The Non-cognitivist Challenge
The Non-cognitivist Challenge
Abstract and Keywords
Sets out the contours of the moral realism that the author wishes to defend, and provides an extended series of arguments against non‐cognitivism and expressivism. Seeks to establish a presumptive case for moral cognitivism.
In this chapter I will introduce the kind of moral realism I favour, and then introduce, and critique, the family of non-cognitivist theories that serves as one of realism's major competitors. Though I will not try to offer a decisive set of objections to non-cognitivism, I do, by chapter's end, take myself to have provided a strong presumption against it, and in favour of moral cognitivism generally.
As we shall see, this presumption, even if warranted, is not enough to land us in the realist's camp. For there is a split among cognitivists, between those who favour realism and those who favour constructivism. In what follows, I explain this division, and reserve my next chapter for an attempt to reveal constructivism's weaknesses. If all goes as planned, we will have, at the conclusion of these first two chapters, a strong case on behalf of moral realism. We will also be face to face with a set of powerful antirealist objections, ones that have done much to convince realism's opponents of its inadequacy. My aim, in the remainder of the book, is to address these concerns in a way that leaves us with a plausible, coherent form of moral realism that supplies satisfying answers to these perennial objections.
I. What Moral Realism Is
There are many different versions of moral realism, and very likely no single description will neatly fit all those theories that have taken the label. At the simplest level, all realists endorse the idea that there is a moral reality that people are trying to represent when they issue judgements about what is right and wrong. The disagreements that arise among realists primarily have to do with the nature of this reality. Despite these disagreements, there is much common ground.
(p.14) We can divide views about the reality of a given domain into three large types. Nihilists reject the reality of the domain. We are all nihilists about goblins and unicorns; such things don't exist. Nihilists about morality believe that there isn't any moral reality at all. We might speak as if there is, and there may be good political, psychological, or strategic reasons for doing so. But in the final analysis, nothing really is right or wrong; these are terms that never refer.
Constructivists endorse the reality of a domain, but explain this by invoking a constructive function out of which the reality is created. This function has moral reality as its output. What distinguishes constructivist theories from one another are the different views about the proper input. Subjectivists claim that individual tastes and opinions are the things out of which moral reality is constructed. Relativists cite various conventions or social agreements. Kantians cite the workings of the rational will. Contractarians cite the edicts of deliberators situated in special circumstances of choice. What is common to all constructivists is the idea that moral reality is constituted by the attitudes, actions, responses, or outlooks of persons, possibly under idealized conditions. In short, moral reality is constructed from the states or activities (understood very broadly) undertaken from a preferred standpoint. The absence of this standpoint signifies the absence of moral reality. If, for instance, there are no individuals who share conventions or make agreements with one another, then for the relativist there simply is no moral reality. For the Kantian, there is no moral reality—no genuine moral obligations or any justified moral claims—if there is no such thing as pure practical reason. Nothing is actually right or wrong, according to the subjectivist, if there are no people, or no personal opinions about matters of fact. For constructivists, what reality there is depends crucially on the existence of the preferred standpoint.1
(p.15) Realists endorse the reality of a domain and do so in opposition to constructivists. Realism is sometimes contrasted with constructivism by invoking the claim that, for realists, morality is mind-independent. There is some truth to this, but it seems ill expressed by the present notion. Any plausible moral theory will claim that some moral truths depend crucially on an agent's mental states. We obviously can't enter a moral assessment of an agent's motivations and intentions without recourse to what is going on in her mind. And the moral status of an action may depend very importantly on how pleased or miserable it makes others, whether it prompts feelings of anger or empathy, whether one who is harmed has given his consent to the treatment, etc. Further, it seems to make eminent sense to say that, absent all agents, there would be no moral facts (as opposed to moral standards)—no particular instances of right and wrong, moral good and evil, etc.2 We must be careful to define realism in such a way as to allow for these home truths.
The way I would prefer to characterize the realist position is by reference to its endorsement of the stance-independence3 of moral reality. Realists believe that there are moral truths that obtain independently of any preferred perspective, in the sense that the moral standards that fix the moral facts are not made true by virtue of their ratification from within any given actual or hypothetical perspective. That a person takes a particular attitude toward a putative moral standard is not what makes that standard correct.
This characterization of realism easily accommodates the possibility that something's rightness or admirability depends crucially on an agent's attitudes. But the dependence is to be understood in a particular way. Any first-order, substantive normative theory worth its salt will require attention to the mental states of agents in a variety of quite complex ways. But realism, being a view about the status of such normative theories, insists that the truth of any first-order normative standard is not a function of what anyone happens to think of it. Such standards, if true, are not made true, and, in particular, are not (p.16) correct in virtue of being vindicated by some process of (inter)personal election or approbation.4
To get a better understanding of realism's commitments, consider the following contrast. An ethical relativism of the sort embraced by Harman (1975) or Wong (1984) claims that what moral reality there is is fixed entirely by certain social agreements. The reality—what really is right and wrong, for instance—is constructed from the content of these agreements. No agreements, no reality. Realists, by contrast, will claim that what really is right and wrong is conceptually and existentially independent of any such agreements. Indeed, for realists, moral reality is conceptually prior to and existentially independent of the moral truths we can construct even from the responses of an ideal observer. It may be, given a precise enough characterization of an ideal observer, that his judgements are inerrant. The realist could allow, in other words, that there is a strong conceptual connection between the responses of a suitably characterized ideal observer and the content of moral reality. But, for the realist, that would be because the ideal observer would never fail to see what was right and wrong anyway. The responses of an idealized observer would not be constitutive of moral truth, but merely bear a very close (perhaps perfect) correlation with a set of truths whose conditions may be fixed without any reference to such an observer.
Realism is sometimes described as the view that moral truths are evidence-transcendent. This could mean one of two things. First, it might be that moral truths are not constituted by the evidence we have for them. If evidence is restricted to beliefs, then this tallies with my conception of realism, since realism will not make the truth of a moral proposition dependent on what anyone believes about it. But sometimes the notion of evidence-transcendence is taken to mean that there might be truths that are in principle unknowable. This has worried many thinkers. Whether their worries are justified depends on the understanding of knowability that we bring to the table.
The claim that something is unknowable is a modal claim, and discussions of modal issues can get very complicated very fast. I want to avoid these (p.17) complexities here,5 and so will enter what I believe to be the most relevant particular claims surrounding the issue. First, realists are committed to the idea that a moral standard might be correct even if no actual person believed it to be; indeed, even if everyone renounced it. Second, realists allow that any actual agent might fail to recognize a moral truth, even after she had made the very best epistemic efforts she could (given her initial psychological profile). Third, realists are not committed to the idea that moral truths are inaccessible to absolutely ideal epistemic agents at the Piercean limit of enquiry. Epistemically ideal agents who have reached this limit will be fully informed. This means that they will know all facts. Moral realists believe that some of these facts are moral ones; so a genuinely ideal epistemic judge will know all moral facts. For such an agent, there will be no unknowable moral truths. But for all the rest of us—that is, for every actual person—the various epistemic liabilities we carry around with us will almost certainly prevent us from ever knowing the whole moral truth.6 This isn't an iron-clad conceptual necessity, but rather an extremely plausible take on our actual epistemic limitations. When it comes to ideal epistemic agents, however, the important point is that their impeccable accuracy reflects an acquaintance with a truth not of their own making. Their perfect knowledge is not constitutive of moral facts, but rather reflects an awareness of what is there, awaiting their discovery.
