We begin with the opening definition of Derek Parfit's famous essay, "Is Common-Sense Morality Self-Defeating?":

There are certain things we ought to try to achieve. Call these our moral aims. Our moral theory will be self-defeating if we believed we ought to do what will cause our aims to be worse achieved.[2]

Parfit argues that the moral theories underlying "common-sense morality" are self-defeating in this way. The point is not that attempting to follow the theory might lead an agent to produce bad results--the road to hell may be paved with good intentions regardless of which moral theory one adopts. The problem is that under certain circumstances, one or more agents' following the theory successfully would produce what according to the theory are worse consequences than their not doing so. Parfit claims that such an objection can be made to almost all moral theories that give different aims to different agents--as opposed to theories such as act-consequentialism (AC), which give all actors who subscribe to them the same substantive aims. On this argument, AC is to be preferred to any agent-relative alternative.

This essay, however, seeks to establish what might be considered the exact opposite position: that the objective consequentialist theories--theories providing criteria of right that are based on the production of certain consequences, irrespective of the subjective beliefs and intentions of the actor--have the most to lose from the claim of self-defeat.[3] Self-defeat is in fact a near-universal attribute of moral theories, but it is only objective consequentialist theories for which self-defeat is truly an objection. In contrast, there exists a class of alternative theories, notably those in which right action is defined in part independently of the consequences, which may be immune to Parfit's objection. As a result, if we take the possibility of self-defeat as seriously as Parfit does, we should be led to reject AC and all its kin in order to adopt one of the alternatives.

The argument will proceed in four parts. First, this essay will provide a definition and analysis of AC, so as to guide further discussion. Second, it will examine the concept of moral self-defeat, and demonstrate why, contra Parfit, such a concept is more dangerous to objective consequentialist theories than it is to common-sense morality. Third, following lines inspired by Donald Regan, it will provide a proof that AC can indeed be self-defeating under certain (relatively common) circumstances, and that it cannot be fully insulated from self-defeat through the addition of general rules.[4] (Indeed, all consequentialist theories marked by what Regan calls "exclusive act-orientation" are vulnerable to the same objection.) Finally, it will expand on Regan's arguments to show that even a consequentialist theory which is not exclusively act-oriented and which evades the standard objection of self-defeat still cannot guarantee optimal consequences through its universal satisfaction. The point is not to show that objective consequentialist theories are somehow uniquely liable to produce worse consequences, but rather that the optimal consequences are not achievable in all circumstances by any theory, even given its perfect satisfaction. This result calls into question why we should regard consequences as the sole measure of right action, and thus undermines the intuitions that might have led us to support objective consequentialism in the first place.

Let the doctrine of AC be defined as follows:

An action by an agent A is right if and only if it produces at least as good consequences as any other act available to A.

Two further features of AC should be noted. First, AC is defined purely as a criterion of right action. It does not offer individual agents a recipe for making moral decisions. As fallible human beings, we are rarely aware of which available act will eventually have the best consequences; in fact, as James Lenman compellingly argues, we will never have even the faintest reason to believe that a given act will produce the best consequences overall.[5] What matters to AC, however, is the objective assessment of the consequences produced by a given act--not our individual and subjective assessment of what an act is likely to accomplish.

As an illustration, compare AC to the decision-theoretic consequentialism (DTC) described by Frank Jackson.[6] On this theory, consequentialism would provide an objective account of what results we ought to desire. DTC would then describe an act as right if and only if it could have been undertaken in light of these desired results and the individual agent's subjective assessment of the relevant probabilities. As an illustration, Jackson proposes the example of a doctor who must treat a patient (call him "Joe") for a non-life-threatening skin complaint. Drug A would relieve Joe's condition but not cure it entirely; drugs B and C would each cure the condition completely for one-half of the population but instantly kill the other half--and no one can predict which will happen in any given case. Those who share Jackson's intuitions would describe the right action as prescribing drug A; the marginal improvement that could be achieved in treating Joe's skin condition simply isn't worth the risk of killing him.[7] But in fact, this act could never be right according to AC, since it is guaranteed to produce suboptimal consequences. The best consequences are, in this case, achieved by prescribing drug B, which just happens to be the drug that will cure Joe entirely. The fact that the doctor did not know this is irrelevant to an objective assessment of the consequences; prescribing drug B was an available act, and had the doctor prescribed drug B on a hunch, this action would have right according to AC.[8] In short, to an act-consequentialist, it can never be wrong to guess right.

