Ethical Theory and Moral Practice

, Volume 14, Issue 4, pp 445–455

Expressivism and Moral Dilemmas: A Response to Marino


    • Department of PhilosophyUniversity of Leeds

DOI: 10.1007/s10677-010-9255-6

Cite this article as:
Baker, C. Ethic Theory Moral Prac (2011) 14: 445. doi:10.1007/s10677-010-9255-6


Simon Blackburn’s expressivist logic of attitudes aims to explain how we can use non-assertoric moral judgements in logically valid arguments. Patricia Marino has recently argued that Blackburn’s logic faces a dilemma: either it cannot account for the place of moral dilemmas in moral reasoning or, if it can, it makes an illicit distinction between two different kinds of moral dilemma. Her target is the logic’s definition of validity as satisfiability, according to which validity requires an avoidance of attitudinal inconsistency. Against Marino’s arguments, I contend that expressivists following Blackburn are able to show how we appreciate the validity of arguments found in dilemma-contexts, and that Marino’s argument concerning the distinction between contingent moral dilemmas and logical moral dilemmas rests on a mistake concerning the logical representation of a contingent dilemma.


BlackburnExpressivismConsistencyMoral dilemmasFrege–Geach problem

Expressivists about morality—those who hold that moral judgements are expressions of attitudes and feelings rather than truth-apt assertions—must give an account of how we are able to use moral statements in valid arguments. This is known as the ‘Frege-Geach problem’ in virtue of its ancestry.1 Simon Blackburn’s seminal paper ‘Attitudes and Contents’ (1988) responds to this problem by translating moral arguments into a ‘logic of attitudes’ (LA) according to which the validity of moral arguments lies in the unsatisfiability of a state of mind which denies those inferences. In other words, denying valid moral inferences will lead one to hold inconsistent attitudes and endorse unattainable goals.2 However, Patricia Marino (2006) has recently argued that grounding logical normativity in the avoidance of inconsistent attitudes is not a viable strategy.3 She claims that there are cases of valid moral reasoning—cases in contexts involving moral dilemmas—which are excluded if we define validity in terms of avoiding inconsistent attitudes. The upshot of Marino’s arguments is this: if LA cannot justify its exclusion of arguments found in moral dilemma-contexts then it is incomplete, in that there are some valid inferences which it cannot account for; but if it can justify the exclusion, it is unsound in that it warrants some inferences which are not in fact allowable.

In this paper I contest Marino’s arguments—I aim to show that moral dilemmas do not present a decisive obstacle to the success of LA. In section 1 I present Marino’s ‘argument from dilemmas’, which aims to demonstrate the incompleteness of the logic of attitudes by showing that moral dilemmas are not only allowable, but are a crucial component of moral reasoning. In section 2 I explain how the proponent of LA can make sense of the role of moral dilemmas in moral reasoning, and go on to rehearse the standard Blackburnian explanation of why moral dilemmas constitute invalidity (roughly, because they are unable to be action-guiding). Marino believes that this latter explanation is untenable because it leads LA into further troubles concerning the distinction between contingent and logical dilemmas; in section 3 I present these worries before arguing that they rest on a mistaken logical characterisation of contingent moral dilemmas. I propose an alternative characterisation and show how it interacts with the principles of LA in such a way as to undermine Marino’s argument.

1 Moral Dilemmas and Inconsistency

As is familiar, Blackburn’s logic of attitudes responds to the Frege-Geach problem by arguing that the validity of moral inferences is grounded in the satisfiability of the sets of attitudes expressed by those inferences.4 Patricia Marino criticises this strategy, arguing that we should not ground logical normativity in satisfiability, and that attitudinal inconsistency is not, in general, to be avoided. While Marino is not the first to have questioned whether attitudinal inconsistency is a strong enough form of inconsistency to base a logical system upon,5 her criticisms have a novel subject: that of moral dilemmas, or conflicts of moral requirements. Her initial contention is that LA is incomplete, because there are examples of valid reasoning, examples involving moral dilemmas, whose validity it cannot underwrite. In this section I outline the main points of her argument.

Consider MMP, an instance of ‘moral modus ponens’ (that is, a modus ponens inference given in a moral context):
  • MMP.

  • P1. If I ought to give money to the homeless, then I ought to give money to Timothy.

  • P2. I ought to give money to the homeless.

