PRÉCIS OF
BREAKDOWN OF WILL
George Ainslie
116A Veterans Affairs Medical Center
Coatesville, PA 19320

Ainslie@Coatesville.va.gov
2/1/01

Presented at the GREMA/CEPR conference,
Economics and Psychology, at the
Universite des Sciences Sociales,
Toulouse, France, June 18, 1999
Breakdown of Will is now available from
Cambridge University Press.
And from Amazon.com

File Bbsp99

SHORT ABSTRACT

There is now ample evidence that both people and lower animals discount the value of expected events in a hyperbolic curve. Such a curve predicts that contradictory reward-getting processes can survive in a stable relationship of limited warfare with each other. It also suggests a means by which this relationship can lead to behavioral consistency, in people at least: the perception of repeated prisoner's dilemma contingencies in recurrent choices. The motivational pattern that should result from this perception resembles traditional descriptions of the will, as well as of compulsive phenomena that can now be seen as side-effects of will: overconcern with precedent, areas of intractable dyscontrol, lapses of awareness, and a loss of intuitiveness. Thus understood, the will is also a factor in the "social construction" of reality and the value of empathic relationships.

Key words: Volition, intertemporal bargaining, picoeconomics, impulsiveness, compulsions, self-control, overcontrol, dynamic inconsistency, akrasia, weakness of will, freedom of will, emotions, social construction, altruism, empathy.

LONG ABSTRACT

There is now ample evidence that both people and lower animals discount the value of expected events in a hyperbolic curve. This curve implies a basic instability of choice, and has thus promoted the problem of estimating value from a trivial matter of psychometrics into the crucial element of self-control. Although hyperbolic discounting seems complex when compared with the conventional, exponential kind, it fits so many aspects of motivational conflict that it can simplify that subject substantially:

With hyperbolic discounting, options that pay off quickly will be temporarily preferred to richer but slower-paying alternatives, a phenomenon that can't be changed by insight per se. However, where people come to look at their current choices as predictors of what they will choose in the future, a logic much like that in the familiar bargaining game, repeated prisoner's dilemma, should recruit additional incentive to choose the richer processes. This mechanism predicts all the major properties that have been ascribed to both the power and freedom of the will.

Intertemporal bargaining also predicts four serious side effects of willpower: A choice may become more valuable as a precedent than as an event in itself, making people legalistic; signs that predict lapses should become self-confirming, leading to failures of will so intractable that they seem like symptoms of disease; there is motivation not to recognize lapses, which might create an underworld much like the Freudian unconscious; and concrete personal rules should recruit motivation better than subtle ones, a difference which could impair the ability of will-based strategies to exploit emotional rewards.

Furthermore, hyperbolic curves predict an urge to satisfy appetite prematurely. This urge should foster efficiency of problem-solving but undermine emotional rewards insofar as they become expectable. The incentive to preserve appetite will select for emotional occasions that are both surprising and sufficiently scarce; two strategies, gambling on unpredictable events and finding dummy activities that approach reward indirectly, can increase long range reward by increasing appetite. This incentive is apt to be a factor in both our "construction" of reality and the value we place on empathic relationships. It can be expected to operate at crossed purposes with the will in many circumstances and thus limit its power.

1. Introduction

Despite the sophistication of modern civilization, people continue to engage in self-defeating behaviors-- in consuming drugs, alcohol and cigarettes, in gambling, in forming destructive relationships, in failing to carry out our own plans, and most simply in procrastinating. In a wide range of situations where a long range payoff is alternative to a poorer payoff that's available earlier, we prefer the poorer payoff temporarily, in the period when it would be imminent. Theories of why we keep making such choices, often in full knowledge of their consequences, no longer mention failures of will (e.g. Becker & Murphy, 1988; Baumeister & Heatherton, 1996; Rachlin, 1995; Polivy, 1998). It's just not a concept that behavioral scientists liked to use in the twentieth century. Some writers have even proposed that there's no such thing as a will, that the word refers only to someone's disposition to choose. Still, the word crops up a lot in everyday speech, especially as part of "willpower," something that people still buy books to increase.

