It Is Impossible to Maximize Good Consequences.

In my last post I argued that utility monsters, genocide of the less happy, and barely happy multitudes raise problems for the maximization clause of the fundamental definitions of consequentialism. Looking again to those definitions, as standardly propounded, I am going to argue that it is impossible to maximize good consequences – even in the unlikely event that a coherent account can be given of the choice alternatives and their consequences.

The Lucky Alternative Problem

Take the last day I gave a lecture on ethical theory.  I was in the classroom at the scheduled time.  I was not on my office computer entering the winning number in an online lottery.  As a result I did not receive a fortune to distribute to the worthiest of charities. I had naively thought that I probably did the right thing in meeting my class on time.  If, however, I were a utilitarian of the classical Bentham-Mill sort, wouldn’t my action have been morally wrong? I didn’t act so as to maximize the net total of good consequences.

This example may irritate you; after all, I could not simply sit down and enter the winning number on an internet lottery. There is the little problem of probability.  Ultimately, you are right, of course; a satisfactory consequentialism must take the improbability of lottery wins into account. However, what I am concerned with here are some well accepted definitions resident in the theoretical underpinnings of consequentialism, and there the lottery is a problem.

It is a problem because “best consequences” are standardly defined in terms of the consequences of all physically possible alternatives. It is not physically impossible to win a lottery. You have seen pictures of real live people who did that very thing.  No force would have prevented me from entering just the right digits had I tried to type them on my key pad. So entering that number on my office computer would seem to be one of the astronomically large number of alternative actions I could have done instead of starting my lecture on time.

Focus for now on the action that has the very best consequences in the most demanding sense, sometimes called, “the objectively right action.”  Even so focused, you may suspect that the lottery example is philosophical trick. I want to persuade you, however, that the lottery is, in fact, illustrative of a genuine, more general, and deeper problem. Lucky alternatives are everywhere.

Maybe you put your right shoe on first today. The alternative of putting your left shoe on first would have had some different consequences, but common sense suggests that those two consequence chains would have quickly merged for all practical purposes. Some other alternative pairs we can expect to diverge dramatically at least for a while, e.g. putting on your right shoe then lacing a roller skate onto your left. 

A few alternatives of ordinary actions would clearly have drastically different consequences that continue in very long runs, e.g. Hitler’s mother’s putting arsenic in his baby bottle. If we give scope to our imagination among physically possible alternatives, it seems almost inevitable that some of them will have better consequences than what we actually did. There will be alternatives very similar to what we actually did that will have at least slightly better consequences and bizarrely different actions may well have dramatically better consequences.

It gets worse. As I said, common sense has it that for most of our alternatives small differences in our actions will tend to make little, if any, difference in long run consequences. Chaos theory, however, gives us good grounds for thinking that common sense is wrong on this point. It is a primary theorem of chaos theory that very similar acts will sometimes have vastly different long run consequences. For twenty five generations of English speakers, “for the want of a nail,” would have come to mind; now it is the “butterfly effect.” We humans, if not in some part chaotic systems ourselves, often send our consequences into chaotic environments. Sometimes putting on your left shoe first may have drastically different long run consequences after all.

Couple the vast number of things that human beings can, under normal circumstances, do if they choose with the butterfly theorem, and it becomes almost inevitable that among our alternatives will be some extreme outliers – actions that would have enormously worse or enormously better long run consequences than the action actually performed. For the definition of objective right, it is the enormously better ones that are the problem, a problem that will be greater if the future human population is very large.  A populous distant future will have great leverage on the total good consequences of our alternatives.  Something with very negative consequences for our generation may turn out, because of consequences thousands of years hence, to be objectively right— if there were such a thing. (I plan to return to the question of the leverage of a populous future for consequentialism in particular and morality in general in a near-future post.)

Lucky Sequences of Alternatives

A consistent consequentialist, of course, is not really concerned with achieving the best possible consequences for a single act but in producing best possible consequences over a lifetime.  Admittedly, when we move from the consequence maximizer at a given moment to the consequence maximizer of a series of actions over an hour, a day, or a decade, the particular example of the lottery loses its punch. It is, lottery-wise, not so obviously false that I did the right thing in meeting my class. I could have won the lottery in the hour before class and still gotten to class on time.  

