, but Smith would have to do more to show that they can’t.

Roughly speaking, noncognitivists say that evaluative judgements are really desires; e.g., the judgement that pleasure is good is really a desire that pleasure occurs. (Of course, noncognitivist accounts get much more sophisticated than that; for now let’s follow Smith and stick to this simple version.) Noncognitivism is thought to be good at accounting for the desire-like features of evaluative judgements - the way in which evaluative judgements are like desires - in, e.g., the explanation of action. It is thought to have trouble with the belief-like features of evaluative judgements. Smith describes three features of evaluative judgements and argues that noncognitivism is “unable to accommodate” all of them. Interestingly, he argues that certain of the desire-like features of evaluative judgements can be understood only in terms of those features. So if noncognitivists are unable to accommodate the features, or can’t do so in a way which also accommodates the desire-like features explicable only in terms of them, then it turns out that noncognitivists have trouble with what is thought to be their home territory: the desire-like features of evaluative judgements.

The three features Smith describes he calls Certitude, Robustness, and Importance. Subjects are more or less confident of their evaluative judgements; e.g. one’s judgement that pleasure is intrinsically good might be quite certain, and one’s judgement that autonomy is intrinsically good somewhat less certain. The degree of confidence its subject has in an evaluative judgement is that judgement’s Certitude. Evaluative judgements are more or less stable under the impact of incoming information and reflection; e.g. one’s judgement that autonomy is intrinsically good might be disposed to endure under whatever new information one receives and new reflection one engages in, whereas one’s judgement that pleasure is intrinsically good might be a passing fancy, ready to be dropped upon a moment’s further reflection. The stability of an evaluative judgement under such circumstances is its Robustness. Finally, evaluative judgements are judgements of degrees of value. One might think autonomy (or a certain amount of it) more valuable than pleasure (or a certain amount of it). The degree of value an evaluative judgement ascribes is its Importance.

Smith argues that the role of evaluative judgements in the explanation of action can be explicated at least partly in terms of these features. Roughly speaking, the idea is that insofar as they are rational, agents’ motivations will be proportional to the degree of their evaluative judgements Certitude, Robustness, and Importance. I’ll return to the role of these features in the explanation of action in section 3. Sections 1 and 2 discuss the question whether noncognitivists can accommodate the features at all. Section 4 returns to a problem for the accommodation of Certitude suggested in section 2.

1. Robustness is Robustness, Importance is Strength.

Two of the features seem to me fairly easy for noncognitivists to accommodate, Robustness and Importance. Indeed, Smith himself seems to believe this is as well.1 Robustness - clearly a feature of beliefs and desires alike - can be understood by noncognitivists as … Robustness. Desires are more or less stable in the face of incoming information and reflection, and there’s no reason to noncognitivists to look for anything further when trying to account for the Robustness of evaluative judgements. Mary’s judgement that pleasure is intrinsically good is Robust to the extent that it is stable under the impact of Mary’s new information and reflection.

Importance looks explicable in terms of strength of desire. If Mary regards (a certain amount of) autonomy as better than (a certain amount of) pleasure, the natural noncognitivist account seems to be that the desire which constitudes her judgement that autonomy is good is stronger than her desire which constitutes her judgement that pleasure is good. To judge that something is very important is to have a strong desire for it, to judge that something is mildly important is to have a weak desire for it, and so on for every point in between.

2. Certitude in terms of betting dispositions?

Certitude looks more tricky. Prima facie it can’t be explicated in terms of either Robustness or Strength, since those appear capable of variation in the face of invariant Certitude (and vice-versa). Two people may judge that pleasure is good, and those judgements may be equally confident, but one may be very Robust and the other not Robust at all. Likewise, one may be very confident that pleasure is a little good; another may be equally confident that pleasure is very good. So an account of Certitude in terms of Robustness or Importance doesn’t look in the offing.

So where might noncognitivists look for an account of Certitude? In this section I’ll suggest that they look precisely where cognitivists might look. Smith himself gestures at an account of Certitude - in terms of betting dispositions - and it looks like noncognitivists might be poised simply to adopt that account. Assuming they can handle a problem that at first might have seemed quite far afield - the problem of molecular jdugements with evaluative components - they seem able to do so. Of course, they will then adopt whatever problems accrue to the betting dispositions account in general. But Smith himself seems to like that account2, so he at least might feel constrained from calling upon any such objection. On the betting dispositions account we look at matrices like the following to determine Certitude, where the judgement that p is the judgement whose Certitude is at stake, the subject must choose between x and y, and O1-O4 are outcomes.

  p ~p
x   O1 O2
y   O3 O4

If the subject would opt for x under such circumstances, then the extent to which O1 serves the subject’s aims better than O3, and O4 better than O2, is the extent to which the subject is confident in p. For instance, if I’m quite sure that it will rain tomorrow, then I’ll be willing to opt for x even if O1 = $1 deposited in my bank account, O4 is $50, and O2 and O3 represent no change.3

What are the prospects for noncognitivism applying this account to evaluative judgements like that pleasure is intrinsically good? As Smith himself points out4, it is somewhat hard to imagine circumstances in which this sort of bet would take place. But that doesn’t seem an in-principle objection to doing so, and in any event it is equally an objection to cognitivist employments of betting dispositions.

