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

Michael Ridge, , University of Edinburgh

Posted 01/15/03

"Certitude, Importance and Robustness for Non-Cognitivists"

Back in the bad old days, it was easy enough to spot non-cognitivists. They pressed radical doctrines with considerable bravado. Intoxicated by the apparent implications of logical positivism, early non-cognitivsts would say things like, "in saying that a certain type of action is right or wrong, I am not making any factual statement..." (Ayer 1936: 107) Like most rebellious youths, non-cognitivism eventually grew up. Later non-cognitivists developed the position into a more subtle doctrine, no longer committed to the revisionary doctrines associated with its forefathers. For example, Simon Blackburn has undertaken the "quasi-realist" project of showing how a non-cognitivist can "earn" the right to the seemingly realist discourse on a less metaphysically controversial and semantically implausible basis by giving a non-cognitivist analysis of realist-sounding semantics and pragmatics (Blackburn 1993).

To retain its distinctively non-cognitivist credentials1, quasi-realism must emphasize the role of desire-like states. To succeed in its aims, quasi-realism must accommodate the belief-like features of moral judgement countenanced by common-sense. This is a delicate balancing act, and Michael Smith's recent paper, "Evaluation, Uncertainty and Motivation" reminds us that moral judgements have at least three important belief-like features - Certitude, Robustness and Importance. Smith argues that because of their emphasis on desire, non-cognitivists cannot accommodate all of these belief-like features. The challenge is an interesting and novel one, but non-cognitivists have ample resources with which to meet it.


Beliefs can be held with greater or less confidence, and moral judgements seem to share this feature. Smith calls this feature of moral judgements 'Certitude'. Beliefs also can be more or less stable in the face of new information or further deliberation and this distinction also applies just as easily to evaluative judgements; Smith calls this feature 'Robustness'. Finally, evaluative judgements can be comparative. I can judge not only that pleasure and beauty are good but I can also judge that pleasure is better than beauty. Smith calls this feature of evaluative judgements 'Importance'.

Smith argues that his own cognitivist theory (see Smith 1994) is well-placed to accommodate these features of evaluative judgement. Smith's theory begins with the apparent platitude that for an agent to think an action in a given circumstance is desirable is for that agent to think that she would advise herself to act in that way in those circumstances if she were in circumstances in which she was best placed to give advice. Smith then suggests that agents are ideally situated to offer advice when their psychologies have been purged of all cognitive limitations and rational failings. The content of this advice is fixed by the desires the agent would have if purged of all cognitive limitations and rational failings. Smith argues that this account is well placed to accommodate Certitude, Importance and Robustness. On this account are just a species of belief, so whatever account we give of these features of belief in general should extend to evaluative judgements so understood.

Smith argues that non-cognitivism, by contrast, is ill-suited to accommodate these features. Smith characterizes non-cognitivism as the thesis that evaluative judgements are expressions not of beliefs that things are a certain evaluative way, but rather are expressions of desires that things be a certain non-evaluative way (316). The problem for non-cognitivism so understood arises from the fact that Certitude, Robustness and Importance are distinct structural features of evaluative judgements. So the non-cognitivist's adequately accommodating these features may require three structural features of desire that in some way that can both mimic Certitude, Robustness and Importance in the case of ordinary factual beliefs and explain how these features can play a role in motivating agents to pursue what they take to be good. Unfortunately, Smith argues, desires "posses just two structural features that look like they will be of any use in the present connection." (316) Desires differ in motivational strength, and this motivational strength may vary over time under the impact of new information. This latter feature is clearly well-suited to characterize the Robustness of moral judgement. The problem is that this leaves only one structural feature - motivational strength - to play the role of Importance and Certitude. Since Importance and Certitude are distinct, it seems unlikely that motivational strength could adequately model both.