It's widely perceived that some factor like will transforms motivation from being a simple reflection of the incentives we face to a process that is somehow ours, that perhaps even becomes us-- a factor that lies at the very core of choice-making. Unfortunately, there has been no way to talk about such a factor in the language of science, that is, in a way that relates it to simpler or better understood elements. Without addressing this factor, science paints a stilted picture of decision-making. I will argue that quantitative motivational research has produced a finding that promises to account for the phenomenon of will-- and the experience of willing-- using elements that are already familiar.

In most disciplines rationality means maximization of expected utility-- or expected reward, as psychology is more apt to call it. That is, rational choice-makers are said to be those who choose more valuable over less valuable alternatives, discounted exponentially from their expected time of occurrence back to the moment of choice. Utility theory imagines an internal marketplace where options compete for adoption on the basis of their utility; rationial choicemakers are those with the most efficient markets. This approach has been a great success at predicting actual preferences, at least among alternatives that are roughly simultaneous-- so much so that the whole gamut of choice-analysts, from sociobiologists through behavioral psychologists to economists and sociologists, accept utility as the elementary basis of choice.

However, utility theory has had trouble accounting for why people experience temporary preferences for poorer but earlier alternatives. In particular, it doesn't explain why we're motivated to forestall these preferences in advance but often fail in the attempt. Proposed explanations have included steep exponential discount curves from delayed rewards (e.g. Becker & Murphy, 1988), limited time horizons (e.g. Logue, 1988), and mechanisms that generate exceptional motivations-- particularly "classical" conditioning (e.g. Loewenstein, 1996, O'Brien et.al., 1986). None of them account for the motive to forestall future preferences (Ainslie, 1992, pp. 24-55, and 1999). A more radical approach is necessary: a utility theory based on nonexponential discount curves.

2. Hyperbolic Discounting Can Cause Temporary Preferences

The main theoretical rival to the exponential discount curve is the hyperbolic discount curve, which is inversely proportional to expected delay and thus more bowed than an exponential curve. With hyperbolic discounting, goods that would be valued the same as with exponential discounting at very short and very long delays would be valued less at middling delays (figure #1).

FIGURE 1

An exponential discount curve and a hyperbolic (more bowed) curve from the same reward. As time passes (rightward along the horizontal axis), the motivational imipact-- the value-- of the goal gets closer to its undiscounted size, which is depicted by the vertical line.

Deeper bowing means that if a hyperbolic discounter engaged in trade with someone who used an exponential curve, she'd soon be relieved of her money. Ms. Exponential could buy Ms. Hyperbolic's winter coat cheaply every spring, for instance, because the distance to the next winter would depress Ms. H's valuation of it more than Ms. E's. Ms. E could then sell the coat back to Ms. H every fall when the approach of winter sent Ms. H's valuation of it into a high spike. Because of this mathematical pattern, only an exponential discount curve will protect a person against exploitation by somebody else who uses an exponential curve. Thus exponential curves seem not only rational, in the sense that they motivate consistent choice over time, but also adaptive. At first glance, it looks as if natural selection should have weeded out any organism that didn't discount the future exponentially.

Nevertheless, there's more and more evidence that people's natural discount curve is not only nonexponential, but specifically hyperbolic. Asked their preferences between pairs of a smaller amount of money at delay D and a larger amount at delay D + a lag L, people regularly reverse their preferences between the same pair as a function of D, and show an overall pattern of choice described by

Value = Value-if-immediate / (1 + (Constant x Delay))

(Ainslie & Haendel, 1983; Green et.al., 1994; Kirby & Herrnstein, 1995). Investigators sometimes report that their data fit still better if the denominator is raised to a power (Myerson & Green, 1995), but this power is usually close to 1.0, and in any case doesn't change the crucial implication of this formula: that the elementary discount curve produces a basic tendency to prefer smaller rewards over larger ones temporarily when the smaller reward is imminently available (figure 2).

FIGURE 2

Hyperbolic discount curves from two rewards of different sizes available at different times. The smaller reward is temporarily preferred for a period just before it's available, as shown by the portion of its curve that projects above that from the later, larger reward.

Other examples include people choosing between various amounts and delays of primary (not token) rewards such as nourishment, video game time, or relief from noxious noise (Ragotzy et.al., 1988; Millar & Novarick, 1984; Solnick, et.al., 1980) and people making real, very long term investment decisions (Harvey, 1994). Laibson and collaborators have found that discount curves which are more bowed than exponential ones better fit people's lifetime patterns of spending and saving, and might explain such notorious anomalies as borrowing at 15% while keeping investments that earn 5% (1997; Harris & Laibson, 1999). An even larger body of animal experiments shows that temporary preference for poorer, earlier rewards is not an artifact of human culture, but a fundamental property of reward (Reviewed in Ainslie, 1992, pp. 63-80).