  • C. Therefore, I ought to give money to Timothy.

In Blackburn’s expressivist system, ‘ought p’ is replaced by ‘endorsement of the goal that p’, and is represented notationally in LA as H!(p) (Blackburn 1988, p.513). MMP is therefore schematised as follows:
  • P1. H!(p)→H!(q)

  • P2. H!(p)

  • C. H!(q)

Blackburn holds that MMP is valid because it avoids attitudinal inconsistency: if I deny the inference, I am unable to form a consistently realisable ideal. Since P1 is classically equivalent to ‘¬H!(p) v H!(q)’,6 accepting the premise ties me to a ‘tree’ of commitments: that is, endorsing the disjunction commits me to endorsing one disjunct should the other become untenable. Here, P2 makes the first disjunct untenable since they are contradictories. So accepting P1 and P2 forces me to accept the second disjunct—H!(q)—which also forms the conclusion of the argument. To endorse P1 and P2 whilst rejecting C is to hold a set of attitudes which is not satisfiable: that is, there is no ideal in which the goals expressed by those attitudes are all realised.7 To put it another way: to endorse P1 and P2 whilst rejecting C is to say that you are tied to a disjunctive tree of commitments, whilst also saying that you are not tied to that same tree. Validity in LA is grounded, then, in the avoidance of attitudinal inconsistency—a moral inference is valid iff we cannot deny it whilst remaining consistent in our attitudes, and MMP is a good inference according to LA because denying it will lead one to have inconsistent attitudes (Blackburn 1988, p.512ff).

Marino highlights a problem faced by this account of validity with respect to moral dilemmas. A moral dilemma is ordinarily thought of as a situation in which I ought to do p and I ought to do q but I am unable to fulfil both of these obligations—I cannot (in some sense) do both p and q.8 In our discussion of expressivism, a moral dilemma should be thought of as a situation in which my attitudes of approval and disapproval conflict—where I have attitudes of approval towards multiple actions which cannot all be performed; where I have attitudes of disapproval towards multiple actions, of which at least one must be performed; or where I have conflicting attitudes of approval and disapproval towards the same action.9

Marino (2006, pp.521–522) uses an example of MMP considered in the context of a moral dilemma. Imagine that I have an existing attitude against giving money to Timothy; for example, I know that he will put the money towards his drug habit. Since I have an attitude against Timothy’s narcotic abuse, and I know that giving him money will fund this activity, I form an attitude against giving him money.10 But at this point in my reasoning, someone presents MMP to me. I accept the premises and am convinced by the argument, and thus form an attitude for giving money to Timothy. My attitudes conflict and I am faced with a moral dilemma over whether to give money to Timothy.

In this case LA convicts me of an error of reasoning because I possess inconsistent attitudes: I both approve and disapprove of the goal of giving money to Timothy. This is an unsatisfiable set of attitudes, and as such my line of reasoning must have been somehow flawed. Marino makes two related points against this conclusion. Firstly, she claims that line of reasoning evidenced by the example is not in fact reprehensible, since there is nothing in principle wrong with a state of mind which embodies inconsistent attitudes—in other words, she takes her example to show there are ways of having inconsistent attitudes of approval and disapproval which are perfectly rational. As such, we should think (contra LA) that my reasoning is valid in the above case. Secondly, she says that one “can endorse logically incompatible goods and still use arguments”; that is, “[the arguments may] seem valid” (Ibid.). I can be convinced by the validity of MMP and still reason well generally, despite the presence of attitudinal inconsistency. But it seems that according to LA, I shouldn’t be convinced by the argument, since an argument is valid if and only if attitudinal inconsistency is avoided (Ibid.).

There is a puzzle here for LA, then—its logical principles predict results which are not evidenced in the case of moral dilemmas. Since validity and attitudinal inconsistency can coexist, the argument goes, validity cannot require the avoidance of attitudinal consistency, and thus satisfiability is not a good ground for logical normativity. Our ability to reason well and appreciate the validity of arguments in the face of moral dilemmas seemingly undermines the foundations of LA, since Blackburn’s logic cannot account for all the data of everyday practical experience of morality. This is surely a serious indictment, as it would be on any theory which fails to account for the relevant data.

2 Moral Dilemmas and Moral Reasoning

As we have seen, Marino disputes the completeness of the logic of attitudes: she claims that it cannot underwrite all valid moral inferences. In particular, it is valid arguments in contexts involving moral dilemmas which LA putatively cannot support.