Moral realism is a form of cognitivism. Cognitivism, like realism, is a term of art that has been used somewhat differently by different philosophers. As I shall understand it, a view is cognitivist if it allows for a central class of judgements within a domain to count as beliefs, capable of being true or false in virtue of their more or less accurate representation of the facts within the domain. Moral realism satisfies these conditions. Realists see moral judgements as beliefs, some of which are true, and true in virtue of correctly reporting moral facts.
Every realist is a cognitivist, though there are many cognitivist theories that reject realism. The different constructivist theories mentioned (p.18) above—subjectivism, relativism, contractarianism, Kantianism, ideal observer theories—are all forms of cognitivism whose acceptance implies a rejection of realism. One hurdle that stands in realism's way is to show why we should opt for realism over any of these forms of constructivism.
Thus there are three distinct kinds of challenge to moral realism. The first is to show why some form of cognitivism is acceptable. The second is to show why realism should be preferred to constructivism. And the last is to show why one version of realism is better than the others. I will consider the first challenge in this chapter, and the second in the next. The intramural debate will be sorted out in Part II.
II. The Non-cognitivist Challenge
Non-cognitivism is the denial of cognitivism. According to non-cognitivism, there are no moral facts or truths. Non-cognitivists have allied this brand of moral nihilism with a specific view about the function of moral discourse. On their view, moral judgements are not beliefs and thus are not truth-evaluable. Moral judgements don't attempt to, and don't ever, state facts. Their purpose isn't to describe any sort of moral reality. Instead, they serve as expressive vehicles, primarily giving vent to our emotions, prescribing courses of conduct, or expressing our non-cognitive commitments. As such, they aren't the sort of things fit to be considered either true or false.
Non-cognitivism originated as a kind of debunking theory. Most people, philosophers and non-philosophers alike, had long believed that some form of objectivism in ethics was true. Objectivism implies two things. First, there is truth in ethics. Second, what truth there is is fixed by things other than parochial opinion; the constitution of ethical truth requires some independence from contingent desires, passions, and standpoints. Non-cognitivists didn't much like objectivism.
According to a standard array of options, this rejection of objectivism would have left a choice between ethical subjectivism and ethical relativism. Ethical subjectivism, in this context, is the theory that (i) an action is right just in case the one who judges the action approves of it, and (ii) a moral judgement is true just in case it accurately reports the sentiments of the one who holds such a judgement. Relativism, by contrast, locates the source of moral truth not in each individual, but rather in some form of social consensus or convention. On this view, a moral judgement's truth is determined by its ability to capture the implications of these social agreements.
(p.19) Both subjectivism and relativism are kinds of cognitivism. Specifically, they are forms of ethical naturalism, a family of cognitivist theories. Naturalists believe that moral facts are a kind of natural fact, i.e. one whose existence can be confirmed by the best natural or social scientific theories.7 Subjectivism and relativism make the rightness of an action dependent entirely on either personal or social approval, and the existence of such approval is, in principle at least, scientifically determinable. Because ethical naturalism is a form of cognitivism, both subjectivists and relativists take moral judgements to be truth-evaluable. And both views imply the possibility of genuine moral knowledge. The non-cognitivists' rejection of cognitivism left them as opposed to subjectivism and relativism as it did to objectivism.
One way of explaining the non-cognitivist's rejection of these standard anti-objectivist theories is by noting the important difference between reductionist and non-cognitivist programmes. Reductionists want to eliminate the mystery in a given area of discussion by reducing the key notions to those that are more tractable. Reductionists don't deny the reality of the reducible domain, because they believe that the reality is the very same reality whose existence passes theoretical muster. Reductionists believe that there really are (e.g.) mental, biological, or moral facts, but they believe this because they believe that such facts are species of neurophysiological, chemical, and other natural facts—all of which are accepted as respectable elements in an ontology. Reductionists hold that we have beliefs about the mental and the moral, that some such beliefs are true, some justified, and, if the winds are good, some may even get so far as to constitute knowledge.
Non-cognitivists, by contrast, reject the reality of the domain that others seek to reduce. Non-cognitivists in ethics, nowadays known as expressivists, will see the notion of a moral fact as akin to that of a fact about devils or phlogiston. There aren't any such facts, and so it doesn't pay to try to reduce this sort of talk to that which describes some genuine aspect of reality. Such talk never refers.
We need some care here, however, since this stance is compatible with both non-cognitivism and a kind of cognitivist view. The cognitivist line says that the relevant language is conveyed by beliefs which purport to be descriptive, but that there is a massive error of presupposition that underlies the vocabulary. If that were so, then all claims within the domain would be false—they would assume the existence of a kind of reality that isn't there. This is an error (p.20) theory; John Mackie (1977) is the most famous proponent of this sort of view in ethics. According to such a view, moral terms have descriptive meanings, and we have moral beliefs, but they are uniformly false.
The expressivist diagnosis denies that there is any such error, since it rejects cognitivism about the domain. It does not see agents as trying, but failing, to describe anything. So we needn't charge people with error. But we can never credit them with knowing the truth, either. Non-cognitivists are spared the task of trying to establish reductive relations between morality and other, recognized domains, because non-cognitivists reject the idea that there are moral propositions, facts, or truths that require reducing. That is why the subjectivist or relativist theories are anathema to non-cognitivists. Despite a shared, deep scepticism about ethical objectivity, subjectivists and relativists insist on the reducibility of moral facts to those of personal psychology or social agreement. Expressivists will have none of it.
According to non-cognitivism, the fundamental mistake made by all such theories is their presupposition that the purpose of moral language is to describe facts. As non-cognitivists see it, the point of moral discourse is not to report some fact about oneself, one's group, or the larger world, but instead to give vent to one's feelings and persuade others to share them (Stevenson 1937, 1963), prescribe some rule of conduct for oneself and others (Hare 1952, 1963), or express one's commitment to norms regulating guilt and anger (Gibbard 1990). Such judgements do not (at least in the first instance8) admit of truth or falsity—indeed, there is nothing that could make them true. There is no world of moral facts against which the truth of a moral judgement can be checked. There are no moral properties whose instantiations can determine the moral qualities of persons, traits, actions, practices, or institutions. There is the familiar world that science speaks of. And there is us, responding to that world. And that is it.
Some early non-cognitivists took this view to stand for the idea that moral predicates are literally meaningless (e.g. Ayer 1936). Their more sophisticated (p.21) successors (Hare 1952, 1963; Blackburn 1993a, 1998; Gibbard 1990; Timmons 1999) might not be so bold as to characterize matters just this way, but the root idea is the same. Expressivists all agree that the incorporation of moral predicates in statements adds no new information. To say that Tom donated a large sum of money and that this was good conveys just a single fact—that of Tom's donation. Nothing further is described or reported, though a favourable attitude towards such donations is expressed. The statement looks like a conjunction of robustly truth-evaluable claims, false if either conjunct is false. But only one conjunct—the first—could be literally false; to assimilate the second conjunct to the first is to commit a category mistake, and be misled by grammatical similarities. ‘Is good’ functions syntactically like any other predicate, but only non-evaluative predicates ever manage to convey information.