Second, AC will be understood in this essay to be concerned with only the marginal consequences of an individual's action--those consequences for which that action is both a necessary and sufficient condition. This interpretation is contested by Peter Singer, and it is worth setting out his position at length:

Although some act-utilitarian writers may have assumed that only consequences for which the act is a necessary or sufficient condition should be taken into account, there is no good reason for an act-utilitarian to do so. An act may contribute to a result without being either a necessary or sufficient condition of it, and if it does contribute, the act-utilitarian should take this contribution into account. The contribution that my vote makes toward the result I judge to be best in an election is a relevant consideration in deciding whether to vote, although it is, almost certainly, neither a necessary nor a sufficient condition of that result; for if this were not so, the act-utilitarian view would leave us with [an election result] which was unconnected with the actions of any of the voters, since what is true of my vote is equally true of every individual vote.[9]

Now, as Regan notes, there is a difference between viewing an election as causally unconnected with the actions of the voters and viewing it as morally independent of the actions of at least some of them. More importantly, however, it is not clear how one would cash out the "contribution" made by an act that is neither necessary nor sufficient to achieve a given result, especially in comparison to the marginal impact of another act. For instance, if there are n people engaged in a joint project which will produce results r, one natural way of measuring the contribution of each person is simply to divide r by n.

An example provided by Regan easily shows the flaws in this approach.[10] Assume there is a population of 50 people, many of whom are engaged in a joint project that will raise $60,000 for charity. However, the joint project actually requires the efforts of only 49 of the 50, such that the assistance of the 50th person (who happens to be our friend Joe) is neither necessary nor sufficient to achieve the desired result. If it were possible for Joe to work independently of the others, he could raise $1,000 on his own. Yet if he were to participate in the joint project, his individual contribution would be assessed at $1,200--even though he produces no additional value at all. An interpretation of AC that required Joe to maximize his "contributory" consequences would produce worse consequences overall, and could hardly be endorsed by a consequentialist of any stripe.[11] Examples such as these cannot prove that, according to its own lights, such a theory is inadequate. But as Regan notes, it is difficult to see why someone truly concerned with the consequences would endorse such an interpretation; in any case where we are given a choice between a marginal and a contributory improvement to the world, the marginal approach would win out.

With AC thus defined, we can turn to Parfit's criticism of "common-sense morality," and understand his argument that AC is immune to similar attacks. Suppose that we have a moral theory T which gives different substantive aims to different people. (A substantive aim is one of those "things we ought to try to achieve," as opposed to the formal aim of following the theory correctly.) Furthermore, suppose that there are two people, each of whom can do one of two actions, with the following results:[12]

Parfit presents a number of cases along these lines, each inspired by the familiar "Prisoner's Dilemma." In the case above, according to T, Alice should always do (1) rather than (2), since in all cases that will better achieve her T-given aims. Bob would apply the same reasoning, and both of us would do (1). However, this would mean that our T-given aims will be worse achieved than if we both did (2). Our seeking to maximize our T-given aims therefore cause those aims to be worse-achieved, which seems to be something of a paradox. The worry is not merely that, in a case where all agents attempt to satisfy T, our substantive T-given aims will be worse achieved; this could happen on any moral theory, and results from simple human fallibility. Let such a case be described as one of "indirect collective self-defeat." The problem is that the theory is "directly collectively self-defeating" (or just "self-defeating" simpliciter), as it is true that, in a case where all agents successfully satisfy T, our substantive T-given aims will still be worse achieved.[13] Indeed, such a case could occur whenever the following conditions hold:

(a) our moral theory gives to each a different aim,

(b) the achievement of each person's aim partly depends on what others do, and

(c) what each does will not affect what these others do.

Any theory that meets these three criteria will run the risk of collective self-defeat, even when it is satisfied by all agents. And as Parfit notes, "If there is any assumption on which a theory should not be self-defeating, it is the assumption that it is universally successfully obeyed."[14]

If Parfit's argument stood, it would be a powerful objection to those moral theories that do not give common aims to all agents. Fortunately, however, it does not stand. Parfit's claims beg the question against alternative theories through an unexamined assumption of consequentialist intuitions. He attempts to counter this charge by showing that even an agent-relative theory (indeed, especially an agent-relative theory) is liable to substantive self-defeat; yet the question-begging occurs at a deeper level. Parfit assumes that the achievement of substantive aims is the proper goal of moral action--that "we successfully obey our moral theory when each succeeds in doing what, of the acts available, best achieves his [substantive] moral aims."[15] It is this assumption that is flawed. One could easily imagine a pair of Kantians who, by telling the truth to a pair of ax-murderers at their doors, knowingly cause each other's deaths--even when protecting the other's life is indeed a substantive aim given by their moral theory.