In responding to Marino’s argument, the first step is to account for the role of moral dilemmas in moral reasoning. That is, we must explain how it is possible to be convinced by valid arguments which appear in dilemma-contexts; an ability which LA seems to shed doubt upon. With respect to the ‘Timothy’ example given above, I contend that we can appreciate the validity of MMP, despite its context in an unsatisfiable set of attitudes, because we can appreciate that the MMP argument is valid when considered on its own. I can consider the modus ponens inference independent of my wider attitudinal context—the context containing the contrary attitude against giving money to Timothy—and, in doing so, I am perfectly able to appreciate the validity of the modus ponens argument. This approach seems prima facie supportable, since it’s not obvious that we must always reason on the basis of all our attitudes at once. When examining the modus ponens inference, I need not actively entertain my conflicting attitudes. I can appreciate the validity of the argument by considering it in isolation from my attitude which contradicts the conclusion. Insofar as Marino’s objection concerns how we can reason well in the face of the attitudinal inconsistency exemplified by a moral dilemma, it can be met by appeal to our ability to reason on the basis of select subsets of our attitudes.

However, this approach only deals with one of Marino’s points. While it explains how we can appreciate the validity of arguments like MMP when they are considered in dilemma-contexts, LA nevertheless tells us that arguments made in these contexts are invalid. We might think that the exclusion of arguments found in dilemma-contexts from validity can be justified simply by appeal to the fact that the attitude-contexts exemplified by these arguments evidence inconsistent attitudes. But for Marino, this is insufficient: she claims that “there doesn’t seem to be anything wrong with my attitudes or my thinking” when I am faced with a moral dilemma, and nor, by implication, when I hold inconsistent attitudes (Ibid.). If this is so then the LA’s notion of validity is objectionably restrictive—we are left a logic which is too strict about what counts as valid.

If expressivists wish to uphold their claim that moral arguments which lead to unsatisfiable attitudes are invalid, they must—given that all arguments involving moral dilemmas are of this kind—tell us why moral dilemmas are, in general, to be avoided. So how can LA justify its exclusion of arguments in dilemma-contexts from validity? In answering this question we should reconsider some of Blackburn’s well-known arguments concerning the inconsistency of attitudes. When a person denies moral modus ponens, for example, that person possesses inconsistent attitudes—and since attitudes are related to goals on Blackburn’s picture, the person’s goals cannot all be realised. The trouble with this kind of inconsistency is analogous to the trouble with holding inconsistent beliefs:

The man who believes that it is raining and that it is not is badly placed to act if he wants, say, to avoid getting wet. But so is the man who believes that it is raining but wants to get wet and not to get wet. (Blackburn 1988, p.509)

Since inconsistent attitudes cannot be action-guiding, we should avoid them—and any moral argument which results in an inconsistent set of attitudes should not count as a good inference. Given the principle that our attitudes should be action-guiding, we should aim to avoid inconsistency in those attitudes. Applying this to dilemma-contexts, we should say that the sets of attitudes evidenced in such contexts are to be avoided because they cannot be action-guiding in virtue of the inconsistency inherent in such sets of attitudes.11

Marino considers this kind of response, but holds that it leads to a further serious problem for LA. Specifically, she claims that the success of the above response entails an illicit logical distinction between two different types of moral dilemma (Marino 2006, pp.524–528). I move on now to consider and reject Marino’s argument for this further obstacle to LA.

3 Contingent Dilemmas vs. Logical Dilemmas

Marino argues that accepting the invalidity of arguments found in dilemma-contexts leads LA into its own kind of inconsistency. Specifically, she argues that LA makes an illicit distinction between ‘contingent’ moral dilemmas and ‘logical’ moral dilemmas, and thus cannot fully support the claim that dilemmas result in logical errors. To roughly define our terms: a contingent moral dilemma is one in which the attitudinal conflict arises only as a matter of accidental fact. For example, imagine that I want to visit both my ailing mother and my ailing brother, but I only have the wherewithal to visit one of the two. The unfortunate state of affairs which prevents me from fulfilling both horns of this dilemma—my lack of funds—is plainly contingent, and so we call this a contingent moral dilemma. A logical dilemma, on the other hand, is one in which my attitudes conflict as a matter of logic—for example, if I both want to visit my ailing mother and also want to not visit her.12 On the basis of this distinction, Marino offers an argument against the strategy pursued in section 2. I will attempt to show that her argument rests on a misunderstanding in how contingent dilemmas should be logically analysed by LA, and thus that her argument provides no obstacle to the Blackburn-style strategy sketched above.