Expressivism has attracted broad support primarily because of two things: the ease with which it comports with a prevailing scientific, naturalistic view of the world, and its ability to avoid the serious failings alleged to beset its cognitivist competitors. As to the first point, expressivists stake out their position precisely by rejecting the possibility that there might be unanalysed normativity in the world. Their ontology is compatible with that proposed by the most advanced science of the day. We needn't add evaluative properties or relations to our world view in order to accommodate moral talk. On the assumption that human beings and their behaviours are resolvable in principle into material constituents and events, the expressivist imports nothing extra into a natural scientist's picture of what the world contains. Ontologically, the picture is simpler than that proposed by cognitivists, who are forced to augment the scientific world view with an additional layer of moral properties. Other things equal, simplicity is to be preferred.
Additionally, expressivism avoids the serious epistemological shortcomings alleged to beset its competitors. Since there are no moral truths to be known, expressivists needn't entangle themselves in the epistemological contortions involved in explaining the perception of a sui generis realm of facts. This has been a perennial worry for non-naturalists—those cognitivists who reject naturalism, and so reject any reduction of moral properties to natural properties (i.e. properties whose instantiations are scientifically verifiable). If, as non-naturalists argue, goodness and rightness are not the very same thing as any natural property, then how can we ever know whether anything is good or right? We have fairly well-worked-out theories for detecting the presence of natural properties in the natural world, but the detection of a realm of non-natural properties raises problems. Non-naturalists agree that something's goodness (p.22) depends on its natural features, but these features are claimed to be distinct from its moral features. So how does appreciation of the presence of natural features give us knowledge of the moral status of a situation? It's really quite mysterious.
While naturalists claim an ability to avoid recourse to an ill-defined intuitive capacity, they fail in the end (according to non-cognitivists) to map out any plausible escape route. For even if, as naturalists claim, moral properties just are a species of natural, empirically confirmable properties, naturalists face the formidable difficulty of explaining how we might gain knowledge of the naturalistic reductions. Once we grant that rightness, say, is maximizing happiness, then we can all agree that justifiably believing in an action's rightness is a matter of deploying the best account of empirical knowledge (to discover the distribution of happiness). But how do we know that rightness is the very same thing as maximizing happiness? The two certainly don't appear to be synonymous, and the method of conceptual analysis has had a dismal history in the moral domain. But if it is an a posteriori identification we are seeking, of the sort that links water and
Lastly, non-cognitivists have a ready explanation for the practicality of ethics. Stevenson (1937) described this as the ‘magnetism’ of moral judgement. Judging something right is thought to entail a motivation to do it, and this is perfectly explained if, as non-cognitivists claim, such judgements are expressions of conative elements of our outlook. If cognitivism is true, however, then moral judgements are beliefs: they are putative reports of a certain kind of fact, can be true or false, and can be held with varying degrees of epistemic justification. On a broadly Humean theory of motivation, which is accepted by all non-cognitivists (and most cognitivists), beliefs alone cannot motivate. Yet moral judgements alone can motivate. Therefore moral judgements are not beliefs. Therefore cognitivism is false. Therefore realism is false.
In a nutshell: what could moral facts be? How could we know them? And how could they move us? Expressivists claim that cognitivists lack satisfying answers to all of these questions. I think we must agree that these are the questions to be asking.
III. A Critique of Non-cognitivism
Suppose for a moment—my worst-case scenario—that the theory I develop in these pages fails to adequately address the non-cognitivist's concerns. Still, (p.23) non-cognitivism doesn't necessarily succeed by default. This for three reasons. First, other realist theories might be more plausible than the one on offer here. Second, a constructivist theory might succeed where realisms fail. And third, non-cognitivist theories fail to accommodate some very attractive features of cognitivism, and are themselves subject to serious challenges. Let me rehearse some of them here.
Cognitivism and Truth Talk
Only cognitivism straightforwardly preserves ordinary talk of moral truth. We appear to take at face value such locutions as ‘it is true that infanticide is wrong’. We allow for the possibility of moral mistake and often characterize it as a case in which a person speaks falsely, or has a false belief. When we experience moral perplexity, we often see ourselves as engaged in a search for the truth about who is in the right, or where our obligations lie. We can well explain the point and persistence of moral disagreement by attributing to agents the presupposition that there is a right answer awaiting discovery. Were they convinced that there was no truth of the matter, most interlocutors would see their continued disagreement as pointless; as pointless as, say, entering an intractable debate about whether red or orange was really the most beautiful colour.
Relatedly, we believe that moral argument can take the logical form of other kinds of argument. We think of sentential operators in moral sentences as truth-functional. The law of excluded middle holds as strictly for moral discourse as for non-moral discourse. We recognize the validity of modus ponens inferences that incorporate moral claims as premisses. We freely use the logical connectives in making and evaluating moral claims. We standardly assess moral arguments as valid or invalid, sound or unsound. This indicates at least a tacit assumption that truth-preservation is an aim of moral argument. Cognitivists have ready, straightforward analyses of such a view of moral argument. Non-cognitivists don't.
In particular, non-cognitivists have always foundered on what Blackburn has called ‘Frege's Abyss’ (though for Blackburn, it is more of a divot than anything to get worried about). The problem, first noted in the contemporary literature by Geach (1960, 1965), is how expressivists can make sense of sameness of meaning in asserted and unasserted contexts. This they must do if they are to preserve the possibility of logical validity, or at least credibly explain away the appearances of validity and soundness in moral argument. The trouble arises if we note, first, that much moral argument employs assertions, negations, disjunctions, and conditionals, and then note that, by (p.24) non-cognitivist lights, the meaning of a moral sentence depends on the accompanying attitude that receives expression therein. Since one's attitude changes if one asserts a claim, negates it, cites it as a disjunct, or embeds it in the antecedent of a conditional, then the meaning of the relevant phrase ought to change as well. But it surely doesn't, and can't, if the possibility of logical moral argument is to be preserved. A minor cottage industry has grown up around this problem. It has yielded some ingenious solutions, but there is good reason to doubt the success of any one of them.9
Cognitivists assume that moral predicates are meaningful and can be used to describe the subjects they are predicated of. We use the indicative mood when issuing moral judgements. We assert that practices, character traits, or states are vicious, morally attractive, or deserving; we state that motives or actions exemplify such things as goodness, generosity, benevolence. When using evaluative language, most people would find it perfectly natural to characterize their doings as instances of describing things as good or bad, or as attributing to things certain qualities—goodness or badness. Moral talk is shot through with description, attribution, and predication. This makes perfect sense if cognitivism is true. The non-cognitivist story cannot be nearly as natural or simple.