Smith's argument implicitly proceeds on the assumption that the non-cognitivist can appeal only to desires. For his argument's main premise is roughly "desires possess just two structural features that look like they will be of any use in the present connection." (316) The suppressed premise is that the non-cognitivist can appeal only to structural features of desires. This suppressed premise might seem reasonable enough, but it does not follow from Smith's characterization of non-cognitivism. On Smith's account, the non-cognitivist maintains that moral judgements do not express beliefs about the ways that things are in evaluative respects, but instead express desires that things be a certain way in non-evaluative respects. This leaves it open that evaluative judgements might express beliefs about the ways that things stand in non-evaluative respects.

Someone sympathetic to Smith's account might at this point grow impatient, complaining, "If non-cognitivists allow this much room for the expression of beliefs in moral judgement then their account has become a familiar form of cognitivism after all, namely subjectivism." Although understandable, this worry is ill-founded. The non-cognitivist sketched here and the cognitivist remain deeply divided on at least two important questions. First, cognitivists maintain that to accept a moral judgement an agent must have a belief with a particular kind of content, and non-cognitivists deny this. Smith, for example, maintains that accepting an evaluative judgement requires a belief about what one would desire if one's desire set were fully informed and maximally coherent. Moorean non-naturalists maintain that accepting an evaluative judgement requires a belief about a certain kind of non-natural property. Different cognitivists differ over the privileged content of evaluative beliefs, but they are all united in maintaining that the acceptance of an evaluative judgement requires adopting a belief with a particular content.

Second, cognitivists maintain, but non-cognitivists deny, that evaluative utterances express beliefs that are thereby guaranteed to provide the truth-conditions for those utterances. Non-cognitivists can allow that accepting evaluative utterances does always involve the having of beliefs as well as desires, but deny that such beliefs are thereby guaranteed to provide the truth-conditions for the corresponding evaluative utterances. For example, perhaps Frank Jackson and Philip Pettit are right to hold that non-cognitivists must admit that moral utterances express the belief that the speaker has the relevant attitude (see Jackson and Pettit 1998). However, it simply does not follow from this that the relevant belief (the speaker's belief that he has such-and-such attitude) provides the truth-conditions for the utterance. Quasi-realists often take seemingly meta-ethical claims (about mind-independence, for example) and argue that they are better understood as first-order claims in disguise. A similar move could be made here, assuming minimalist theories of truth and truth-aptness are defensible (which, for present purposes, I must simply assume). For example, on a minimalist theory of truth perhaps my saying that 'The belief that Pettit approves of X provides the truth-conditions for 'X is good'' just is my saying that X is good if and only if Pettit approves of X, and my saying the latter, in turn, may just be expression of an attitude of approval towards all and only the Pettit-approved stuff. 2

Once this much is clear, Smith's own theory provides a clue as to how a non-cognitivist might meet his challenge. Perhaps non-cognitivists should accept Smith's suggestion that to say someone ought to do something is to say they would advise themselves to do it if they were ideally situated to offer advice. It does at least seem very plausible that anyone who says that someone ought to do something is committed to supposing that someone ideally situated to offer advice would advise them to do it.

Interestingly, accepting this much does not commit us to Smith's theory. For Smith's theory also incorporates the controversial thesis that to say someone is ideally situated to offer advice is simply to say that they are fully informed of all the facts and that their desires are maximally coherent and unified. A version of Moore's Open Question Argument arguably casts doubt on this further proposal. Plausibly, a competent user of evaluative language could reject the idea that being fully informed of all relevant facts and having maximally coherent and unified desire set is sufficient for being and ideal advisor. Someone could insist that an ideal advisor must also be altruistic, and that merely being fully informed of all the facts and having a unified and coherent desire set does not guarantee altruism (being fully informed may make one jaded rather than altruistic). Insofar as the supposition that altruism is also essential for being an ideal advisor does not necessarily betray incompetence with evaluative language, Smith's proposal that statements of the form 'X is desirable' are analytically equivalent to statements about what one would advise oneself to do if one were fully informed and one's desire set were maximally unified and coherent looks implausible. Equally, though, someone could reject the idea that ideal advisors as such must be altruistic without thereby betraying linguistic confusion. Altruism is just one example, of course. Other competent speakers might insist that an advisor must care about justice, or the word of God, or equality to count as ideal and any of these positions plausibly can be accepted or rejected without thereby betraying conceptual incompetence. So being fully informed and having a maximally coherent and unified desire set plausibly does not capture what the meaning of 'ideal advisor'.