In order to fully comprehend Marino’s argument and the issues that arise from it, we will need to briefly revisit the semantics for LA as presented in Blackburn. As we have seen, validity in LA is defined in terms of satisfiability—and a set of sentences is unsatisfiable iff every route of approximation from that set to a final ideal leads to a contradiction. The final ideal is an unrealised yet possible ‘world’ that can be consistently thought of (Blackburn 1988, pp.513–514). In order to discover whether a set of attitudes L is satisfiable, we must perform a process of “step-by-step idealisation” to verify that the set allows us to imagine a consistent ideal (Ibid.). This process begins with the step from L to a second set of attitudes, L*—where L* is the next level of approximation to the ideal from L. Blackburn gives four rules which govern this and subsequent progressions:
  1. 1.

    If H!(p) is a member of L, then p will be a member of L*. (My approval of p implies that a world which includes p will be closer to the ideal.)

  2. 2.

    If H!(p) is a member of L, then H!(p) will also be a member of L*. (This reflects the idea that if a goal is laudable, then it’s still a laudable goal in a world in which it is achieved.)

  3. 3.

    If T!(p) is a member of L, then the set of next approximations to the ideal must contain at least one set, L* which has p as a member. (This is because T!(p), or tolerance of p, says that p is consistent with an ideal world.)

  4. 4.

    If p is a member of L*, then p will also be a member of future approximations to the ideal L**, L***, etc.13


The ‘ideal’ is reached when a subsequent approximation based on these rules doesn’t produce any new sentences. Where L contains disjunctions (including material conditionals, which are equivalent to disjunctions) we can form different routes to an ideal, as opposed to a single route to the ideal.

Let us return to Marino’s argument, which begins by considering whether the principles of deontic logic might undermine Marino’s original argument from section 1. She observes that the ‘closure principle’ of deontic logic—that is, if I ought to do p, and doing p involves doing q, then I ought to do q—is not valid in LA. If we take H!(p) to approximate ‘I ought to do p’, then the principle becomes (pq)⇒(H!(p)→H!(q)), which is invalid, because {(¬p v q), H!(p), T!(¬q)}—the set of sentences comprising the affirmation of the premises together with the denial of the conclusion—is satisfiable (Marino 2006, pp.524–526).14 However, the principle of deontic consistency—H!(p)⇒¬H!(¬p)—is valid, since {H!(p), H!(¬p)} is clearly unsatisfiable. Now Marino notes that the form of the ‘Timothy’ example from section 1—in which the dilemma faced is a contingent one—relies on the principle of closure, since I form an attitude against giving Timothy money because giving the money implies funding his narcotic abuse.15 However, if the closure principle fails, then I cannot validly infer my attitude against giving money to Timothy. So this case—and, more to the point, contingent dilemmas in general—need not, unlike logical dilemmas, be unsatisfiable according to LA. That is, it need not evidence an inconsistent set of attitudes. To see why, imagine that we have the set of sentences {H!(p), H!(q), (p→¬q)}—the third member of which is supposed by Marino to capture the essence of a contingent dilemma (Ibid.) Since unendorsed facts fall away upon the first approximation, the conditional is lost, and we’re left with just {H!(p), H!(q), p, q}*, which is a consistent set of attitudes. Because the principle of closure is not in play—that is, the conjunction of H!(p) and (p→¬q) does not imply H!(¬q)—we cannot infer a conflict of attitudes. Contingent dilemmas are satisfiable, then, and do not exemplify logical errors according to LA. Logical dilemmas, on the other hand—those which instantiate sets of attitudes of the form {H!(p), H!(¬p)}—can easily be shown to be unsatisfiable by way of the principle of deontic consistency.