This may solve the lottery paradox, but it only makes the more general problem worse. When we look to the best consequences of a sequence of actions we have not only the possibility of picking up many independent weird outliers at different times, but also the possibility of weird combinations of outliers. You play and win the lottery, then visit a casino and win at roulette. Next you head to the horse track, after which you give the total amount of your winnings to (what turns out in the long run to be) the best of charities or the optimum combination of charities. Over time, objective rightness cherry picks the best consequences of all combinations including all outlier combinations of outliers and inliers. Surely hitting the objectively right series of actions is going to be geometrically less likely than getting right a single action. So over time the probability of acting objectively rightly is effectively nil.

Lucky Alternatives and Consequentialist Definitions

What all this establishes is that, understood in terms of consequence maximization for any very wide set of alternatives, “Jenny did the objectively right thing,” is, if meaningful, almost always going to be false. Is this any real problem for consequentialism, however?  There is the consolation that most of the practical work of consequentialism is not done by assertions of objective rightness or of best possible consequences. The heavy lifters are the action guiding statements and the expressions of praise or blame.

Where the lucky lottery and its extensions and generalizations pose a problem is for the definitions linking objective rightness with consequentialism’s action guiding principles. The objective right is what we are told to shoot for.  The way consequentialists have often put it is that we should guide our conduct by trying to do the objectively right action, within the constraints of efficiency of deliberation. 

If we do not try for the best consequence alternative, then we can be blamed. Under our conclusion that the objectively right action will usually be a weird outlier, however, this would be a disastrous way to proceed.  I would waste all my time just generating a partial list of weird outliers, and hitting the right one would be at least as hard as  hitting that winning lottery number.

We do not improve the situation by recasting the definition in terms of trying, not to hit the objectively right action, but trying to get reasonably close to it.  All the alternatives reasonably similar to a weird outlier are bound to be pretty weird themselves. Worse, an action very close to a wonderful outlier may not be wonderful at all.  If the winning lottery number is 1-10-13-19-21-28, then you will not throw a party because you entered 1-10-13-19-21-27. Again, where there are chaotic variables we can expect near misses of the objectively right act to be poor choices pretty frequently. So a sensible consequentialism will not lead us to try to perform the objectively right action or an action as similar as possible to that good consequence maximizer, if there is one.

Lucky alternatives, then, do seem to pose a problem for the way theorists of maximizing consequentialism have typically defined their key concepts and for the way consequentialists of a less theoretical bent have usually explained their moral system.  Before trying to solve the problem by improving on the definitions, I want to exhibit a yet more serious flaw in the standard formulations set out in terms of the consequentially best of all the agent's alternatives. The solution to the lucky alternatives puzzle will have to accommodate this more fundamental problem as well.

The Phantom Consequences of Non-Actions

Of all the supposed alternatives before the agent, at most one will be acted upon. It will have consequences. Do the other, non-actual actions also have consequences?  They must, of course, if “right action” picks out the alternative with the best consequences. But do non-actions really have physically or metaphysically sound consequences? (Well, there would be a degenerate form of maximizing consequentialism in which the action actually performed always had the best consequences because none of the other alternatives had consequences at all. This would not be very satisfactory.)

Theorists of consequentialism normally simply assume that that the chain of consequences of an unperformed action, stretching from the instant at which that act was not performed into the unbounded future, exist in some metaphysically respectable way.  Why should we think that?  Let us pass by the problem that there is no unique instant correlated with an action that was never performed.

Consider the metaphysical bona fides of the first paragraph of John F. Kennedy’s Second Inaugural Address, on the assumption that a particular actual trigger pulling had instead been an alternative with the rifle barrel one degree to the right. There is no very good reason to suppose that this extended counterfactual refers to anything, at least any one thing, at all.  Actual events have causal consequences because they are actual. Alternative actions give rise to no causal chains. 