Consider a very simple example that reveals the comparative advantages here. Suppose someone tells me that virtue deserves reward. The cognitivist glosses the claim by saying that there is such a thing as virtue, and desert, and reward. The sentence means just what it says, and is true just in case virtue and reward are related as it says they are. But for expressivists, when we say that virtue deserves reward, we aren't describing a thing (virtue), and saying of it that it has some property (that of deserving reward). There is no such thing; there is no such property. So we must be evincing a non-descriptive attitude. What would that be?
There are two problems with expressivist efforts to provide an answer. First, expressivists will have difficulty translating the general formula ‘X deserves y.’ We must, at the least, be evincing an attitude of approval when contemplating the state of affairs in which X gets y. But surely there's more (p.25) involved than that. For we can register such approval in any number of other cases where desert isn't involved. ‘Wouldn't it be nice were X to get y’; ‘it's good that X has y’, etc. It isn't clear how expressivists are going to analyse the attitude that receives expression in desert claims, as opposed to the various other claims that register our approvals. Second, even if expressivists can distinguish desert claims from others, they don't believe in virtue. ‘Virtue’, like other ethical terms, denotes nothing; it never refers. So not only are we asked to find a suitable candidate for the special kind of approval that gets expressed in desert claims; we have nothing, really, that serves as the object of such approvals. Perhaps expressivists can provide a plausible translation of this simple claim. But surely cognitivists have the upper hand here. They don't require a paraphrase to begin with.
The first problem, just discussed, is really a particular instance of a quite general difficulty, that of accounting for the rich diversity of moral predicates. If expressivism is true, the essence of any moral statement is the expression of a fundamentally non-representational attitude towards some natural state of affairs. If, as it seems, we convey and mean something slightly different when we say of an action that it is virtuous, right, mandatory, supererogatory, kind, beneficent, admirable, conscientious, attractive, desirable, laudable, saintly, or fine, then non-cognitivists must explain this by citing a different attitude that receives expression in each case. But our attitudes don't seem nearly as diverse or fine-grained as the predicates we standardly deploy in moral assessment.10 Cognitivists straightforwardly account for these differences by referring to different meanings, contents, or properties that are exemplified when these different assessments are true. Expressivists must either deny that these predicates signify different assessments, or identify a different attitude for each predicate. Neither route seems very promising.
Cognitivism also has some appeal when dealing with issues of moral motivation. If moral judgements are moral beliefs, capable of truth or falsity; and if moral beliefs, like other beliefs, do not necessarily motivate those who hold them; then it is at least conceptually possible that an agent issue sincere moral judgements and yet fail to be motivated. And this does seem a conceptual possibility: we can imagine agents who, through debilitating mental or physical illness, or an enervating lassitude, or an antipathy to morality, are left utterly cold by the moral judgements they take to be true. Non-cognitivism, by contrast, takes moral judgement to essentially involve the expression of attitudes that are (p.26) motivating. So any judgement that fails to motivate, according to non-cognitivism, cannot possibly be a moral judgement. The counter-intuitive consequences of this position are discussed at length in Chapter 6.
Cognitivism preserves our talk of moral belief and the possibility of moral knowledge. To say of someone that he believes certain actions to be wrong, or others to be good, is to say nothing strange, but something we all understand. To turn some such beliefs into knowledge is recognized by most to be at least a possibility. Despite reluctance that stems from egalitarian impulses, we often talk of some being morally wiser than others, and typically abandon our reluctance when pressed with concrete comparisons. And even if we are sceptical about the number of situations in which people or societies have registered moral improvement, we are less dubious about moral regression, and either idea presupposes the possibility of mistake in moral beliefs and outlooks.
Some contemporary non-cognitivists want to allow for the possibility of moral error. But this isn't meant to signify the possibility of falsehood. On their view, moral error isn't a matter of saying or believing what's false. Instead, error might be diagnosed in one of two ways. First, it might essentially involve a kind of incoherence. Second, it might involve a failure to satisfy some other standard meant to govern attitude formation.
The first diagnosis identifies moral error with the case in which one holds attitudes that are inconsistent with one another. Practical attitudes are inconsistent if they counsel actions that are not jointly realizable. But why think of this as any kind of error? If I am committed to norms that govern my work behaviour, and also committed to being a good parent, why I am making some kind of mistake if my domestic commitments require that I sacrifice my work ethic? And even if attitudinal conflict is what moral mistake consists in, which of the conflicting attitudes should be singled out as the faulty one? Presumably that attitude which conflicts with the relevant highest-order attitude that I hold. But what if there is conflict within my highest-order attitudes? What if they are incomplete, or specified at such a level of generality as to fail to arbitrate in particular cases? Then all we can say is that there is a mistake, but we will lack the resources for identifying the culprit. This isn't merely an epistemological problem. In the absence of a determinate highest-order attitude that can adjudicate attitudinal conflict, the non-cognitivist must say that there is no particular mistaken attitude. So any case of attitudinal conflict involves a mistake, and yet in some cases, there may be no attitude that is (p.27) itself mistaken, so no determinate answer as to which should be modified or abandoned.
This isn't a contradictory view, but the cognitivist has a more palatable account of the nature of moral mistake. When attitudes conflict—and, for the cognitivist, the relevant attitudes are moral beliefs—at least one is false. This is exactly what we want to say when we charge someone with moral error. Consider: even if we grant that a conflict of non-cognitive attitudes must involve mistake, is this really the central error we are trying to identify when convicting someone like Himmler or Mao of gross immoralities? Indeed, the attitudes of such fanatics needn't involve any internal conflict at all. And this seems a serious shortcoming for the non-cognitivist analysis of moral error, since we do think that people can be morally mistaken even if they've built themselves a perfectly coherent set of attitudes. For such attitudes might be pitched in precisely the wrong direction.
Non-cognitivists might instead claim that moral error involves commitments that would be excluded from a set of best possible attitudes. Not just any incoherence will mark error; and coherence is not sufficient to guarantee freedom from moral mistake. So we can set aside the worries just mentioned. But which criteria must be satisfied to secure us from error? Reducing moral error is a matter of registering improvement in one's sensibility—one's more or less integrated set of attitudes. Yet plausible candidates for measuring such improvement are themselves normative conditions. At a minimum, attitudes are improved to the extent that they are formed against consideration of relevant information, free of bias, apportioning a due or proper weight to each individual's interests. But the notions of relevance, bias, and propriety are normative. There is no reason to be confident that we can understand the concept of a sensibility's improvement, or that of a set of best possible attitudes, without relying on other normative concepts. And so there is no reason to think that we can make sense of moral error without importing some normative constraints. If, as non-cognitivists claim, there really aren't any such things, then there really isn't any such thing as moral error or improvement. Alternatively, if these constraints, being normative, are really nothing other than an expression of a non-cognitivist's own commitments, then we might wonder why they have earned the right to serve as standards of evaluation for anyone else.