A more serious problem arises if we consider that what is at stake here are subjects’ betting dispositions in the event of their making certain judgements. That is, we consider what the subject would choose in the event that she judges that if she chooses x and p then O1, and if she chooses x and ~p then O2, and so on. So far as assessing your confidence goes, it doesn’t matter what actually will happen if you make certain choices. What matters is what you judge will happen. Suppose you can either take the bridge or the valley road, and if the bridge doesn’t collapse you’ll find at the other side a million dollars. If you opt for the valley road that doesn’t display a low confidence in the bridge if you didn’t realize the million was there. And if you judged the million was there, and still opted for the valley road, that does display a low confidence even if the judgement was false and there was no million.

So noncognitivists can adopt the betting dispositions account only if they can make sense of judgements like if I choose x and pleasure is intrinsically good then I’ll get a million (or whatever). This is, of course, a classic problem for noncognitivists, variously known as the problem of embedded contexts, molecular judgements with evaluative components, or the Frege/Geach problem. The particular version of the problem relevant here is whether they can make sense of judgements like that if sadistic pleasures are intrinsically good, I’ll be surprised, where the overall judgement is a conditional and the antecedent is evaluative. Obviously, no desire for sadistic pleasures is expressed, so it’s not clear what noncognitivists are going to say about such judgements.

Smith might suggest that if calling upon the betting dispositions account saddles noncognitivists with this problem, so much the worse for noncognitivists’ ability to do so. But that would be a mistake. We already know, independently of the question of Certitude, that noncognitivists have to explain molecular judgements with evaluative components. Moreover, noncognitivists have already gone to great pains to provide such explanations. Hence, if Smith’s objection to noncognitivism’s ability to handle Certitude amounts to no more than an objection to their ability to handle such judgements, the objection is neither new nor as simple as Smith makes out. He would have to examine and find wanting the relevant work of Allan Gibbard and Simon Blackburn, et al.5

3. Certitude and practical rationality.

Smith argues that Certitude plays a role in the explanation of action, and that accounts of it thus have to be evaluated with respect to whether they can explain that role. In this section I question just what Smith has in mind by this role, and whether his account of it is defensible. Despite this, I argue that there is indeed a potential problem here for the noncognitivist account of Certitude described in section 2, a problem I grapple with in the next section.

Smith says that “in so far as they are rational, agents are more strongly motivated to do that about which they are more confident, as between options which they judge to be equally desirable”.6 If Mary can either bring about a certain amount of autonomy or a certain amount of pleasure, and she judges those amounts to be equally good - but she is more confident of her judgement that the pleasure is good at all - then insofar as she’s rational she’ll be more motivated to bring about the pleasure.

Contrary to some of Smith’s advertising, this sounds like a straightforward normative judgement (compare: “insofar as they are moral, agents are more strongly motivated to do that which will yield more happiness” - this is a moral judgement, not a psychological judgement). Smith clearly means for the judgement to be, in some sense, motivational, but it seems to me a normative claim about what is required for being rational.

As such, I’m not sure it’s true. The first question we have to ask is: what kind of value does Smith have in mind? Does he mean moral value? Or value for the agent? (Surely not, say, aesthetic value.) Presumably Smith would say that he has in mind value simpliciter, or all-things-considered value. Now one might very well doubt that there is such a thing (as have, e.g., Philippa Foot and Judith Jarvis Thompson7). But even if there is, I’m not sure rational motivation tracks it in the way suggested by Smith.

For is it really irrational to say “more valuable all-up, yes, but hooey to that - I’m doing what’s best for me”? Many have thought not. And what is perhaps even a harder case: is it really irrational to sacrifice one’s own good for the good of a beloved, if the result is less good all-up? (Perhaps one sacrifices one’s life to save a child from paraplegia or unjust imprisonment.) Likewise, one may sacrifice one’s own good for a valued project - a work of art, say - even if one grants that value all-up isn’t thereby best served. Altruism and dedication as well as selfishness may lead one to opt for something other than what it most valuable, and it isn’t clear that irrationality is always the appropriate charge (although other charges may be made; e.g. immorality in the case of selfishness).8 And notice that the connection is supposed to hold no matter what thinks valuable. If I think, for whatever crazy reason, that pain is intrinsically valuable, is it really irrational for me to be more strongly motivated to pursue pleasure? We might rather say that my evaluative judgements are irrational, and any motivation derived from or constitutive of them. If it isn’t always rational required to do what one judges most valuable, it isn’t always rationally required to do what one is most confident has value among alternatives judged equally valuable.