Perhaps this dispute arises from a mere confusion as to the nature of substantive aims. Parfit claims it is possible to include under the rubric of "substantive aims" even goals concerning "what we do," rather than what we achieve by our actions (e.g., "that I do not steal," as opposed to "that my children do not starve").[16] Many agent-relative theories might thus be self-defeating; it might be perfectly possible that another's moral actions might cause me to harm the innocent more than otherwise.[17] But while another's moral actions might cause me to kill (e.g., by placing an innocent person in the path of my car), it is unclear how they could cause me to murder--or, indeed, to commit any act defined in part by my intent or state of mind, so long as we are assuming my complete and faithful obedience to T. Parfit does not present a single example where someone else's T-inspired actions might directly cause aims of this nature to be worse achieved. Under what circumstances would another's acting out of respect for humanity, say, lead me for reasons of morality to steal, or knowingly to make a false promise? I do not see how another's acting on DTC, for instance, could possibly cause me to undertake an action that would seem to me at the time to fail to maximize expected benefit, while I still obey DTC.[18] Defining "that I do not murder" or "that I act as would seem to me best" as substantive aims would falsify Parfit's criterion (b), that "the achievement of each person's aim partly depends on what others do." A subjective theory that relied on the agent's state of mind would therefore escape the charge of self-defeat.

The problem seems to be that a moral theory T may have certain "aims" the optimal achievement of which is guaranteed by obedience to T. Acting "as would seem to me best," if it can be described as an "aim" of DTC, is one example; if I am already obeying the theory, I do not see how another's action could cause me to achieve this aim any better. If we treat such self-guaranteeing aims as substantive, then Parfit's critique no longer applies to theories whose essential aims are self-guaranteeing.

If, however, we choose to treat such self-guaranteeing aims as "formal" rather than "substantive," then Parfit's argument proves far too much. In that case, we need not resort to elaborate collective examples to show that a subjective theory can be self-defeating. Consider the example of decision-theoretic consequentialism given above; the doctor following DTC will prescribe the less-than-optimal drug A, even though optimizing the consequences in the same way as AC is obviously among the theory's substantive aims (on this new understanding of "substantive"). The complete cure of the disease is clearly, in Parfit's terms, something we "ought to try to achieve." DTC is therefore directly individually self-defeating--a condition that Parfit does not even recognize as possible.[19] Following DTC will therefore result in the DTC-given aims being worse achieved, and DTC can therefore be self-defeating regardless of whether group action is involved.

Indeed, this result can be made far more general. Let us call a theory T₁ "contained within" a theory T₂ if and only if the set of all acts of which T₁ approves is a subset of those approved of by T₂. (Note that all theories are contained within themselves.) It then follows that any theory T that is not contained within AC may in certain cases be directly individually self-defeating. Consider the version of AC in which the best consequences are those most effective in achieving T-given substantive aims (call it AC_T); since T must in some case approve of an act of which AC_T would disapprove (by virtue of not being contained within it), an agent A's satisfying T in this case will cause his or her T-given aims to be worse achieved compared to undertaking the alternative act proposed by AC_T.[20] Thus, Parfit's notion of substantive self-defeat would apply to almost any moral theory distinct from AC--even those which give common aims to all agents. DTC, for instance, clearly gives all agents the same substantive aims, but it is easy to imagine cases where it is either individually or collectively self-defeating.

Rather than enshrine AC as the optimal moral theory, however, Parfit's argument shows us only that self-defeat is an inappropriate test. The worse achievement of substantive aims is a mundane feature of widely accepted moral theories. What Parfit must show is that this mundane feature in fact constitutes an indictment of the theory; i.e., that what matters most is the achievement of substantive aims--or, in other words, the consequences.[21] Parfit argues that the achievement of aims "may not matter morally," but does "matter in a way that has moral implications"; part of the reason why we ought to try to achieve our substantive moral aims "must be that, in this other sense, their achievement matters." In other words, he relies on the consequentialist intuition that if it is wrong to do act X, then it is bad for acts like X to happen, or for the consequences of acts like X to be realized. And if we held that intuition, we would already be consequentialists. Those who place greater importance on the nature of the act itself, or on the intentions with which it was undertaken, will not concede that the worse achievement of its substantive aims is an objection. Parfit's argument would therefore fail to convince anyone who did not already agree with him when he started.