This bears on the dialectical situation as follows. If Marino is correct then we cannot accept the comments of Blackburn’s that I reported in section 2; i.e. we cannot accept the incoherence of inconsistent attitudes as an explanation of why dilemma-contexts constitute errors. LA is unable to live up to these comments in the light of the failure of closure—it cannot exclude every dilemma-context from its notion of validity, since contingent dilemmas need not embody inconsistent attitudes. Therefore, the success of my section 2 conclusions, coupled with the success of Marino’s argument here, would entail that LA is unsound in that it underwrites the validity of inferences which are not in fact allowable. Now we might think that this line of thought can be easily countered by Blackburn; that he can accept Marino’s contention that contingent dilemma-contexts are consistent whilst logical dilemma-contexts are inconsistent, and take this to show that contingent dilemmas are allowed on LA after all even though logical dilemmas are not. But Marino claims that the problem is deeper than this—that it is a desideratum of an account of contingent and logical dilemmas that they not be given differential logical treatment, since they feel the same to the agent:

From the point of the agent, logically conflicting attitudes are similar to contingently conflicting ones. If I want to eat cookies and I do not want to eat cookies, I face the same difficulty as I do if I want to have cake and I want to have ice-cream but I am only allowed one dessert. In each case, I face a puzzle over what to do. (Marino 2006, p.525)

If Marino’s argument forces LA into making this illicit distinction between contingent and logic moral dilemmas, then serious doubt is cast on its viability. However, I contend that Marino’s argument here is deficient—specifically, that it misinterprets the logical structure of a contingent dilemma. Marino represents the set of attitudes evidenced by a contingent dilemma as {H!(p), H!(q), (p→¬q)}. According to her interpretation, what makes this set of sentences a dilemma, over and above the attitudes in favour of p and q, is the conditional ‘(p→¬q)’ (which is equivalent to ‘¬(p & q)’). This interpretation of the dilemma is correct insofar as doing q excludes doing p—but, as it stands, it is too weak. ‘¬(p & q)’ just says that at least one of p or q is false; but in a contingent dilemma, it is not merely the case that either p or q fails to hold. Rather, the trouble is that a conjunction of p and q is not practically attainable. In other words, when I am faced with a contingent dilemma with respect to p and q, I am faced with a situation in which I am unable to make p and q true together—and our logical characterisation of the contingent dilemma must take account of this inability. ‘¬(p & q)’ itself indicates nothing of this practical unattainability, which is a crucial element of a contingent dilemma. How, then, can ‘¬(p & q)’ be strengthened then in order to meet this desideratum? One way is to apply a modal operator to the negation, that of practical necessity16: ‘¬(p & q)’ is practically necessary in that p and q are not true together at any world which I can bring about.17 That is, ‘¬(p & q)’ is true, and there is nothing I can do about it. The issue isn’t simply that I do not give Timothy money without his spending it on drugs; more than this, there is no world that I can bring about where I give Timothy money without his spending it on drugs.18 Notationally, I will express practical necessity as ‘□p’. ‘□p(p)’ behaves like the standard modal operators, in that ‘□p(p)’ is a well-formed formula of LA iff ‘p’ is also a well-formed formula. Taking into account the practical necessity of the dilemma forces us to revise the notational interpretation of a contingent moral dilemma as follows: {H!(p), H!(q), □p(p→¬q)}.

The presence of a practical necessity operator in the set of attitudes changes how the idealisation process proceeds. Ordinarily, when we are analysing the satisfiability of a set of sentences, unendorsed facts about the world do not carry over to the first approximation to the ideal, and it is on this basis that ‘(p→¬q)’ does not (on Marino’s interpretation) carry over to the first level of approximation.19 This feature of the idealisation process reflects the idea that, when imagining the ideal, we shouldn’t let avoidable and unfortunate facts about the way the world happens to be get in our way (Blackburn 1998, p.514). But facts about what is practically necessary are not avoidable from the perspective of an agent; that is exactly what the operator is meant to capture. Because of this, formulae which are subject to the □p operator should not fall away at the first level of approximation to the ideal. We need a supplementary idealisation rule, taken along with Blackburn’s rules 1–4, to capture this:
  1. 5.