Understanding Normative Questions
We commonly ask ourselves what we should do (or think or feel) in a given situation. For non-cognitivists, there isn't anything we should do, really. (p.28) Doing different things will bring about different consequences, but no result is such that one ought to do it, since no result—no state of affairs in the world— could possess value or be obligatory. Value and obligation are normative notions that never refer. There is no such thing as normativity; we live in a value-free world, the world as science describes it. But then what is going on when we ask ourselves, in any given case, how we ought to proceed?
The expressivist analysis, presumably, will be that we are asking what the results would be were we to take certain courses of action, and then asking ourselves what attitude we would have toward the results. But that cannot be all; we weren't asking a descriptive question, but a normative one. We may be puzzled not about whether we would respond in a certain way, but whether it would be fitting or appropriate to do so.
For expressivists, the answer to that question will not be given by the truth about what has value, since there is no such truth. Instead, the answer must be rendered in terms of our other attitudes. But how is the answer to be delivered? Any response will invoke a normative constraint on attitude formation—it will tell us how our other attitudes ought to be used in shaping the ones under scrutiny. For if it only tells us how our attitudes actually influence other attitudes, then it isn't answering our question. We want to know whether the influence exerted is legitimate.
Let us suppose that non-cognitivists will avow the importance of consistency among attitudes. So knowing how to act will be a matter of determining the attitudes one has toward the act and its results, and then selecting that act about which we have attitudes that best cohere with our other attitudes. Assume for now that we can understand the notion of ‘best cohering’ in a wholly non-normative way. Still, we can always ask why we ought to select just those attitudes that best cohere with our other ones. The answer here, too, must refer to some function of our actual attitudes. We ought to value coherence because we do value what coherence makes possible, i.e. we have a favourable attitude towards how things go when we exemplify coherence in attitude. But of course we can always ask why we should have these further attitudes towards the things that coherence makes possible, and so on.
This picture generates three problems. The first is that the series of replies to these questions about our attitudes never addresses our fundamental concern: the appropriateness of the attitudes we have. Each question in the series is answered by reference to our existing attitudes; yet each question represents another variation on the one that keeps getting deferred, namely: are our existing attitudes fit to be the primary source of evaluation? Since evaluative attitudes are, for the cognitivist, just beliefs, cognitivists will assess the fitness (p.29) of the relevant attitudes by reference to whether they are true. Of course this isn't an easy matter—we have already seen the challenge to cognitivist epistemology, and will try to answer it in Part V—but it has the benefit of at least squarely addressing our normative concerns. The non-cognitivist, by contrast, continually refashions the question of what we ought to do so that its answer can be given just by an introspective psychological enquiry. But asking and answering normative questions does not seem to be the same thing as asking and assuring ourselves about the implications of our existing mental states.
This leads directly to a second, related problem. Suppose that we have traced the replies to our normative questions back to an attitude that is foundational: we can cite no further, more general attitude that justifies it, though many other attitudes are justified by being derivable from it. The reason we value coherence, for instance, is that when we exemplify it we tend to secure our own aims. As a general matter, we want to secure our aims. Why? We just do. If we repeat the question, we will get a causal story, which is entirely beside the point in this context. We aren't seeking an aetiology of the want, but a justification of it.
The problem for non-cognitivism is that when we reach a point of citing a ‘brute’ desire—a want or liking that is self-standing, whose warrant derives from no other pro-attitude—then we have identified something that lacks a justifying reason, and so is arbitrary. For it certainly isn't self-justifying—that concept hasn't much credence in a non-cognitivist picture. Yet basing evaluation ultimately on attitudes that are arbitrary is problematic. The arbitrariness isn't problematic because recognition of it will sap one's energy or cripple one's determination.11 It is problematic because it infects all justificatory efforts. If our evaluative attitudes rely for their justification on attitudes which themselves lack justification, then the whole network is corrupt.12 And the cognitivist is in better shape here, because her ultimate moral commitments can avoid arbitrariness if they are true, and if she believes them because (p.30) they are true. Many such commitments will fail to meet these two conditions, and this will have similarly serious consequences for the resulting network of moral attitudes. But if the cognitivist (who is not also an error theorist) is correct, then we can at least aspire to avoid arbitrariness, and some of us will succeed.
The last problem here is related to the arbitrariness worry. If evaluative attitudes can somehow be justified, their justification must be routed through other, existing attitudes. Yet evaluative networks differ markedly in their content from person to person. There is no reason to be optimistic that the very same set of attitudes will emerge in every person after all the internal tidying up is done. And this generates a relativism, for non-cognitivism lacks any basis for giving justificatory priority to one evaluative network over another. Each individual will, of course, very likely prefer her own set to another's. So each individual will probably not take a relativistic attitude toward her own values. But relativism may be true even if no one endorses first-order relativism (the theory that makes rightness relative to individual or social preference). Indeed, as we shall now see, expressivism seems forced to such a view, if it allows justification at all for one's first-order evaluative attitudes.
Expressivism implies that there cannot be just a single standard for determining the justification of evaluative attitudes. For who would set it? If the standard is set by any particular expressivist, then we can always ask why we should accept it. After all, the standard will be a normative one, and so, per expressivism, can't really be true, can be, in fact, only an expression of the particular non-cognitivist's attitudes. Why must everyone accept such strictures? But if no one person may set the standard, then either there are no standards (and so no justification of any evaluative attitudes), or everyone gets to set the standards (in which case we have relativism), or there (p.31) is a standard fixed independently of anyone's attitudes about what the standard should be (in which case we have realism). So one who wishes to endorse non-cognitivism is faced with the options of rejecting the justification of all evaluative attitudes, or embracing relativism. It isn't a happy choice.
Here we can see why the recent efforts of Blackburn (1998) and Timmons (1999) are bound to run into difficulties. Both thinkers want to ally their expressivisms with the possibility of moral truth. Blackburn characterizes his view as ‘quasi-realism’, and Timmons describes his ‘irrealist’ efforts as part of an ‘accommodation project’: both see the above-mentioned attractions of cognitivism and realism quite clearly, and they want to convince us that we can account for the objective pretensions of morality (Gibbard's fine phrase) while eschewing the metaphysics that has standardly underwritten them. As they see, most of the explanatory benefits that accrue to cognitivism—its superior ability to conform to our ordinary talk of moral truth, facts, knowledge and belief, our common-sense understandings of moral error, predication, and argument, etc.—rest on the truth-evaluability of moral claims. If non-cognitivism can take over this possibility without sacrificing its metaphysical and epistemological economy, then philosophers such as Timmons and Blackburn can have their cake and eat it, too.
The route they take to achieve this result is run via the deflationary theory of truth—to affirm a statement as true is merely to evince one's endorsement of it. On such a view, citing a judgement's truth is not recording some further fact or attributing an additional property, but merely signalling one's agreement with its content. The problem is that once we allow truth a place in ethics, then no matter how economical one's metaphysics, the ancillary views of the non-cognitivists commit them to ethical relativism.
Both Timmons and Blackburn are very hard on relativism. They each realize that relativism conflicts with the appearances: the semantic norms that constrain truth in ethics do not (even implicitly) relativize such truth to the attitudes of the speaker or the mores of her community. We judge our moral views to be true, simpliciter. The quasi-realist and accommodationist want to take such thoughts at face value. That doesn't seem possible.