Neither is it obvious that being fully informed or having a maximally coherent and unified desire set is necessary for someone's being an ideal advisor. Being fully informed of all the facts might make someone jaded and cynical, and we might for that reason alone coherently doubt the credentials of such an advisor. Perhaps more provocatively, Richard Holton has argued that the Muggletonians were neither conceptually confused about the meanings of evaluative language nor could they possibly have understood ideal advisors as Smith suggests any competent user of evaluative language must (see Holton 1996; see also Bigelow and Smith 1997). The Muggletonians were a bizarre religious sect that took the story of the tree of knowledge very seriously, and held that reason and knowledge were deeply pernicious and to be avoided. Holton makes an interesting case that the Muggletonians would not have recognized their fully informed selves as ideal advisors but as Satan-like entities, and that it is conceptual arrogance to accuse them of not understanding evaluative language but highly implausible to suppose their evaluative judgements concerned what they would want if fully rational.

There is room for considerable debate about these arguments, but this is not the place to adjudicate that debate. Insofar as Smith's challenge raises a new problem for non-cognitivists that problem should not be parasitic on problems with their deployment of Moorean Open Question Arguments. The point in discussing these arguments is simply to provide some motivation for an account that accepts Smith's suggestion that in saying someone ought to do something an agent is indicating that they would want themselves to do it if ideally situated to offer advice but gives a non-cognitivist account of what it is to take an advisor to be ideal. The suggestion is that to say that an action is desirable is to express one's pro-attitude in favor of performing actions insofar as a certain kind of advisor would recommend them, and one's belief that the action in question would be recommended by such an advisor. A speaker's conception of an ideal advisor is captured by the relevant pro-attitude, and in effect amounts to the speaker's conception of the good. This pro-attitude functions like a concept (it is a 'quasi-concept'), and so is general, applying to all actions insofar as they would be recommended by a certain kind of advisor. The belief that such an advisor would recommend this action is the speaker's application of this concept to this action. Notice that this belief will not be a belief about the agent's desires but rather a belief about a certain kind of advisor, where the relevant kind of advisor is fixed by the agent's desires. There is a difference between believing that my favorite advisor would recommend something and believing that a fully informed advisor would recommend it; this distinction remains even if my favorite advisor happens to be a fully informed one.

On this account, evaluative judgements involve both non-cognitive attitudes and beliefs. It nonetheless is a version of non-cognitivism because no particular content is privileged as the content of evaluative beliefs as such; there can be massive disagreement about what makes an advisor an ideal one. The contrast with congitivist accounts like Smith's, according to which we both express beliefs with a particular content (e.g., as to what any agent whose desire set is maximally informed, coherent and unified would desire Jones to do) is striking. Nor need the relevant beliefs concern the speaker's own desires (unlike subjectivism), some non-natural property (like Moore's account) or any other particular property one might care to name.

Neither are the beliefs expressed by such utterances on this non-cognitivist account thereby guaranteed to provide the truth-conditions for the corresponding utterances. This is the second reason the proposed account does not collapse into any form of cognitivism. I can on this account coherently admit that your utterance expresses your conviction that God would favor Jones's truth-telling but deny that this makes it true that it Jones's telling the truth is desirable. In denying that this belief provides the truth-conditions for 'Jones's telling the truth is desirable' on this account I simply express my own non-cognitive attitudes in favor of a certain kind of advisor and my belief that such an advisor would not recommend actions insofar as they are favored by God. In this respect, minimalist account of truth and truth-aptness are important resources for defending non-cognitivism.