I see one way to save traditionally defined consequentialism from this counterfactual consequences problem. It is a somewhat heroic expedient: assume the truth of physical determinism and define an alternative to be a completely specific possible world state. Then the laws of physics should generate one unique set of (counterfactual) consequences.  I will suppress the suspicion that counterfactual world states are themselves metaphysically defective along with the concern that, even if counterfactual world states exist in some respect, it might not be a respect to which physical law could apply. For argument sake, I will also continue not to worry about the fact that non-acts are associated with no unique world state time slices, and will even be unconcerned about the relativity of time slices in Einstein’s Special Theory.

If you are willing to accompany me to this level of heedlessness, you will have some hope for “consequences” in the traditional consequentialist definitions. In fact, the situation is slightly less bleak than I have made out. Determinism is not strictly required. Indeterminacy of the probabilistic sort favored by current quantum mechanics will do. There will then exist an “expected utility” summation of the consequence tree, assuming that no branch thereof is infinite and that it does not have infinitely many branches. (Do not hang up on the word “expected;” this is a summation that may exist only in metaphysical space, wholly beyond our ken.)

There would be no such summation for a physical indeterminacy not governed by probability, however. So, for example, the idea that human decision making is creative, and absolutely undetermined (strong free will), cannot coexist with consequentialism as traditionally defined. In any event, this rescue of the concept of “consequences,” constrained as it is, turns out to put a great deal of pressure on the idea of “alternatives” to which I now turn.

The Conceptual Bankruptcy of “All Alternatives”

The problem is this: the set of “all alternatives,” typically used in the basic definitions, is, upon close examination, physically/metaphysically arbitrary in a way that makes the concept utterly self-defeating.

As we have just seen, to give some hope that alternatives have determinant consequences, we are driven to think of each alternative as a fully specific possible state of the world at a moment of time. That we end up at this point following traditional definitions is no great surprise.  Maximal specificity is, in fact, the majority view among those theoreticians who have considered in depth the question of alternatives. Fully specific alternatives do, however, invite the question how free we are to be with the contents of these specific counterfactual states of the world. The question is particularly pressing when it comes to that part of the state description that deals with what is in the head of the person acting.

Last night I was sitting at my piano.  I may well have played the first note of the Liechtenstein national anthem.  If I didn’t, certainly I could have played that note, whatever it might turn out to be. Could I have played the whole anthem?  Was that one of my alternatives? Well, I don’t know Oben am jungen Rhein, and I am almost certainly not a good enough pianist to play it even if I had the sheet music, not being much of a pianist at all. So there is an obvious respect in which I could not have played it even had I wanted to.  Yet there is also a respect in which I could have played it.  After all, my piano has all its keys, and there was no anti-Liechtensteiner lurking to blow my head off at the third bar. Perhaps with the music in front of me and a great deal of practice, I could blunder my way through the piece. If so, then there is a possible state of my brain that would have resulted in my playing (generously understood) the anthem.

Of course, that conjectured state of my brain was not its state last night. But then, no alternative, from the most bizarre to the most pedestrian, had a corresponding actual brain state – else it would be the act performed rather than an alternative to that act. (I guess this is not so for the radically undetermined free choice school, but we have already seen that they must part ways with traditionally defined consequentialism at the level of the existence of consequences of unperformed actions.)

Just how large a deviation from what is actually in the actor’s head are we to permit in generating the alternatives to the action performed? Do we also need to adjust the state of the world outside of the actor’s brain to be consistent with such a brain state’s being actual? I hope it is obvious that there are no non-arbitrary answers to these questions.

I could have played what would in fact be the first three notes of the piece, but not the first three bars? Was one of my alternatives to do a back somersault off the piano bench?  Is it relevant that the most obvious somersault scenarios would involve my being coerced into the maneuver at gunpoint or persuaded to it by a substantial monetary proffer – neither of which possibilities was in the offing anywhere near my actual piano last night?  I submit that we simply cannot draw a non-arbitrary line between the “to be included” and “to be excluded” variations of the actual state descriptions.