In fact, the only theories which are truly vulnerable to arguments from self-defeat are those consequentialist theories which place ultimate value on the achievement of substantive aims. If the universal satisfaction of these theories can produce worse consequences than their universal violation, then it is very difficult to see what would recommend their adoption.

Can a theory such as AC be self-defeating? Consider the following example. During the 1993 flood of the Mississippi, cities would occasionally call for volunteers to help lay sandbags to keep the river from overflowing its banks. Suppose that one year there is a particularly terrible flood, one that threatens the entire city with destruction. Moreover, assume that what matters is ultimately not the sheer number of sandbags laid, but whether the number of sandbags exceeds some threshold that nobody knows in advance. In other words, there is some number of sandbags sufficient to block the water from spilling over; any fewer than that number results in the whole structure being swept away, and any more than that number results in no additional benefit. However, no one knows how high the threshold is, and so everyone fears that the city will be destroyed.

Now suppose that our friend Joe, whose luck appears to have entirely run out, is a representative resident of this town, and that one night the city makes an emergency call for volunteers. Suppose further that, in addition to laying sandbags, he has an alternative action available, namely working from home to generate income to donate to charity.[22] We can rank the n residents of the city in ascending order from 1 to n, according to their ability to generate income from home. Moreover, let us assume that everyone is equally effective at laying sandbags, and that the threshold of sandbags will be met so long as i people participate (with 1 < i < n). If the destruction of the city would be sufficiently bad to outweigh the potential income generated by all n residents, the optimal outcome will clearly be produced if persons 1 through i lay sandbags, and if the rest stay home and work.

However, this information will not be available to Joe, who knows neither the magnitude of i nor where he falls in the scale from 1 to n. (Assume also that n is large enough that he cannot communicate effectively with his fellow residents.) He may reason as follows: even if many tens of thousands have volunteered to lay sandbags, regardless of what i might be, the city will likely miss it or exceed it by several thousand at least. As a result, there is an astronomically low subjective probability that Joe's particular contribution of sandbags will be necessary and sufficient for the city to be saved. Since only the marginal contributions matter, it is almost certain that he could better satisfy AC by staying at home and working. The same logic holds for all other residents; the effort they might expend laying sandbags has a very minimal probability of making a difference, and virtually any benefit that could be achieved by working from home would be preferable. Thus, no one would choose to volunteer to lay sandbags, and in a city of act-consequentialists, everyone would drown.

Thus far, one might claim, I have only identified a case of indirect self-defeat; the residents try to do what AC demands, but fail due to their limited knowledge. Yet what such an analysis ignores is that once everyone is at home generating income, they are all satisfying AC. The efforts of a single person to lay sandbags would be fruitless; it would be better for Joe to stay at home. Thus, it is true for all agents that they are performing the available act that produces the best consequences; were any of them to change their acts and lay sandbags, the consequences would be worse. But even though AC is universally fulfilled, the AC-given aims are worse achieved. Indeed, the consequences might even be better if no one satisfied AC. Suppose that all n residents volunteer to lay sandbags. Since i < n, it is true of Joe (and thus of each agent) that he ought, according to AC, to stop laying sandbags and go home; the city would still be preserved, and additional income could be generated on the side. Thus, each agent is violating AC. Yet because the city is saved, this outcome remains better than the case where everyone stays at home, satisfies AC, and drowns.

Alternatively, one might claim that the framework of individual actions is an inappropriate lens--that since the best result would be produced if persons 1 through i laid sandbags, this is the only arrangement that satisfies AC. But this does not speak at all to the choice faced by Joe, who cannot coordinate with all n residents to achieve this pattern. His only available acts are to stay at home or to lay sandbags; he cannot directly affect the actions of others. And in the case where everyone else is staying at home, following their example will be the available act that produces the best consequences overall. The universal satisfaction of AC causes AC-given aims to be worse achieved than by its universal violation, and this is precisely our definition of direct collective self-defeat.

Faced with such an argument, Parfit would likely respond that although it meets the above definition of direct self-defeat, it does not meet his. On Parfit's definition, unlike the one presented in this essay, a theory T is directly self-defeating only if it is certain that following T will cause our T-given aims to be worse achieved. And as Regan shows, AC is always consistent with producing the best consequences. Suppose Joe is among a group of agents whose acts jointly produce the best possible outcome (e.g., only persons 1 through i are laying sandbags); were he to change his actions, the outcome would be worse. Thus he must be satisfying AC by acting in such a way as to produce the best consequences. So there is never a time when an individual could, on his or her own, produce a better outcome by disobeying AC--indeed, that is a trivial consequence of its definition.