The crucial point for present purposes is that this version of non-cognitivism has ample resources with which to distinguish Certitude, Robustness and Importance in a plausible way. My confidence that telling the truth is good in such-and-such circumstances is simply my confidence in an ordinary belief - my belief that the relevant sort of advisor would approve of truth-telling here. So whatever general account we offer for confidence of ordinary descriptive beliefs (in terms of counterfactual betting behavior, e.g.) can be extended to cover the relevant evaluative judgements. Furthermore, this account will cover both judgements of instrumental and non-instrumental value. For on this account to judge that something has non-instrumental value is just to believe that an ideal advisor would approve of it and recommend it for its own sake and not merely for its consequences. So my confidence that pleasure is good for its own sake is simply my confidence that an certain kind of advisor would approve of pleasure for its own sake and not merely as an end. Once again, this is an ordinary belief about what a certain kind of agent would prefer, so no special story needs to be told about confidence.

Furthermore, an agent's confidence in her evaluative judgements so understood can play just the motivational role Certitude is supposed to play. All else being equal, the more confident I am that an action would be good the more motivated I should be to perform it. Similarly, all else being equal the more certain I am that an ideal advisor would recommend a given action the more likely I will be to perform the action, given my desire to perform actions insofar as they would be recommended by such an advisor. This is just a special case of a general phenomenon that everyone must accommodate - if an agent wants to perform actions with a given feature F (e.g., being favored by a certain kind of advisor) then the more confident the agent is that a given action is F the more strongly motivated the agent will be to perform the action. Perhaps the best explanation of this will appeal to the idea of confidence analyzed in terms of counterfactual betting behavior in forced choice situations or perhaps the best explanation will be rather different. The crucial point is that nobody denies that confidence in ordinary beliefs can somehow influence one's motivation in these ways. So there seem to be no new problems for the non-cognitivist here.

The robustness of my evaluative judgement simply consists in facts about how easily I could be brought to modify my judgement. The Robustness of my evaluative judgement is just its resistance to such changes in the face new information and further deliberation. It should be clear enough that Robustness is distinct from Certitude on this account and that Robustness is a diachronic concept while Certitude is a synchronic one.

Finally, Importance is easy enough to make sense of on this account. To judge that one outcome is better than another or one kind of thing is better than another is simply to believe that the relevant sort of advisor would prefer the one to the other. How good one judges something to be, more generally, is a matter of how much one supposes the relevant sort of advisor would approve of it.

Nor is it difficult to see how Importance, so understood, can play a suitable motivational role. Insofar as one is rational, all else being equal, one will also prefer what one takes to be more good to what one takes to be less good. For to judge something is good is to approve of it insofar as an ideal advisor would approve of it - the greater the approval of the object of evaluation the speaker takes it that an ideal advisor would have, the more the speaker approves of it. Of course, the 'all else being equal' clause does some real work here, as it did in the other cases. If one thinks an ideal advisor would more strongly approve of pleasure than knowledge but one is less certain that an ideal advisor would approve of pleasure at all then this will make a difference. The greater one's uncertainty about pleasure's value in contrast with one's greater confidence in the (lesser) value of knowledge, the more pleasure it will take for you to forego knowledge for pleasure. Again, this is no problem on the proposed account. In general, an agent's otherwise stronger desire for one outcome can be less strongly motivating than her otherwise weaker desire for another outcome if the agent is much more certain about the attainability of the more weakly desired outcome. In a similar fashion, if I desire to act as an ideal advisor would recommend then the more certain I am that pursuing knowledge would be recommended by an ideal advisor the more strongly motivated I will be to pursue knowledge here. I also think an ideal advisor would approve even more of pleasure, but I am much less certain that an ideal advisor would approve of pleasure at all. So if I forego knowledge for pleasure then I risk going for something that does not actually satisfy my desire to act as a certain kind of advisor would recommend at all. Depending on how great my uncertainty is, it might make sense for me to go for what I take to be a lesser good I am more sure is a good than what I take to be a greater good but am much less sure really is a good at all.

In sum, once the non-cognitivist makes room for a substantial role for belief in her account without abandoning her non-cognitivism, Smith's challenge can be met. In the last section, I consider an objection to this proposal.


One might complain that the account is fine as far as it goes, but still omits something important. What about an agent's confidence in her conception of the good itself? An agent's conception of the good on this account surely just is that agent's pro-attitude in favor of actions insofar as a certain kind of advisor would favor them. When it comes to the agent's conception of the good, it is hard to see how beliefs come into it. A number of points should be made about this objection.