Let us grant Smith’s claim, though. Can it be used against noncognitivism? I don’t think so, so long as noncognitivists have an account of Certitude which is otherwise adequate. For they can just call upon noncognitivism about rationality judgements. It seems natural to extend noncognitivism about value judgements to noncognitivism about rationality judgements. If so, then noncognitivists are free to assent to whatever claim Smith makes about what is rationally required in the realm of Certitude. (The relative independence of normative moral judgements from metaethical ones like noncognitivism seems to obtain in the realm of rationality judgements as well.) Of course they would have a very different interpretation of what they are doing when they judge that it is rationally required to be most motivated to do what one judges most valuable, putting aside differences in confidence (perhaps their view would be that one is expressing one’s acceptance of norms requiring such motivation). But whether that interpretation is correct just is the question of noncognitivism (about rationality judgements), so in this context Smith couldn’t assume it false in order to argue against noncognitivism about evaluative judgements. He must allow the noncognitivist their every possible move, and noncognitivism about rationality judgements seems a natural one. If it is true then noncognitivists can easily assent to Smith’s claims about the rational requirements involving Certitude.

4. The need for higher-order desires.

I don’t think, then, that noncognitivists need worry about the rational requirements imposed by different degrees of Certitude, assuming they have an account of it which is otherwise adequate. But is the account offered in section 2 otherwise adequate? A problem for it jumps out, as well as a certain sort of solution to it. (Both the problem and the solution are remarkably close to things Smith says, so perhaps they are what he had in mind all along.) The problem appears if we consider betting situations in which the choice is between the things whose value is in question. Suppose that we are betting on the value of autonomy, and the value of the outcomes is not determined by something independently of autonomy (like money) but autonomy itself. For instance, suppose we are choosing between a certain amount of autonomy and a certain amount of pleasure, and we judge that that amount of autonomy is slightly better than that amount of pleasure, but we are quite a bit more confident that the pleasure is valuable at all. So the matrix might look like this.

  autonomy is good autonomy isn’t good
autonomy   +10 0
pleasure   +9 +9

In such cases, the accounts of Certitude and Importance offered push in opposite directions. For the greater Importance of autonomy was said to consist in a greater desire for autonomy; but the greater confidence in the value of pleasure consists in our tendency to choose pleasure under such circumstances. Can we have both a greater desire for autonomy and a tendency to choose pleasure?9

We can if our Certitude involves other attitudes which supplement our desire for pleasure, or oppose our desire for autonomy. This suggests a higher-order desire account of the sort described by Smith.10 It’s also worth pointing out that this comports nicely with views actually expressed by such defenders of noncognitivism as Gibbard and Blackburn.11 There are a variety of possible views here. In addition to Smith’s suggestion (that the relevant state is a desire to have the first-order desire), we might consider whether it is a desire to act upon the first-order desire, or a desire to consider whether one should have it, or some combination of these. Any number of these views seems capable of doing the job needed. Suppose, for instance, that the degree of confidence one has in an evaluative judgement is (at least partly) the strength of one’s second-order desire to act on the evaluative judgement.12 This may explain why, although our (first-order, evaluative-judgement-constituting) desire for autonomy is greater than our (first-order, evaluative-judgement-constituting) desire for pleasure, we choose pleasure. Our much stronger desire to act upon our desire for pleasure combines with our slightly weaker desire for pleasure to outweigh the combination of our much weaker desire to act on our desire for autonomy and our slightly stronger desire for autonomy. Such an account seems capable of giving us what we need to distinguish Importance from Certitude.13


1. See section 4, where the problem really comes down to Certitude.

2. Smith’s own attitude to the betting dispositions account is equivocal. On the one hand, he suggests that subject’s dispositions are how we “measure” their Certitude, and that Certitude “gets revealed in” such dispositions (307). This suggests that confidence and betting dispositions are distinct things which bear an epistemic relation. But he concludes his description of the role of betting dispositions in determining confidence levels by saying “Facts about subjects’ different levels of confidence are thus, as we might put it, synchronic facts about their evaluative judgements” (307; italics added). And he also says “Whereas facts about subjects’ levels of confidence are fixed synchronically by how much they would bet on the propositions they accept under circumnstaces of forced choice, facts about how stable the levels of confidence in the propositions subjects accept is under the impact of incoming information and reflection are plainly fixed diachronically. It is fixed by changes in how much subjects would be willing to bet on one outcome as opposed to another over time” (308). These suggest a more intimate relationship between betting dispositions and Certitude.