    If □p(p) is a member of L, then p will be a member of L*.20


Taken along with rule 4, this guarantees that p will be a member of each subsequent approximation to the ideal. Rule 5 reflects the idea that we should require our imagining of the ideal to remain within the bounds of practical possibility: if attitudes are to be understood as closely related to courses of action, as Blackburn holds, then it makes little sense for us to imagine ideals which are impossible for us to bring about. To do so cannot be action-guiding. Further support for rule 5 can be garnered by noting the untenability of a weaker alternative: that we only require our imagining to remain within the bounds of physical possibility (i.e. accordance with the laws of physics). This is surely far too liberal a principle to constrain our imagining: certain interpretations of contemporary physics allegedly tell us that it’s physically possible for a person to ‘quantum-tunnel’ through a brick wall; but we surely shouldn’t allow an ideal appealing to a phenomenon like that to be satisfiable. We need a modality stricter than physical possibility to constrain our imagining of the ideal, then—and practical possibility seems like a good candidate for such a notion.21

Let’s apply this to our contingent moral dilemma case: according to rule 5 (in conjunction with rule 4) □p(p→¬q) means that (p→¬q)—and thus, equivalently, ¬(p & q)—must be a member of all subsequent branches in the idealisation process. How does this impact on the satisfiability of contingent dilemmas? We have, in such cases, the following set of sentences: {H!(p); H!(q); □p(p→¬q)}. p and q must both appear in each subsequent approximation to the ideal because of the presence of H!(p) and H!(q). But because of the □p operator, ¬(p & q) must be true on each branch, and thus every route to an ideal must also contain either ¬p or ¬q. This means that every branch must lead to a contradiction, and thus that the set of sentences is unsatisfiable. No consistent ideal is imaginable—so each contingent dilemma must, contra Marino, result in an unsatisfiable ideal. There is, therefore, no distinction—illicit or otherwise—between the results that LA provides for contingent and logical dilemmas, since both are unsatisfiable according to LA.22

Before considering how Marino might respond to this strategy, I must compare it to (and distinguish it from) a modification to LA proposed by Bob Hale which is similar in spirit.23 While considering how LA should account for mixed conditionals—that is, those with a descriptive antecedent and an evaluative consequent (or vice versa) such as “if Timothy has no money then he ought to be given some”—Hale proposes that the idealisation process should hold some of the subject’s beliefs fixed, in addition to their attitudes (Hale 1993, pp.355–358). On rules 1–4, the mixed conditional would fall away at the first level of approximation since its main connective is not an attitude-operator—it is of the form (p→H!(q))—but Hale’s proposal is that the conditional should be taken as non-negotiable in the idealisation process (Ibid.) This will require a rule analogous to rule 5, in that such beliefs should be held fixed at each stage of the idealisation process.24 This can be related to our contingent moral dilemma case as follows: if we take the conditional element of the contingent dilemma— ‘(p→¬q)’ —to be a fixed belief, then we can derive the same result as I have above, namely that contingent dilemmas are inconsistent, without introducing any modal claims. So if this addition is already mandated by the phenomenon of mixed conditionals, why should we add an additional modal component to LA?

Using Hale’s ‘fixity of beliefs’ strategy does get the same results here, but it will not allow us to rectify the main lacuna in Marino’s reasoning: namely, the failure to correctly represent contingent dilemmas in the logical notation. In order to apply the belief-fixity strategy without appealing to practical necessity, we would presumably represent the conditional element of the dilemma as β(p→¬q), where β(p) denotes that p is a belief fixed for the purposes of idealisation. With a rule parallel to rule 5 for the β operator, the idealisation process would then proceed exactly as I have described above. But this logical representation of the contingent dilemma faces the same problem that I raised above: β(p→¬q) does not capture what it is to face a contingent moral dilemma. Leaving aside matters about the necessity and sufficiency of belief in this conditional element for the status of being in a moral dilemma, this characterisation is fully subject to my worries from above—when I believe I am in a moral dilemma, I do not simply have a “germane factual belief” that at least one of p and q is false; I believe that a conjunction of p and q is not practically attainable (Ibid.). So it seems that we can only incorporate the belief operator in a correct representation of the contingent dilemma if we also incorporate the practical necessity operator, i.e. β□p(p→¬q). This shows that Hale’s fixity strategy does not pose a direct alternative to my introduction of the practical necessity operator to address Marino’s argument.