To see why, note first that Timmons endorses a contextualist semantics according to which statements true in one context might be untrue in another. In the context of morally disengaged investigation, Timmons is a moral nihilist—there are no truths at all (1999: 244). We have to be speaking and thinking within a moral outlook in order for moral judgements to be true. (p.32) This is Blackburn's view as well (1998: 50).13 And from within those outlooks, people will usually assert their commitments in an unqualified way—they will insist on the correctness of such judgements for all, not just for those who share a way of life or the outlook whose constituent views are receiving expression.
As a record of moral phenomenology, this is surely accurate. But it doesn't seem sufficient to fend off the worry that this sort of view is a form of ethical relativism. There is nothing that makes moral judgements true, according to Blackburn and Timmons—no ‘external’ moral facts that moral judgements might accurately describe, nothing in virtue of which such judgements, when true, are true.14 Since that is so, there is no basis other than one's own outlook for evaluating the competing moral claims made by others with different outlooks. Though few will judge their own views to be true only relative to their own outlooks, the absence of any moral facts outside particular outlooks makes it the case that the judgements rendered within one outlook are no more true than those of a competing outlook. The views of each incompatible outlook are equally (un)true. This is relativism.
Blackburn and Timmons see relativists as making a first-order ethical claim that identifies the right-making feature of actions as their conformity to existing social norms. They don't themselves find much to recommend in such a view. But relativism's failure as an ethical recommendation must be measured by norms, and the non-cognitivists' settled view is that any such norms have no greater authority than the individual commitments that underwrite them. So another's commitment to a brand of relativism is not inferior to one's own rejection of such a doctrine, except by the lights of an antirelativist whose concerns exclude such a commitment. The standards by which we assess the propriety of a set of commitments are none other than those actually endorsed by different agents. Agents will endorse different standards of assessment, no one of which is allowed to be uniquely authoritative. It follows that a moral claim is true only relative to an endorsed standard. Hence relativism.
(p.33) So the addition of moral truth to the non-cognitivist's arsenal does not advance the cause very much. It creates the appearance of being able to accommodate much of morality's objective pretensions, but does so at the cost of the objectivity it had hoped to account for. As we have seen, this problem isn't restricted to those non-cognitivists who seek to incorporate the notion of moral truth. For a non-cognitivism that stays the traditional course, and resists talk of moral truth, is still beset by relativism about justification. On any non-cognitivist theory, the justification of a moral claim is nothing more than its conformity to a moral outlook with no intrinsic justification. The justificatory superiority of one outlook over another is a matter of its being endorsed as such. But different outlooks will receive endorsement from different people, and there is no intrinsically normative standard by which to register justificatory superiority. In other words, the justification of one's moral views is relative to one's outlook, and there is no stance-independent standard to evaluate the relative merits of these outlooks. So long as non-cognitivism entertains the possibility of justifying moral claims, it cannot escape relativism. The only alternative is to give up the possibility of ethical justification. Either way, non-cognitivists fail to accommodate a central feature of our understanding of moral thought.
A further problem for non-cognitivism stems from its rejection of the existence of irreducible normativity. According to non-cognitivists, to say that things are justified, obligatory, or good is not to describe things as having such features—there aren't really any such features—but to express a favourable attitude toward them. But can this attitude (or these different attitudes) be understood in a non-normative way? Here we confront two distinct but related charges of circularity.
The first claims that the attribution of propositional attitudes is itself a normative endeavour.15 Attributing such attitudes analytically commits one to normative claims, and conceptually presupposes an understanding of normative vocabulary. For instance, to attribute to an agent a desire to lose weight, and a belief that dieting is the only efficient way to do so, entails a commitment to the view that the agent ought to go on a diet. One can't sensibly make this sort of attribution without being committed in this way. But then we have a circle: to see someone as holding a moral judgement, one (p.34) must attribute to her some desire or favourable attitude. But to make such an attribution is itself to hold some normative commitment. Therefore the analysis of normative language itself presupposes notions of normativity. Therefore expressivists cannot make sense of ethical discourse without recourse to unreduced normativity. And this is contrary to the deepest motivations of non-cognitivism: to explain the world in a wholly naturalistic way.16
The second charge stems not from commitments that are presupposed in the attribution of intentional attitudes, but instead from the non-cognitivist analysis of normativity itself. To say that someone ought to Φ is to express a favourable attitude to Φ-ing. But ‘favourable’ is itself a normative notion. The challenge is to unpack it without the use of normative concepts. This may be impossible.
This isn't the very same problem as the one of distinguishing different attitudes to map onto the diverse moral predicates that we take for granted in moral talk. The present problem goes deeper. The challenge is to make sense of the fundamental concept of a pro-attitude, which is the key to the non-cognitivist's understanding of normativity. If expressivists are to sustain their critique of cognitivism, they have to show that moral judgements are something other than beliefs. For if they are beliefs, they might be true, they might yield knowledge, they might refer and so describe moral facts. Non-cognitivists contrast beliefs with what I have called favourable attitudes. These often pass for desires. But what distinguishes a desire or pro-attitude from a belief ? The best account that I know of relies on the notion of direction of fit.
There are actually many different accounts of direction of fit. Consider a prominent one, that offered by Michael Smith (1987, 1994: 113 ff.):
[A] belief that p tends to go out of existence in the presence of a perception with the content that not p, whereas a desire that p tends to endure, disposing the subject in that state to bring about p. (1994: 115)
As Smith recognizes, this view will not account for desires about the past, or desires that others undertake certain actions. But leave that aside. (Smith offers a solution to this problem in 1994: Ch. 4, n. 7.) We should also wonder whether the perception that Smith speaks of isn't just a synonym for belief, in which case Smith is explicating the notion of belief by relying on the very same notion, which seems problematic, even if it isn't his intention in this (p.35) passage to offer an analysis. Leave that aside as well.17 The real problem is that Smith's view does not succeed because he explicates the notion of direction of fit, and so the distinction between beliefs and desires, without recourse to normative concepts.
Beliefs cannot be identified solely by their tendency to extinguish upon a perception that their content isn't realized. In the first place, it isn't clear that an individual belief is the sort of thing that could have such a tendency. A particular belief that p either will or will not persist in the face of appreciation of contradictory evidence. The class of beliefs may exhibit such a tendency, but that is just shorthand for the claim that most beliefs do extinguish upon a perception that their content is false. But this point, which is surely true, does not give us any basis for making a determination in particular cases as to whether a person believes that p, as opposed to desires that p. For beliefs sometimes do persist even after the agent recognizes that the belief is false. And desires sometimes do extinguish after a perception that their content is unrealized. From the fact that one continues to hold that p, even after recognizing that not-p, one cannot infer anything about whether the agent's intentional attitude is one of believing or desiring that p. A man who perceives all of the excellent evidence for the straying affections of his wife may nevertheless continue to believe in her fidelity. Even if we can speak of a single belief's having a tendency to extinguish or persist, this man's attitude may have no tendency whatever to extinguish, despite his repeated perception that things are not as he takes them to be. Yet the attitude can be a belief for all that.