3. Of course, speaking of the outcomes in terms of money is a simplification. Ultimately we’ll want to portray the outcomes in terms of how well the agents ends are served.

4. P. 307.

5. See, e.g., Allan Gibbard, Wise Choices, Apt Feelings: A Theory of Normative Judgment [Cambridge, Mass.: Harvard University Press, 1990], ch. 5; and Simon Blackburn, Ruling Passions [Oxford: Oxford University Press, 1998], ch. 3.

6. P. 309.

7. See Philippa Foot, “Utilitarianism and the Virtues”, Mind 94 (1985), 196-209; Gilbert Harman & Judith Jarvis Thomson, Moral Relativism and Moral Objectivity [Oxford: Blackwell, 1996], ch. 5.

8. One might wonder whether a tight connection between rationality and value might be forged if the kind of value in question is instrumental; i.e. what will serve one’s aims. Even if that is so, it doesn’t seem to help Smith’s case, since he doesn’t seem to be talking about instrumental value (his main examples are pleasure and autonomy, which seem to be in the realm of non-instrumental value). But even then it is not clear that one is rationally required to do what promotes one’s own aims. If sympathy leads me to sacrifice my life-long aim of buying an island in the Carribbean to which to retire, and I instead donate the money to charity, it sounds a bit harsh to call me “irrational”. (One might suggest that in this case, during the time of my action, I was indeed doing what most promoted my ends; the idea being that my ends shifted during that time from buying an island to helping those in need. This, however, raises the question of whether it is even possible to do other than what [one sees as] most promoting one’s ends; in which case the question of the irrationality of doing so doesn’t even arise. See Simon Blackburn [Ruling Passions, ch. 6] for an illuminating discussion of this issue.)

9. This is not just a problem for the betting dispositions account. Any noncognitivist account is going to have to allow the tendency to give up more easily one’s commitment to autonomy the less certain one is of its value, even in the face of an invariant view about just how valuable it is. This pushes the noncognitivist to suppose that Certitude involves some further state, beyond the desire constituting the evaluative judgement in the first place, which can supplement or oppose its strength. Thanks to Greg Littmann for discussions of this and other matters.

10. P. 317ff.

11. See Gibbard, Wise Choices, ch. 8; Blackburn, Ruling Passions, ch. 3, and “Securing the Nots: Moral Epistemology for the Quasi-Realist”, in Moral Epistemology, New York, Oxford University Press, ed. Walter Sinnott-Armstrong, 1996. They are apt to put their view in terms of higher-order norms or evaluations, but of course on their noncognitivism these amount to (very broadly speaking) desires. 12. A desire to act upon it may or may not accompany any given desire. One may desire to swerve into the oncoming lane, or take heroin, without desiring to act upon that desire. On the other hand, one may very well desire to act upon one’s desire to eat, take care of one’s kids, or help the needy.

13. Smith says that it’s difficult for him to see how to square the higher-order desire account with “the observation that a subject who is less certain about the value of autonomy and more certain about the value of pleasure may none the less be rationally required to desire to be autonomous more than she desire to experience pleasure” because she judges autonomy to be more valuable (318). But this doesn’t seem difficult at all, given the point made in the last section. Unless Smith can establish that noncognitivism about rationality judgements is false, the noncognitivist about evaluative judgements can call upon it to explain how the rational requirement may obtain. He also charges that making Certitude a matter of second-order desires “seems arbitrary”. But whether it is so depends on the evidence for it. Presumably much of that evidence will simply be whatever supports noncognitivism in the first place. If noncognitivists can give us good reason to believe noncognitivism, and they can show that the closest thing to Certitude in evaluative judgements involves a higher-order desire, then the account loses its appearance of arbitrariness. In any event, this charge falls far short of showing that noncognitivists are “unable to accommodate” Certitude and its implications for rationality, Smith’s ultimate charge.

A final point I’d like to make, unrelated to the foregoing, relates to Smith’s charge that contemporary metaethics suffers from a “tendency to treat the belief-like and desire-like features of evaluative judgements in isolation from each other” (306). I would suggest that there is hardly such a tendency. A standard Cognitivist strategy for explaining the desire-like features of evaluative judgements is to show that those features are to be expected, given the content of those judgements. A standard Noncognitivist strategy for explaining the belief-like features of evaluative judgements is to show that those features are to be expected, given the particular desire-like nature of those judgements. This hardly bespeaks a “tendency to treat the belief-like and desire-like features of evaluative judgements in isolation from each other” (page number in Smith). [Jon: cites]

back-symp.gif - 1.6 K