The certainty of self-defeat is surely a far worse consequence for a theory than the mere possibility of self-defeat in certain situations. But is the latter really something that the act-consequentialist can afford to ignore? Remember, Parfit originally defined self-defeat as believing that "we ought to do what will cause our aims to be worse achieved." And the flood case does not point merely to the possibility of indirect self-defeat, where adherents of AC will try and fail to achieve their goals. Rather, the city's residents all succeed in satisfying AC, and still have not managed to produce the best consequences. An alternative theory, even a narrower gimmicky one ("An act is right if and only if it will produce the best consequences, and moreover persons 1 through i should lay sandbags in the Great Flood of 2033") would, if universally satisfied, be guaranteed to produce as good or better consequences than AC alone.[23] And such cases, I would claim, are hardly unusual. Although major floods are thankfully rare, many of the moral choices we make--whether to vote in a given election, whether to walk across the grass in a college quad, whether to participate in a joint project for the improvement of society--depend for their consequences on the actions of others.

Moreover, it must be remembered that what we are addressing here is a criterion of right, not a decision procedure. Were a given decision procedure P self-defeating only under extraordinarily unlikely circumstances, such rare self-defeat might not constitute a reason to abandon it; following P would still produce the moral outcome enough of the time to make P preferable to any other usable procedure. But if a criterion CR, under the same unusual circumstances, described the same act as both wrong and right, we would be immediately moved to reject it as contradictory. The likelihood of those circumstances being realized is irrelevant; CR must be judged on theoretical grounds, and a theory that is formally flawed in this way is deeply unsatisfactory. 

I do not claim that self-defeat shows that AC is self-contradictory, but only that it suffers from a formal flaw, the impact of which is not dependent on the likelihood of the circumstances that make it noticeable. A criterion of right can only be defended on theoretical grounds, and not on the grounds that its widespread satisfaction might produce good consequences. For one thing, on this standard, AC would certainly be rejected in favor of some form of rule utilitarianism, which would guarantee best consequences by its universal adoption. For another, applying such a consequentialist rule to the choice of criteria ignores the fact that the actual consequences in practice will be produced by fallible individuals who attempt to satisfy, rather than actually satisfy, the chosen criterion. Since a good consequentialist would look to actual over theoretical consequences, what would matter for such a choice would be the question of indirect self-defeat, and whether such a rule would be well-fitted to the crooked timber of mankind. If we care about the actual consequences, we cannot choose our moral theory in a fairyland where omniscient agents act solely from benevolent motives.[24] And given Lenman's argument that we are indeed clueless as to the full consequences of our actions, there is no reason why we should endorse (on consequentialist grounds) a criterion such as AC over one such as DTC, which takes account of our limited knowledge and abilities.[25] The possibility of self-defeat negates any advantage that a criterion such as AC might possess.

How ought the consequentialist to respond to these claims? The difficulty for AC arises because it is indeterminate; it will be universally satisfied if everyone stays home or if persons 1 through i lay sandbags. The obvious response, then, might be to create a number of rules that identify the best arrangement in every case for which AC is indeterminate, and then require individuals to pursue that outcome to the exclusion of the others. Yet this response is utterly inadequate to save consequentialism from self-defeat. For instance, if the rule is that "Persons 1 through i should lay sandbags, and everyone else should stay home," it will guarantee the best consequences if universally satisfied. But if it is not universally satisfied, it will be individually self-defeating; if person i neglects to lay sandbags, someone else must be able to violate the rule and save the city. Taking self-defeat seriously deprives the consequentialist of any grounds on which to choose among AC and such rule-based models; AC can guarantee the best consequences on the individual level, but is consistent with worse consequences for the group, while rule-consequentialism can guarantee the best consequences for the group, but is consistent with worse consequences for the individual.

Furthermore, revising the rule to make it conditional (e.g., "Persons 1 through i should lay sandbags, and everyone else should stay home, except that it is acceptable to diverge from this pattern when others' doing so means that following it would produce worse consequences") would only create the same problems that attend AC. In a case where everyone stays at home, they are again satisfying the rule, because others' diverging from the pattern means that laying sandbags would be fruitless. But even though the rule is again universally satisfied, the city will be flooded and the worse consequences produced.