Since the point of introducing the idea of direction of fit is to supply a basis for distinguishing beliefs from desires, Smith's account will not do. And the reason is that his account fails to invoke normative standards in the explication of direction of fit. A belief that p is an attitude that ought not to persist after a non-discounted perception that not-p; a desire that p is an attitude that is not necessarily such that it ought to extinguish upon such a perception.18 This is (p.36) what licenses us in saying that the forsaken husband's belief was (epistemically) irrational; he failed to do what he ought to have done.
And now the problem for non-cognitivism. If we explicate the very idea of a pro-attitude by reference to normative standards, then the non-cognitivist cannot eliminate normativity from the world by analysing moral judgement as the expression of pro-attitudes. And this is a crucial failing for non-cognitivism, given its stated aims. According to non-cognitivism, our moral judgements are not trying to fit the world; they are not beliefs, because they are not trying to convey any information. And it's a good thing, too, since (according to non-cognitivists) there isn't any such information to be conveyed. There is no relevant world that a moral judgement might fit—a world of moral facts—that could match our description of it, that could make our moral judgements true or false. But if that is so, then we shouldn't conceptualize moral judgement as belief. And thus we should see it as the expression of pro-attitudes, i.e. attitudes that are essentially motivating, that essentially dispose us to do something. But these attitudes can be understood only by invoking normative notions. They are essentially attitudes defined by means of a normative function—that of being such that they ought to (or oughtn't necessarily to) extinguish under certain conditions. So we haven't really got rid of normativity after all. Thus the central aim of a non-cognitivist programme is unfulfilled, and we lose incentive to abandon or reconceptualize all of the phenomenologically appealing features that cognitivism about ethics entails.
The ontological motivations that prompt the development of non-cognitivism should incline the non-cognitivist to be suspicious of all normative claims. If brute normativity is a problem in ethics, it should be a problem anywhere else. And this means that the non-cognitivist must either try to reduce such normative notions as reasons, rationality, legitimacy, justification, relevance, appropriateness, and warrant, or supply a non-cognitivist diagnosis of our employment of them. There is no internal inconsistency in a non-cognitivist about ethics seeking reductions in other domains. But naturalistic reductions of normative concepts aren't promising (more on this in Part II), and non-cognitivists have traditionally relied on this to buttress their antireductionist, expressivist diagnoses.
Yet such diagnoses seem far-fetched when it comes to the normative concepts just mentioned. Are we really doing nothing more than expressing a non-cognitive commitment when assessing a belief as justified, an inference as warranted, an argument as sound? When a person believes a conditional and its antecedent, we say of her that she ought to believe the conclusion, that (p.37) she is warranted or justified in doing so, that she has good reason to accept the consequent. We certainly seem to be describing her situation, and conveying our assessments by means of beliefs. We think there really is such a thing as being epistemically justified; some agents and beliefs merit the tag, and when they do we speak the truth by describing them as justified. If non-cognitivism is correct, however, then this picture is all wrong. There really aren't any such things as reasons (for belief, for action, etc.). We talk and think as if there were, but there is nothing in the ontology that answers to this notion. What we do when we deploy the concept is to express a favourable attitude to something or other. And nothing more.
It is difficult to see such a view as anything other than deflationary. Though that would be perfectly in keeping with early non-cognitivist accounts, such as Ayer's, the most sophisticated recent efforts—those of Gibbard, Blackburn, and Timmons—seek to vindicate and preserve our ordinary understanding of moral thinking. Despite their intentions, however, it seems to me that in this respect, as in so many others, non-cognitivism requires quite serious revisions in our ordinary understanding of ourselves as moral agents engaged in moral deliberation.
Non-cognitivists charge that realists lack satisfying accounts of moral ontology, motivation, and epistemology. These charges are serious. I devote the bulk of Parts II, III, and V, respectively, to efforts to supply satisfying realist responses to these worries. If the replies have any merit, then we have good reason to prefer realism over non-cognitivism, because non-cognitivism faces many problems of its own.
Though I don't claim to have given an exhaustive inventory, or to have considered all of the non-cognitivist responses to the cited problems, the vulnerabilities of non-cognitivism seem real enough. It lacks a natural account of our talk of moral truth, moral knowledge, the capacity for moral error, the descriptive and assertoric appearance of moral discourse, and the logical forms that moral argument can take. Non-cognitivists haven't yet been able to account for the rich diversity of moral predicates, cannot take at face value what appear to be ultimate, irreducible normative questions, have difficulty escaping moral relativism, and rest their substantive moral convictions on arbitrary foundations. Non-cognitivists cannot allow for the conceptual possibility of an amoralist, and are faced with two distinct problems of (p.38) normative circularity: one to do with the analysis of normative language and the attribution of intentional attitudes, the other to do with the non-cognitivist analysis of normativity itself.
Admittedly, I have not gone into the sort of detail that would be required to conclusively establish the soundness of any of these criticisms. But the best of the extant non-cognitivist efforts (Gibbard 1990; Blackburn 1998; Timmons 1999) seem to me to offer less than compelling replies to these worries. This is merely autobiography; I don't try to substantiate this scepticism in what follows. So the proper attitude to take, even for those who accept most of what is to come, is that realism is only presumptively warranted. Whether it is superior to all forms of non-cognitivism depends not only on its merits (or failings), but on those of its non-cognitivist competitors.
This points up a limitation that should be confessed at the outset. No single work could reasonably attempt to render a thorough verdict on the victor in the metaethical stakes. It is surely hard enough, but also hardly sufficient, to develop a plausible metaethical theory that integrates views on metaphysics, philosophy of mind and language, action theory, rationality, and epistemology. One must also identify ineliminable errors in competing theories, and this is itself a task as taxing and extensive as that of developing the positive theory.
The battery of charges recorded above has done much to convince me of non-cognitivism's failings. But it would be premature to consider non-cognitivism false. It may be able to sustain its claims to an attractive metaphysics, epistemology, and theory of motivation. It may be able to defuse the criticisms I have just levelled. I don't believe it succeeds on either front. I do believe that it would take another book to show it so.
That said, the criticisms I've enumerated in this chapter do seem to me sufficient to create a strong prima facie case against non-cognitivism. But whether it is conclusive depends on the viability of non-cognitivist replies. It also depends, very importantly, on the plausibility of positive cognitivist theories. Let us turn our attention to these.
(1) This characterization of constructivism differs from that advanced by Brink (1989: 19–20). He sees constructivism as the view that there are moral facts or truths which are constituted by the evidence we have for them. Understood this way, constructivism has problems distinguishing between epistemic justification and truth; it implausibly claims that a moral judgement is true just in case, and because, the best evidence says it is. Brink rightly rejects such a view (1989: 31–5), but constructivists such as Firth (1952), Rawls (1980), Korsgaard (1996a), Carson (1984), Scanlon (1998), or Milo (1995) would reject the claim that moral truth is constituted by the best evidence for it, and so would reject the equivalence or identity between justification and truth that Brink attributes to constructivism. Because of this, I think it best to proceed with the characterization of constructivism I offer above, which incorporates the views that Brink subsumes within the class (those of James 1907; Bradley 1914; Pierce 1914; Blanshard 1939; Dummett 1963; Quine 1968; Putnam 1983; Rorty 1980), and also easily explains why the views of these and the philosophers mentioned above are neither realistic nor nihilistic.