Some of Marino’s comments suggest that she takes my proposed revision of LA to be unavailable to Blackburn. She writes that, when we say that Blackburn’s rules make it inconsistent to approve of two states which could not coexist,

this must be read as “states of affairs that could not coexist” in any logically possible world. If we read it as “states of affairs that could not coexist in our world,” or “states of affairs that could not coexist in any possible world” where possibility is understood as physical possibility, or conceivability, then it is false. (Marino 2006, p.525)

If correct, this will surely apply to my theses concerning practical possibility as well as physical possibility and conceivability, and therefore undermine my rescue of LA. Marino claims that this result follows from Blackburn’s definition of unsatisfiability:

A set of sentences L is unsatisfiable iff each route to a set of final ideals S results in a set of sentences S one of whose members contains both a formula and its negation. (Blackburn 1988, p.514, emphasis mine)

But Marino’s conclusion does not necessarily follow from this statement. While it is true that one way for a set of sentences to be unsatisfiable is for it to contain endorsements of two logically conflicting actions, such as p and ¬p—that is, attitudes which cannot be realised in any logically possible world—my above points on practical necessity show that this is not the only way for an approximation to the ideal to contain “both a formula and its negation”. Marino’s conclusion follows only if the idealisation process entitles us to disregard at the first step any fact about the world towards which we express no attitude. If there are facts about the world which we are not entitled to disregard in our process of approximation to the ideal, then contradictions—“a formula and its negation”—can be derived when attitudes conflict with these unavoidable facts about the world. As I have argued above, we are not entitled to disregard truths about what must be the case—in the sense of practical necessity—when we undertake the idealisation process. This fits in with Blackburn’s wider theses, in that the avoidance of inconsistency is meant to assure that attitudes will be action-guiding. Imagined ideals which appeal to practical impossibilities cannot be action-guiding, and thus we cannot disregard these impossibilities in our imagining of the ideal.25

The upshot of this is that LA gives the same logical treatment to contingent dilemmas as to logical dilemmas. Because of the practical necessity inherent in contingent dilemmas, sets of attitudes evidenced by such contexts are, just like those evidenced in logical dilemmas, unsatisfiable. This shows that the Blackurnian response to Marino’s argument elucidated in section 2 does not fall foul of the contingent/logical dilemma distinction. All moral dilemmas come out as unsatisfiable according to LA, and thus the logic is not unsound. In securing this conclusion, LA is rescued from an argument which purported to show that it cannot account for our practical experience of morality, this time in terms of the relationship between contingent and logical moral dilemmas. As I have argued, LA can support the putative data that our phenomenological experiences of contingent and logical moral dilemmas are similar enough for us to expect no deviation in the logical results that LA provides for the two varieties of dilemma.

4 Conclusion

To recap: Marino’s argument against LA’s principle of attitudinal inconsistency is that either it cannot account for examples of valid reasoning in the context of moral dilemmas or, if it can, then it makes an illicit distinction between contingent and logical dilemmas. I argued in section 2 that LA can account for our ability to reason well in dilemma contexts. What’s more, I have argued that Blackburn’s comments on the error of holding inconsistent attitudes do not fall foul of the distinction between contingent and logical moral dilemmas, so long as we correctly represent contingent dilemmas in the notation of LA and give due attention to the modal element of these dilemmas. If these arguments are sound then Marino’s argument fails, and moral dilemmas do not present any distinctive problems for the foundations of LA. While there may be other reasons to call into question LA’s reliance on the principle of attitudinal inconsistency, considerations about moral dilemmas do not give us cause to do so.


See Geach (1960, 1965).


Blackburn’s expressivist system is one among several extant accounts. Notable alternatives have been developed by Gibbard (2003), Horgan and Timmons (2006) and Schroeder (2008), among others. Blackburn’s account has been subject to much critical attention—for instance, van Roojen (1996) presents influential arguments against Blackburn which are in some ways related to the issues discussed in this paper. Nevertheless, Blackburn’s account remains highly influential on current debates surrounding expressivism. In this paper I focus entirely on the problems Blackburn’s account faces with respect to moral dilemmas, and put other objections to one side.


While Marino’s paper also attacks Gibbard’s response to the problem, I will focus exclusively on her arguments against Blackburn’s approach.


More will be said about LA and the notion of validity as satisfiability as we proceed through Marino’s arguments.


Another notable critic of this point is van Roojen (1996).


That is, (pq) entails (¬p v q).


In section 3 I will explore the Hintikka-inspired semantics surrounding this account of validity in more detail, but for the moment I will leave matters at this informal level.


See Gowans (1987) for one overview of matters concerning moral dilemmas.