This is the final nail for the coffin of the traditional definitions of consequentialism in terms of the best consequences of any alternative action. It is not just that those definitions need to be gussied up with an epistemic component to take care of the lucky lottery and the like. The traditional definitions are metaphysically hopelessly defective at such phrases as “best consequences among all alternatives.”

A Proposal

Let me try to fix the problem. Recall that the function of the defective definitions is to link the ideal of the consequential best with the action guiding and praise and blame components of the theory – the working principles. The “objective right” theorists have attempted to render “consequential best” in a rigorous and objective fashion – that is a fashion not relative to the frailties of individual inclination, circumstance, and ignorance. But this, we now have good grounds to believe, is bound to fail.

The first step in the right direction is to recognize that maximization programs do not require the existence of a determinant maximized ideal towards which they aim.  Any claim as to the absolute maximum number of points that could be scored by a World Cup football/soccer side in a regulation length game would depend upon assumptions so numerous and so arbitrary as to be silly. The coach who entreats her players to score as many points as possible is, however, not talking nonsense. Similarly, doing the consequentially best that one can should not be understood as getting as close as one can to a consequentially ideal world. It is a matter of seeking to do ever better, within practical constraints that are themselves generated by the same goal of doing ever better overall.

Cutting all definitional linkage to the best consequences among all alternatives, we can, as a first approximation, set out the working principles of maximizing consequentialism along the following lines: Guide your action by first generating as many and as promising alternatives as it is reasonable to consider given the importance of the decision and the time and resources available. Then choose the alternative that has the best reasonably foreseeable consequences, given those same constraints. Praise those who so guide their actions, and blame those who do not.

A concept of best consequences remains in these principles, but it can now be understood to be nothing more exotic than the winner of a comparison of actual consequence estimates of a manageable number of candidates.

Alternatives have not the complete specificity of the theorists' concept, but are, instead, the sorts of alternatives that we actually think about. Perhaps I considered picking up a cup of coffee on the way and so getting to my lecture a minute late. There are, no doubt, millions of slightly different ways I could have gotten that coffee. The non-bizarre ones I lump together into that single alternative. If you would like, you could think of an alternative for these purposes as a collection of fully specific actions, but I would advise against thinking of it as a set of such actions. Sets have determinant membership, whereas an alternative is vague at the edges, as is typical of the objects of our mental lives.

Before introducing the above working principles, I said that they were only a first approximation. One reason for this caveat is that the “reasonableness” modifier, so prominent in my formulation, is not quite correct. The conscientious maximizing consequentialist will not be satisfied with a reasonable performance.  She will strive to be at her most creative, given the time and resource constraints, in coming up with alternatives that might be better. She will analyze consequences with the greatest sensitivity to data, perspicacity of analysis, and soundness of judgment, again subject to the practical constraints. We may not want to blame someone who has deliberated reasonably, but special praise is due to one who acts on a better than merely reasonable deliberation. This modification is necessary to conform to the spirit of maximization in maximizing consequentialism – to do the consequentially best that we can.  

I conclude that real world consequentialist deliberation is not necessarily misguided even though standard definitions are terrible for the following reasons:  there is no such thing as the set of all an agent's alternatives; it is doubtful that unselected alternatives have consequences, even if there were such a set whose members had consequences, almost all positive assertions of objective right would be false and it would be madness to seek either the sets best consequence member or an approximation thereto; and, per my prior last post, there is no always acceptable notion of maximization because of the average good, total good problems.  

All we need to do is to ditch the idea of a rigorously objective criterion of right and its underlying concept of “best consequences of all alternatives.” We should satisfy ourselves instead with consequence oriented action guiding and praise and blame principles, sensitive to the specifics of the circumstances, including those circumstances that may bring deontology into the deliberations and modify, restrict, or displace consequential reasoning.   

A version of this post, with most of its content and some nice graphics appeared in Philosophy Now, July/August, 2014. Last time I tried, the version of the article available online was missing the nice graphics.