(2) Moral facts are the things in virtue of which the truth conditions of assertoric moral claims are satisfied; moral facts are the truth-makers of true moral assertions. Understood in this way, moral facts are distinct from moral principles. These latter, as explained in Chapter 12, are best understood as conditionals. These conditionals (in conjunction with relevant non-moral facts) can explain why the moral facts are as they are. But the conditionals—the moral principles, standards, rules, or laws—are to be distinguished, in my taxonomy, from the particular moral facts that they help to explain.
If there are no moral agents, then, on a credible view of the content of moral standards, the antecedents of such standards will never be satisfied. And so, in the absence of moral agents, there will be no moral facts. But, for the realist, the absence of moral agents does not signal the absence of moral standards. These do not depend for their truth on the existence of those they might apply to.
(3) I owe this felicitous phrase to Ron Milo.
(4) How should we classify views that make the determination of moral truth dependent on God's ratification of the relevant standards? Such views seem most naturally grouped with realism, even though, on these accounts, there is a constructive function that explains the correctness of the proper moral standards. For my purposes it doesn't much matter how we solve this taxonomic problem. All divine command theories incorporate a constructive function, to be sure, but it is also the case that the relevant standard-setting attitudes are not those of any human being—the attitudes (no matter how refined) of us mere mortals do not do the relevant work. Place these theories as you like; I don't think that anything that follows will be importantly affected by assigning such views to one camp rather than another.
(5) I discuss some of these in Chapter 10.
(6) Realists may concede the existence of a class of unknowable moral truths, but this limitation must be accompanied by an explanation. For instance, some realists have a particular view about the sorites paradox which leads them to the view that there is a precise though unknowable point that marks the absolute division between the point at which a vague predicate does and does not refer. There is a precise number of grains of sand that makes a heap, or a perfectly determinate division between acts that are right and wrong. This is the epistemicist view of vagueness (see Cargile 1969; Sorenson 1988; Williamson 1990, 1998). Realists don't have to take this line, but some do. The key point is that if they do, they owe an explanation for the unknowability of the truths in the restricted domain; as a general matter, realists can reject the view that epistemically ideal agents might remain ignorant of truth.
(7) For more on naturalism, including a discussion of whether we can give an adequate account of what it is to be a natural property or fact, see Chapter 3.
(8) Matters are complicated here by the recent attempts of Timmons (1999) and Blackburn (1998) to ally their non-cognitivisms with the claim that moral judgements are truth-evaluable. I discuss difficulties with their proposals below. For now, it is sufficient to note that both thinkers accept the claim that moral judgements do not describe anything, for there is no moral reality awaiting description. And since theirs is not an error theory, it isn't as if moral judgements mistakenly or inadvertently (attempt to) describe any other sort of reality. For Timmons and Blackburn, as for all other non-cognitivists, there is nothing that can make moral judgements true—no moral facts or moral reality that they could possibly correctly represent, nothing they are true of. Such a view, in conjunction with their rejection of an error theory, is enough to distinguish them from both constructivists and realists.
(9) For attempts to solve the problem, see Gibbard 1990: 83–102; Blackburn 1988, 1993b, 1996: 84–5, 1998: 68–77; Stoljar 1993: 91–7. For criticisms of their efforts, see Schueler 1988; Hale 1993; van Roojen 1996; Unwin 1999. Some critics see potential for solving the Frege–Geach worry, but at the cost of the non-cognitivism that is at the heart of these expressivist theories: see Sinnott-Armstrong 1993: 299–301; Wedgwood 1997; Dreier 1999. The cumulative force of these critical papers entails a dilemma: either remain an unreconstructed non-cognitivist, in which case the Frege–Geach problem impedes hopes for logical moral argumentation, or solve that problem, but only by introducing modifications that transform the expressivism into a version of cognitivism.
(11) Wiggins (1976) levelled this charge against non-cognitivism. See Blackburn 1987 and Timmons 1999: ch. 4 for replies. At best, Wiggins's charge reveals the upsetting pragmatic implications of wholeheartedly embracing non-cognitivism. But even if Wiggins's take on our psychological propensities is correct, this would not show that non-cognitivism is false (though it would be further evidence that our moral practice proceeds as if it were). Compare the predictions of dismay and lassitude that would eventuate were most of us to embrace hard determinism. The dire consequences of such an embrace would not refute determinism, though such consequences would give us some (non-epistemic) reason to stop believing it.
(12) The best effort to resist this conclusion is given by Timmons (1999: ch. 5), who argues that unjustified beliefs may nevertheless confer positive justification on other beliefs. Whatever plausibility such a claim has depends on his contextualist thesis that these ‘unjustified justifiers’ can, in other contexts, receive justification. In what follows, I try to show that non-cognitivists lack the resources for escaping from the arbitrariness worry. Here is a preview: to do the work required of them, these unjustified justifiers must receive justification in some context. The relevant contexts are two: either a philosophical one, in which we refrain from substantive ethical engagement, or a morally engaged one, in which we advocate some first-order moral views. According to non- cognitivists, there can be no such thing as genuine justification of normative views in philosophical contexts. But (I claim) any non-cognitivist justification that arises in engaged moral contexts will entail a form of self-intimating relativism. So the introduction of a contextualist epistemology will not save the non-cognitivist from arbitrariness worries.
(14) Indeed, it isn't at all clear what it would be for the right-hand side of a Tarski sentence to be realized in the moral case. According to Blackburn and Timmons, ‘abortion is wrong’ (for instance) is true if and only if abortion is wrong. But they give us little help when it comes to understanding what it would be for the right-hand side to hold—what sorts of conditions would have to be fulfilled, what kinds of states of affairs would have to be realized in order for anything like that to be the case. Given their naturalism, their rejection of any representational function of moral discourse, and their rejection of moral properties (of the sort whose instantiations could serve as truth-makers for moral claims), the nature of the right-hand side of moral equivalences is quite mysterious.
(17) Humberstone (1992) offers a good discussion of the various ways one can read Smith (1987) on this matter. Humberstone thinks that Smith isn't attempting to offer analyses of belief and desire, and so can avoid the circularity worry. Copp and Sobel (2001), by contrast, press just this concern, and argue that it is fatal to direction-of-fit analyses of belief and desire.
(18) The ‘non-discounted’ aspect of the perception is needed in order to accommodate an example of Copp and Sobel (2001: 47). They present a case in which a driver ‘sees’ a patch of wet and wavy road ahead, though one continues to believe that the road is, as a matter of fact, flat and dry. This happens all the time in very hot conditions. An experienced driver will not allow such perceptions, often encountered under similar conditions, to extinguish what is really a belief. Nor ought such perceptions to eradicate the belief. So not every perception with a content that not-p is sufficient to serve as the basis of a test for distinguishing beliefs from desires that p. Only those perceptions that an agent does not discount can play such a role.