In his (2008), Mark Schroeder considers the question of whether we should think of expressivism as genuinely endorsing the possibility of moral dilemmas. I will assume for argument here that Marino’s cases should be thought of as genuine dilemmas.


Marino uses the more familiar case of lying to one’s little brother—but the differing case does not affect the argument. Both examples are, of course, fairly contrived and uninteresting compared to actual cases of moral dilemmas. Nevertheless, their simplicity aids us in understanding the arguments at hand.


In more recent work, Blackburn presents further distinct support for the ‘inconsistency’ principle, as follows. Recall that assenting to H!(p)→H!(q) commits me to the disjunction of ¬H!(p) and H!(q). If I am also subject to a moral dilemma involving p and ¬q—that is, I endorse both H!(p) and H!(¬q)—then I am also, by implication, claiming not to be committed to the tree of disjunction involving ¬H!(p) and H!(q). In such a case, as Blackburn says, “we can make no intelligible interpretation of [my state of mind].” (Blackburn 1998, p.72).


That is: in a contingent dilemma, I want p and also want q, where p and q exclude one another contingently; in a logical dilemma, on the other hand, I want p and also want ¬p.


Ibid. Blackburn admits that rule 4 requires revision (p.514) but for the purposes of this discussion we will consider it in its original form.


As an unendorsed fact about the world, (¬p v q) drops away at the next level of approximation, which means there is no way to reach a contradiction in the idealisation process. This follows from rules 1–4 as outlined above.


Again, Marino gives her example in terms of the ‘little brother’ case, but the point stands.


This notion of practical necessity is an alethic modality, and is not to be confused with the deontic modality called “practical necessity” in Williams (1981).


Introducing a practical necessity operator here does not undermine the status of the dilemma as contingent. It is still contingent in that it is neither physically nor logically necessary that ¬(p & q).


If this isn’t true, then there is no genuine dilemma—in that case, I should simply aim to bring about a world in which I give Timothy money and he spends it on something other than drugs.


Although recall that Blackburn’s rules (cf. section 1) do require that any unendorsed statements which make it through to the first approximation should then be carried through to subsequent levels of approximation. (Blackburn 1988, p.513–514).


A corollary of this seems to be that similar rules must hold for stronger senses of necessity, e.g. physical necessity. This result seems independently plausible.


But what about cases where I desire that some practically necessary fact p not be practically necessary—that is, cases where I have the set of attitudes {□p(p); H!(¬□p(p))}? The consistency of this state of mind according to LA depends on the status of two logical principles concerning the ‘□p’ operator. The first approximation to the ideal will, on idealisation rules 1–5, be {p; ¬□p(p); H!(¬□p(p))}. This will come out as consistent only if we reject two principles: firstly, the iteration of □p, i.e. ‘(□p(p))⇒(□pp(p))’; secondly, a further idealisation rule that where a set of sentences L contains ‘□p(p)’, each further approximation to the ideal must also contain ‘□p(p)’. If you accept one of these principles about the ‘□p’ operator then you must also hold that it is inconsistent to express a positive attitude towards the negation of a practically necessary fact. I thank Jon Robson for pointing out these implications to me and for helpful discussion of logical issues surrounding the practical necessity operator.


Note that we could, on the grounds I have provided here, devise an alternative ‘closure principle’. On the account I have given here, ‘□p(pq)⇒(H!(p)→H!(q))’ comes out as valid; so perhaps we should interpret the principle of closure in this way instead.


I thank an anonymous referee for drawing my attention to this similarity.


I make no comment on the success (or otherwise) of this expressivist account of mixed conditionals here. As Hale points out, at best there are serious problems with the approach.


If I am mistaken here, and Blackburn’s work commits him to the weaker ‘logical possibility’ claim, then my argument should be read as providing a revision to LA rather than defending Blackburn’s original form of LA. It seems clear to me that logical possibility is far too liberal a notion to guide our imagining. To repeat examples in the above style: it is logically possible for me to be in both Yorkshire and Norfolk at the same time, but it’s clear that allowing such possibilities into my imagining of the ideal cannot be action-guiding, and thus fails to hold to the spirit of LA.



Thanks to Jon Robson, Daniel Elstein, Aaron Meskin, Peter Simons and an anonymous referee for helpful comments on my arguments in this paper.

Copyright information

© Springer Science+Business Media B.V. 2010