Indeed, Regan presents a simple but devastating proof that no "exclusively act-oriented" theory can guarantee the best consequences regardless of others' decisions to satisfy or violate it. Exclusively act-oriented theories are marked with the property that for any person, given the behavior of others, the theory endorses a non-empty set of natural actions (such as "lay sandbags" or "stay home").[26] These theories are concerned solely with the eventual actions, and not with the method, decision procedure, or intent with which they are undertaken. Suppose that a theory T is exclusively act-oriented. Imagine how T would address Joe in the case where all other actors are staying at home. If T is not to be individually self-defeating, it must endorse only that course of action which would produce the best consequences--namely, that Joe should stay at home, rather than uselessly lay sandbags. But in that case, every other agent staying home is also satisfying T, meaning that T is universally satisfied. But given that everyone is staying home, the city will be destroyed, and T cannot be said to have guaranteed the best consequences through its universal satisfaction. The only way a theory T could avoid self-defeat is if it is possible for everyone to stay at home without having satisfied T; in other words, if some of the factors ignored by exclusively act-oriented theories--method, decision procedure, intent, etc.--are taken into account.

In the place of exclusively act-oriented theories, Regan suggests instead his favored consequentialist theory, that of cooperative utilitarianism (CU). In short, CU requires agents to employ a certain decision procedure, identifying other agents willing to cooperate in a given project and then acting to produce the best consequences in light of the available cooperation. CU, Regan shows, has the desirable property of "adaptability"; if satisfied, it guarantees that the class of agents who satisfy it are producing the best possible consequences, both as a group and as individuals.

CU is, however, still an objective theory; it describes actions as right only if this decision procedure succeeds, and if the agents in fact produce the best consequences. And Regan's proposal, no matter how carefully crafted, can still suffer from flaws similar to those that plague the cruder versions of consequentialism. The presentation of CU assumes that good or bad consequences are produced only by the eventual choice of action (e.g., laying sandbags or staying at home). Occasionally, however, the decision procedure employed can have consequences as well--by setting inappropriate examples for others, etc.--and these consequences must be taken into account.

To employ an extreme illustration of the point, suppose that there exists a "mad telepath," who will read Joe's mind as he makes a minor moral decision and, if Joe at any point employs the decision procedure required by CU, will blow up the Macy's department store. (I hasten to add that the example is Regan's, not mine.) Assume that the destruction of Macy's would outweigh any possible benefit from Joe's actions. How should Joe proceed? Perhaps he will decide that it is better to pick the wrong alternative in his minor decision in order to save the customers of Macy's--but how would he decide this? If the correct theory for testing the rightness of actions is CU, then the only way for Joe to learn that he should not apply CU on the "first-level" decision is for him to apply CU on the "second-level" decision of whether to apply it on the first, and Macy's will still be destroyed. Under such circumstances, it would be best for Joe never to apply CU, and instead to act on a hunch; the consequences of satisfying CU at all would be worse than the consequences of violating it, and CU is directly self-defeating.

If Regan were right in his claim that a similar "mad telepath" argument could be constructed for exclusively act-oriented theories, then perhaps we would still decide that CU is superior to any other consequentialist theory. And one could certainly imagine a mad telepath who would create intractable practical problems for Joe, perhaps by threatening to blow up Macy's if Joe acts with any intent of satisfying AC. But surely AC, which is an objective theory, would then require that Joe perform the appropriate action without intending to do so. There cannot be a mad telepath who conditions the destruction of Macy's on the actual satisfaction of AC, because doing so would result in a contradiction.[27] The "mad telepath" argument therefore can only be applied to non-exclusive-act-oriented theories; it is a flaw that they uniquely possess. Yet exclusive act-oriented theories, as we have seen above, suffer from the possibility of self-defeat. And as Regan concludes, "The ideal consequentialist theory--a theory which guarantees in all cases that the best possible consequences all things considered will be produced by whatever collection of agents satisfy the theory--is a logical impossibility."[28]

What lessons can be drawn from this result? I would not argue that the impossibility of perfection forces us to abandon objective consequentialism. Indeed, the conditions under which CU is self-defeating seem rather unusual indeed. Rather, I am arguing that the impossibility of perfection makes attention to the consequences somewhat less pressing. If Parfit had been right, and an objective consequentialist theory were free of the danger of self-defeat, it might have been preferable to its consequentialist rivals, and perhaps even to non-consequentialist alternatives. Certainly CU would have much to recommend it if it had not been liable to its own set of theoretical objections. Yet if there can be no objective consequentialist theory that is free of such flaws, the prospect of choosing a criterion of right in the hopes of avoiding self-defeat seems dim. Moreover, if any objective theory that seeks to optimize consequences will under certain circumstances compromise its own goals, perhaps that is a reason for us to care somewhat less about consequences. A theory such as DTC makes no attempt to hide the fact that it cares about the actor as much as the action--that it values the standpoint of the subjective agent as well as that of the objective observer. If no great advances can be made by limiting ourselves to the objective consequences, why should we not give such theories a second look? If, as Regan fears, we cannot find an approach to maximization that clearly states what is required of both individuals and groups, then "quite possibly we must give up the idea that maximizing good consequences is what morality is all about."[29]