First, the scope of the putative problem for non-cognitivism has at this point been dramatically reduced. So far as this objection goes, the non-cognitivist can accommodate the distinctions between Certitude, Importance and Robustness so far as ordinary first-order judgements as to what things are good and bad. It is only when we come to the more distinctively philosophical idea of an agent's conception of the good itself, as opposed to her application of that conception to the world, that problems seem to arise. Furthermore, Robustness of an agent's conception of the good is easy enough to understand even here. The Robustness of my conception of the good is simply a function of how easily my fundamental desire to follow the advice of a certain kind of advisor would be in the face of possible new information and deliberation. Importance, by contrast, raises no real problems because it seems not to apply when it comes to one's conception of the good itself. Admittedly, an agent might think that one conception of the good is better to adopt than another. A conception of the good can in this way be self-effacing and not best by its own lights, but this too is easy enough to account for on the proposed account. That will just be a judgement that the relevant ideal advisor would recommend adopting this conception rather than itself in these circumstances. Once we start talking about importance of a conception of the good simpliciter where this is distinct from the importance of people's accepting that conception we stop saying anything intelligible. So Certitude is the only feature that threatens to raise a problem, and it threatens to do so only in the case of an agent's conception of the good itself and not in the case of ordinary everyday evaluative judgements Any remaining problem therefore is contained to a very narrow range of cases.

Second, the quasi-realist project only requires making sense of the belief-like features of evaluative judgements that are clearly countenanced by common sense. Once we come to a speaker's conception of the good as such, it is not at all clear that the idea of Certitude has intelligible application. Ordinary folks may be perfectly happy with questions like, 'How confident are you that pleasure is good for its own sake and not just for its consequences?' and 'How sure are you that knowledge is good in virtue of its intrinsic properties?' It is not so obvious that ordinary folks understand questions like 'How confident are you in your concept of goodness as what God would command?'. To some degree, the non-cognitivist could admit that their account does not make sense of such questions without embarrassment. We can be more or less confident that a concept (or quasi-concept) applies to the world; it is not at all obvious that we can be more or less confident in our concept (or quasi-concept) itself.

In addition to applying their evaluative concepts to the world, speakers do make judgements about those concepts themselves. Smith, for example, expresses his own beliefs about what words like 'desirable' mean. Presumably, these kinds of meta-ethical judgements are just ordinary beliefs about the meanings of words and not first-order evaluative judgements. In that case, we can understand Certitude here in whatever way we should understand Certitude more generally for ordinary beliefs.


Smith's challenge is a good one and if the non-cognitivist were restricted to modelling these features only in terms of the structural features of desires then the challenge would be hard to meet. Fortunately, this restriction is not justified. Non-cognitivists can make room for a significant role for belief in their account without abandoning their non-cognitivism. Perhaps unsurprisingly, making room for beliefs in this way provides crucial resources with which the non-cognitivist can make good sense of some of the more sophisticated belief-like features of evaluative judgement that figure in Smith's challenge.


1.Blackburn actually disapproves of the label ‘non-cognivitivism’ for his view, but his position does fall into the same tradition, very broadly speaking, as the positions of Ayer, Stevenson and Hare.

2.See also my review of Jackson and Pettit’s article on BEARs at


Ayer, A.J. 1936. Language, Truth, and Logic. Gollancz. Second Edition 1946.
Blackburn, S. 1993. Essays in Quasi-Realism. Oxford: Oxford University Press. _____. 1998. Ruling Passions. Oxford: Oxford University Press.
Bigelow, J. and Smith, M. 1997. "How not to be Muddled by a Meddlesome Muggletonian." Australasian Journal of Philosophy 75: 511-527.
Holton, Richard. 1996. "Reason, Value and the Muggletonians." Australasian Journal of Philosophy 74: 484-7
Jackson, F. and Pettit, P. 1998. "A Problem For Expressivists." Analysis. 58: 239-251.
Smith, M. 1994. The Moral Problem. Blackwell, Oxford.

back-symp.gif - 1.6 K