Authors have proposed several discounting models to explain the temporary preference phenomenon without abandoning exponential discounting, such as:

-- a step function in which immediate events are valued extraordinarily and events at all delays are discounted exponentially (Simon, 1995);

-- an exponential discount rate whose exponent itself varies as a function of delay (Green & Myerson, 1993);

-- the summation of separate exponential discount rates for association and valuation (Case, 1997); and

-- random variation in discount rate (first suggested by Strotz, 1956, and recently enlarged upon by Skog, 1999).

However, the data behind each of these are scanty compared to those behind hyperbolic discounting; and none of these data contradict the simpler hypothesis of hyperbolic discounting.

Does the apparent universality of hyperbolic discounting mean that utility theorists through the ages have been wrong-- that philosophers and bankers and welfare economists should have been calculating the worth of goods using the deeply bowed curves of figure 2? This can't be true, either. We saw in the overcoat example that exponential discounting is better than hyperbolic, in the sense that exponential discounters win out in competition with hyperbolic discounters.

The conventional answer would almost certainly be that since only exponential curves produce consistent preferences, they're the ones that are objectively rational, and that people should learn to correct their spontaneous valuations to fit them. After all, science has long known that many other experiences are described by hyperbolic curves of felt intensity as a function of objective stimulation (Gibbon, 1977), and that people can learn to correct such impressions. You develop "object constancy." Can't people learn to value reward in proportion to its objective amount, just as we gauge objective brightness and distance?

That's what conventional utility theory calls for; but despite information from clock and calendar such an adjustment seems to occur irregularly, sometimes not at all. It usually takes some kind of effort ("willpower" again) to evaluate a smaller imminent satisfaction as less desirable than a greater one in the future. This is where the analogy of delay to other sensory impressions like length breaks down:

You may move through time toward a goal just as you move through space toward a building. The formula describing your spontaneous valuation of a goal is close to the formula for the retinal height of the building (Y=l/X, where Y is the magnitude in question and X is the distance to the building or goal). But the building doesn't seem to get larger as it gets closer, whereas the goal often seems to get more valuable. Insofar as you fail to make the correction in value that corresponds to your correction of retinal height, poorer goals that are close can loom larger than better, distant goals.

This failure shouldn't be surprising. If reward is the fundamental selective force of choice, then however you perceive or categorize it, you're still acted upon by its direct influence. While a raw process that constitutes the active determinant of value can be perceived abstractly, it shouldn't occur differently because of this abstraction.

The self as a population. Hyperbolic discounting shifts the main problem for utility theory. We're no longer at a loss to explain choices that are short-sighted and temporary. Now we have to account for how people learn the adaptive controls that let them behave like bankers. How does an internal marketplace that disproportionately values immediate rewards grow into what can be mistaken for the long range reward-maximizer of conventional utility theory?

We can no longer regard people as having unitary preferences. Rather people may have a variety of contradictory preferences that become dominant at different points because of their timing. The orderly internal marketplace pictured by conventional utility theory becomes a bazaar of partially incompatible factions, where in order to prevail an option has not only to promise more than its competitors, but to act strategically to keep the competitors from later turning the tables. The behaviors that are shaped by the competing rewards must deal not only with obstacles to getting their reward if chosen, but with the danger of being un-chosen in favor of imminent alternatives.

An agent who discounts reward hyperbolically is not the straightforward value estimator that an exponential discounter is supposed to be. Rather she will be a succession of estimators whose conclusions differ; as time elapses these estimators shift their relationship with one another from cooperation on a common goal to competition for mutually exclusive goals. Ulysses planning for the Sirens must treat Ulysses hearing them as a separate person, whom he must influence if possible and forestall if not. Obviously if what you do in a situation regularly gets undone later, you'll learn to stop doing it in the first place-- but not out of agreement with the later self that undoes it, only out of realism. You'll keep trying to find ways to get what you want from the earlier vantage point, steps that won't get undone, and take precautions against a future self who will try to undo them. You'll be like a group of people rather than a single individual; subjectively, however, the results of learning to do this may feel like no more than choosing among an array of intentions.