It must be remembered that the advantage, if any, of an objective criterion of right such as AC is its theoretical simplicity and elegance. In its actual implementation in a flawed and fallible world, it may turn out to be as difficult (and as indirectly self-defeating) as any of its rivals. Yet if such criteria lead to inappropriate assessments under certain situations, no matter how unlikely, their theoretical achievements are seriously undermined. In abandoning practical for theoretical benefits, these theories have at the same time lost any resources with which to defend themselves against theoretical objections.

We therefore return to the opening claim of Parfit's essay: "There are certain things we ought to try to achieve."[30] Morality requires us to attempt certain projects, and to apply our subjective beliefs in certain ways. The error made by objective consequentialist theories is that they instead identify certain things we ought to achieve, forgetting that their simultaneous achievement may, even in theory, be impossible. "For us," as T.S. Eliot wrote, "there is only the trying. The rest is not our business."[31]

[1] I am indebted to Ralph Wedgwood for his insightful comments and suggestions.

[2] Derek Parfit, "Is Common-Sense Morality Self-Defeating," Journal of Philosophy 76:10 (Oct. 1979), pp. 533-45, 533.

[3] I will occasionally refer to such theories simply as "consequentialist theories," while recognizing that this term would normally include subjective theories such as the decision-theoretic consequentialism described below. Subjective consequentialist theories will be indicated explicitly.

[4] Donald H. Regan, "Utilitarianism and Co-operation" (Oxford: Clarendon, 1980).

[5] James Lenman, "Consequentialism and Cluelessness," Philosophy and Public Affairs 29 (2000), pp. 342-370. Lenman points out that even an apparently heroic act, such as saving a drowning child, may produce worse consequences 50 generations hence when the child's descendant commits terrible war crimes. Indeed, if future consequences are not discounted over time (and there is no reason why they should be, since future individuals have just as much claim to be objects of moral concern), then almost any act would, over the course of the next several millennia, produce immense consequences for either good or ill. Since we can have no knowledge (or even justified beliefs) concerning the extent or direction of these thousand-year effects, Lenman claims, consequentialism is essentially useless as a guide to action.

[6] Frank Jackson, "Decision-theoretic Consequentialism and the Nearest and Dearest Objection," Ethics 101:3 (April 1991), pp. 461-482.

[7] Does the rightness of the doctor's decision to prescribe drug A depend on the decision procedure? DTC could be interpreted in two ways, either saying that an act is right if it is the result of the appropriate weighing of probabilities (i.e., the proper decision procedure was followed by the agent), or if it could have been the result, regardless of which decision procedure was actually employed. I would favor the second interpretation, as it allows us to state that a doctor who prescribed drug A for other reasons (by accident, flipping a coin, etc.) eventually wrote the right prescription for someone in his circumstances, but employed a method of prescription that was wrong. (Flipping a coin is, according to DTC, an unacceptable method of practicing medicine.) Note that on this interpretation, DTC is a subjective moral theory--depending as it does on the agent's own knowledge and beliefs--but it is also exclusively act-oriented in the sense described below.

[8] Cf. Regan's discussion of "ought implies can" on pp. 173-4.

[9] Peter Singer, "Is Act-Utilitarianism Self-Defeating?," Philosophical Review 81:1 (Jan. 1972), 94-104.

[10] The example presented here is modified from the original.

[11] An even more concerning result can be found in the second example, where a joint project requires the work of all 50 individuals to raise $20,000 for charity. Of the 50 individuals, at least 49 will participate. However, the project is so inefficient that individuals could raise more money ($1,000 each) by working independently. Assuming that Joe is unable to persuade the others to abandon the group project, he clearly ought to join them and realize the $20,000 gain, rather than allow the project to fail. Yet his individual contribution to the group project would be assessed at only $400, making it appear that he would be better off raising only $1,000 on his own.

[12] The example is taken from Parfit, p. 535.

[13] Note that the latter definition differs from Parfit's, as will be discussed below. Cf. Parfit, p. 534.

