Desire, preference, and utilitarianism

Desire-satisfaction utilitarianism (DSU) holds that the right thing to do is what maximizes everyone’s total desire satisfaction.

This requires a view of desire on which desire does not supervene on preferences as in decision theory.

There are two reasons. First, it is essential for DSU that there be a well-defined zero point for desire satisfaction, as according to DSU it’s good to add to the population people whose desire satisfaction is positive and bad to add people whose desire-satisfaction is negative. Preferences are always relative. Adding some fixed amount to all of a person’s utilities will not change their preferences, but can change which states have positive utility and which have negative utility, and hence can change whether the person’s on-the-whole state of desire satisfaction is positive or negative.

Second, preferences cannot be compared across agents, but desires can. Suppose there are only two states, eating brownie and eating ice cream (one can’t have both), and you and I both prefer brownie. In terms of preference comparisons, there is nothing more to be said. Given any mixed pair of options i = 1, 2 with probability p_i of brownie and 1 − p_i of ice cream, I prefer option i to option j if and only if p_i > p_j, and the same is true for you. But this does not capture the possibility that I may prefer brownie by a lot and you only by a little. Without capturing this possibility, the preference data is insufficient for utilitarian decisions (if I prefer brownie by a lot, and you by a little, and there is one brownie and one serving of ice cream, I should get the brownie and you should get the ice cream on a utilitarian calculus).

The technical point here is that preferences are affine-invariant, but desires are not.

But now it is preferences that are captured behavioristically—you prefer A over B provided you choose A over B. The extra information in desires is not captured behavioristically. Instead, it seems, it requires some kind of “mental intensity of desire”.

And while there is reason to think that the preferences of rational agents at least can be captured numerically—the von Neumann–Morgenstern Representation Theorem suggests this—it seems dubious to think that mental intensities of desire can be captured numerically. But they need to be so captured for DSU to have a hope of success.

The same point holds for desire-satisfaction egoism.

Utilitarianism, egoism and promises

Suppose Alice and Bob are perfect utilitarians or perfect amoral egoists in any combination. They are about to play a game where they raise a left hand or a right hand in a separate booth, and if they both raise the same hand, they both get something good. Otherwise, nobody gets that good. Nobody sees what they’re doing in the game: the game is fully automated. And they both have full shared knowledge of the above.

They confer before the game and promise to one another to raise the right hand. They go into their separate rooms. And what happens next?

Take first the case where they are both perfect amoral egoists. Amoral egoists don’t care about promises. So the fact that an amoral egoist promised to raise the right hand is no evidence at all that they will raise the right hand, unless there is something in it for them. But is there anything in it for them? Well, if Bob raises his right hand, then there is something in it for Alice to raise her right hand. But note that this conditional is true regardless of whether they’ve made any promises to each other, and it is equally true that if Bob raises his left hand, then there is something in it for Alice to raise her left hand.

The promise is simply irrelevant here. It is true that in normal circumstances, it makes sense for egoists to keep promises in order to fool people into thinking that they have morality. But I’ve assumed full shared knowledge of each other’s tendencies here, and so no such considerations apply here.

It is true that if Alice expects Bob to expect her to keep her promise, then Alice will expect Bob to raise his right hand, and hence she should raise her right hand. But since she’s known to be an amoral egoist, there is no reason for Bob to expect Alice to keep her promise. And the same vice versa.

What if they are utilitarians? It makes no difference. Since in this case both always get the same outcome, there is no difference between utilitarians and amoral egoists.

This means that in cases like this, with full transparency of behavioral tendencies, utilitarians and amoral egoists will do well to brainwash or hypnotize themselves into promise-keeping.

In ordinary life, this problem doesn’t arise as much, because as long as at least one person is more typical, and hence takes promises to have reason-giving force, or if public opinion is around to enforce promise-keeping, then the issue doesn’t come up. But I think there is a lesson here and in the previous post: for many ordinary practice, the utilitarian is free-riding on the non-utilitarians.

A defense of natural law eudaimonism

My main objection to natural law ethics has for a long time been that it looks egoistic because it is eudaimonistic. One version of that worry is the “one thought too many” objection: You should just do good to your fellow humans because they are who they are, because they are your fellow human beings, or something like that, but definitely not because doing so leads to your flourishing.

I think there is a nice—and probably well-known to people other than me—response to this version of the worry, and to many similar “one thought too many” worries. To put this “one thought too many” worry more abstractly, the worry is that the metaethics will infect the reasons for action in an unacceptable way. But the response should simply be that, first, what metaethics asks is this question:

1. What makes the reasons for action be reasons for action?

Here, read “reasons” factively as “good reasons” or even “good moral reasons” (I don’t actually distinguish the two, but many do), not as motivations. And, second, insofar as R is my reason for my action, I am acting on account of R, not on account of R being a reason. Compare: what causes the fire is the match, not the match’s being a cause.