[14] Parfit, p. 542.

[15] Parfit, p. 540.

[16] Parfit, p. 543.

[17] Parfit. p. 544.

[18] Indeed, it seems part of the definition of DTC that it could not. Perhaps, for theories relying on the agent's intent or state of mind, one could construct a case involving time inconsistency: perhaps avoiding an immoral action now would, due to a similar choice by another, result in changes in my character such that I will commit an (even worse) immoral action later on. But then it seems we have abandoned the critical assumption of universal satisfaction. Such a theory might be indirectly self-defeating, given my weak character and fallibility, but it would not be directly self-defeating, because we are no longer assuming (as Parfit does in all other cases) that all agents have at all times successfully obeyed their moral theory. By doing (1) rather than (2), Alice did the right thing, even though her T-given aims could have been better served if both had done (2) than if both had done (1). This assumption of irreproachable conduct would no longer hold in a case of changing character.

[19] Parfit, p. 535.

[20] Does this result also apply to any theory S that is extensionally distinct from AC_T? Not necessarily--AC approves of those acts that produce the best or equal best consequences, and so one could construct an artificial S that chose only one action among the equal best alternatives to endorse. In that case, in any given moral decision, there will still be at least one option of which S approves, but it is not true that S is individually self-defeating.

[21] These consequences are, however, not necessarily valued in an agent-neutral way.

[22] The income will be bequeathed to the same causes, should everyone die in the flood.

[23] This result appears to validate the claim of D. H. Hodgson, that "to act upon the act-utilitarian principle would probably have worse consequences than would to act upon more specific moral rules, quite independently of the misapplication of that principle." (D.H. Hodgson, Consequences of Utilitarianism, p. 3, quoted in Singer, p. 94.)

[24] Indeed, many defenses of the act-consequentialist society have relied on such assumptions of omniscience. Take for example Singer's defense against Hodgson's communication argument. Hodgson claims that in a world full of fully rational act-consequentialists, communication would be impossible; a representative agent Alice could never be sure that Bob was telling her truth rather than offering a noble lie, and could not act on the information he offers. Singer offers an amusing response to this objection: if telling a lie to Alice would cause her to act in such a way so as to produce better consequences, then surely Bob could merely say so, and Alice (pursuing the best consequences) would of course respond appropriately. Yet this objection is only appropriate to a world in which Alice and Bob both consider each other to have perfect judgment in moral matters. If Alice believes Bob's judgment to be potentially flawed, she might well worry that he is telling her a lie out of a false judgment as to its potential benefits; and if Bob believes Alice's judgment to be flawed, he might well lie to her out of a fear that the truth will produce worse consequences. Thus, Hodgson's argument might well hold, and society could disintegrate. So long as humans are still fallible--a fact unlikely to change this side of Heaven--AC can still be indirectly self-defeating on a massive scale. And though Singer is right that such indirect self-defeat "would still not quite refute act-utilitarianism," he is also right that accepting the truth of this claim "would be to seriously embarrass the act-utilitarian, for he could hardly continue to advocate his doctrine" (Singer, p. 94).

[25] Endorsing a criterion such as AC on the grounds that it produces the best consequences also seems somewhat circular, but I am not certain if this constitutes a real objection to the approach.

[26] Regan does not offer a complete definition of exclusive act-orientation, but rather identifies a necessary criterion that all exclusively act-oriented theories must meet.

[27] Suppose that Joe makes the minor moral decision in such a way as to satisfy AC. Then Macy's is destroyed. Thus, his action is not the available act that would produce the best consequences. Thus, AC is not satisfied, and Macy's is not destroyed. But then Joe's action did have the best possible consequences--and so on.

[28] Surprisingly, the theory of DTC sketched above would be resilient to the attacks of a mad telepath. Since the interpretation in this essay requires only that an act could have been selected through the proper decision procedure, assuming the same facts about the external world and the agent's knowledge and beliefs, DTC is exclusively act-oriented, and thus the mad telepath cannot use as a condition the actual satisfaction of DTC. Moreover, if the mad telepath looks only at the application of a certain decision procedure, the theoretically right course would be to do the DTC-endorsed act on a hunch, while still satisfying the theory. A mad telepath would of course present insoluble practical problems for an agent guided by DTC, but the same would be true for any moral theory whatsoever; it is only because the theoretical objection applies to some theories but not others that it has argumentative force.

[29] Regan, p. viii.

[30] Parfit, p. 533 (emphasis added).

[31] T.S. Eliot, "East Coker," Four Quartets.