Thus, the natural lawyer should say that what makes the fact that an action promotes the good of my neighbor be a reason is that I flourish (in part) by intentionally (under this description) promoting the good of my neighbor. But the reason for the action is that the action promotes the good of my neighbor, not that I flourish by intentionally promoting the good of my neighbor. The natural law answer to the metaethics question (1) is this:

2. R is on balance a reason for action if and only if, and if so then because, I flourish by acting on R.

We do in fact flourish by intentionally promoting the good of our neighbor. Note that (2) does not by itself yield any egoism in our motivations. We could imagine selfless beings that flourish only insofar as they are intentionally promoting the good of their neighbor as a final end, and who are blighted insofar as they are intentionally promoting their own good or flourishing. We are, of course, not such selfless beings, but we don’t learn the fact that we are not such beings from (2). In fact, (2) is fully logically compatible with us being such beings. Hence, the metaethical theory (2) cannot by itself give rise to the “one thought too many” worry I started the post with. (Of course, some natural lawyers will go beyond (2). They may say that in fact our happiness is the end of all our actions. If so, then I think they are subject to the “one thought too many” worry.)

It is important to add a little bit to the above story. While it is true that “this benefits my friend” is typically reason enough, and that I don’t need to act on the second order fact that “this benefitting my friend is a reason”, we also do have such second order reasons. That there is a reason for an action is itself a reason for action. A parent might tell a child: “You have good reason to do this, but I can’t explain the reason right now.” In that case, the child could well be acting on the second-order reason that there is a first-order reason. (The child could also be acting on a first-order reasons to please the parent).

Here is another kind of case. I start off without any belief about whether R is a reason for action, and R leaves me cold. Maybe I am completely insensitive to considerations of privacy, and the fact that an action promotes someone’s privacy just leaves me completely cold. But I observe my virtuous friends, and see that they are acting on reasons like R, and I notice that their so acting contributes to what I admire about them. I conclude that R is in fact a good reason for action. But that’s purely intellectual. I am still left quite cold and unmotivated by the fact that some proposed action A falls under R. But what I can do at this point is to act on the second-order reason that A falls under a good reason. I can even say what that good reason is. But I cannot act on it itself, because it leaves me cold.

These are, however, non-ideal cases. If I know that R is a good reason, I should strive to form my will to be motivated by R. It will be better to act on R than to act on the knowledge that R is a reason. And thinking about these cases makes the response to the “one thought too many” worry about natural law even more compelling, I think. It does promote my flourishing to promote my flourishing, though I think that it doesn’t promote my flourishing as well as promoting the flourishing of others does. So that kindliness to others promotes my flourishing is a reason for benefiting others, just not as good a reason as that it benefits others. But such “not as good reasons” are important for our moral development: we are not yet in the ideal state, and so that “one thought too many” is still needed.

This helps make me feel a lot better about natural law ethics. Not quite enough to embrace it, though.

Present moment ethical egoism

One of the least popular ethical theories is present moment ethical egoism (pmee): you ought do what produces the state that is best for you at the present moment.

But pmee has a very lovely formal feature: it can be used to simulate every other ethical theory of permissibility, simply by changing the value function, the function that ranks states in terms of how good they are for you at the present moment. To simulate theory T, just assign value 1 to one’s presently choosing an action that T says is permissible and value −1 to one’s presently choosing an action that T says are impermissible. In this way, pmee simulates Aquinas, Kant, virtue ethics, utilitarianism, non-present-moment egoism, etc.

This formal feature is not shared by non-egoistic consequentialist theories. For the only way a consequentialist theory can simulate a deontological theory is by assigning an overwhelmingly large negative value to wrong choices. But this gives a result incompatible with many deontological theories, namely that you should choose to commit a murder in order to prevent two other people from doing the same.

The formal feature is also not shared by egoistic but not present-moment theories. For on some deontological theories, it is wrong to commit a murder now in order to prevent oneself from choosing two murders later.

Here is another curious thing. Basically, the only present thing in my present control is my present choice. This means that pmee cannot be a consequentialist theory in the typical sense of the word, because all causal consequences take time, and hence every causal consequence within my present control is in the future. In other words, it is the value of the present choice that pmee needs to focus on (both in itself, and in a larger context).

But once we see that it is the value of the choice itself, and not the causal consequences of the choice, that pmee must base a decision on, then given the fact that the most compelling value that a choice has is its moral value, it seems that pmee tells one that one should do what is morally right. And what is morally right cannot be defined in terms of pmee on pain of circularity.

This is a Parfit-like thought, of course. (Maybe even exactly something from Parfit. It’s been a while since I’ve read him.)

Hindsight and altruism

Forty years ago you and your brother foughts about who would get to play with the new train. You remember the fight, but not the outcome. At the time, you really wanted to win the fight and play with the train. Now you no longer have any strong preference for having been the one who got to play with the train.

In the view from distant hindsight, we do not have strong preferences that we were the ones who obtained some good—at least if we are talkiing of material goods. Now Simone Weil thinks, and I think common sense supports her, that the view from distant hindsight is the true and objective way of looking at a situation. Since it is not rational to see it as strongly preferable that we had long ago received some material good rather than someone else getting it, neither should we have such a stroong preference in regard to near-future goods. This is an argument for altruism and against egoism.