You could call the mental operations selected for by a particular kind of reward the person's interest in that reward: Interests within the person should be very like interests within a social group, those factions that are rewarded by ("have an interest in") the goal that names them (e.g. "the petroleum interest," "the arts interest"). Since a person's purposes should be coherent except where conflicting rewards dominate at successive times, it makes sense to name an interest only in cases of conflict. I wouldn't be said to have separate chocolate and vanilla ice cream interests, even though they're often alternatives, because at the time when I prefer chocolate I don't increase my prospective reward by forestalling a possible switch to vanilla. But I may have an ice cream interest and a diet interest, such that each increases prospective reward in its own time range by reducing the likelihood of the other's subsequent dominance. Put another way, I don't increase my prospective reward in either the long or short range by defending my choice of chocolate against the possibility that I may change to vanilla; but I increase my prospective long range reward by defending my diet against ice cream, and I increase my prospective short range reward by finding evasions of my diet for the sake of ice cream. Whichever faction promises the greatest discounted reward at a given moment gets to decide my move at that moment; the sequence of moves over time determines which faction ultimately gets its way.

Where the alternative rewards are available at different times, each will build its own interest, and one interest will be able to forestall the other only if it can leave some enduring commitment that will prevent the other reward from becoming dominant: If my diet interest can arrange for me not to get too close to ice cream, the discounted prospect of ice cream may never rise above the discounted prospect of the rewards for dieting, and the diet interest will effectively have won. However, whenever the value of ice cream spikes above that of dieting, the ice cream interest may undo the effect of many days of restraint. The ultimate determinant of a person's choice is not her simple preference, any more than the determinant of whether a piece of legislation becomes law is simple voting strength in the legislature; in both situations, strategy is all.

3. Will as Intertemporal Cooperation

The hyperbolic discounting hypothesis can account for most familiar self-control tactics: The majority involve using your currently dominant motivation to change future options or motives: You can use external devices like physically committing yourself-- tie yourself to the mast à la Ulysses -- or arranging for social influence. You can restrict your attention; or you can either excite your emotions or nip contrary emotions in the bud (see Ainslie, 1992, pp. 130-144, and in press). However, these tactics are less adaptable, and often less available, than what is usually called willpower. Willpower may be the commonest way that people deal with temporary preferences, but also the most mysterious. What is there about "making a resolution" that adds anything to your power to resist changing motivations? When people have given up smoking or climbed out of debt they mostly say they "just did it." Words like willpower, personal rules, character, intention, and resolve are often applied, but don't suggest how people actually resist temporary preferences for shortsighted options.

Writers over the centuries have described some specific properties: The most robust idea is that will comes from turning individual choices into a matter of principle. As early as the fourth century BC Aristotle proposed this idea (referring to dispositions to choose as "opinions"): "We may also look to the cause of incontinence scientifically in this way: One opinion is universal, the other concerns particulars... (Nichomachean Ethics, 1147a24-28)." Galen said that passion was best controlled not by looking at individual opportunities but by following the general principles of reason (1963, p. 44).

The Victorian psychologist Sully observed that the key property of the will was to "unite... particular actions... under a common rule," so that "they are viewed as members of a class of actions subserving one comprehensive end (1884, p. 631)." Recently two researchers from the behavioral school have reaffirmed this idea. Heyman (1996) and Rachlin (1995) have both suggested that choosing in an "overall" or "molar" pattern (respectively) will approach reward-maximizing more than a "local" or "molecular" one.

Hyperbolic discounting implies just such an effect, assuming only that the discounted impact of successive rewards is roughly additive. Summation of rewards hasn't been studied much, but a few experiments suggest that the hyperbolically discounted effects of each reward in a series simply add, at least in pigeons (analyzed in Mazur, 1997 and in Ainslie, 2001, p. 213-214). Data from some other experiments suggest that there is a more general but complex form of the matching equation that lets it predict choice of extended reward series even better (Grace, 1994; Mazur, 1997), but it doesn't change the implications for our analysis.

From here on I'll assume that the discount curves from rewards that are chosen together add together, which is the simplest assumption as well as the one that goes along with the available data. If I'm at least roughly correct, then we have an explanation for why making choices in categories can lead to less impulsiveness:

Consider a series of larger, later rewards and their smaller, earlier alternatives-- for instance, philosopher Michael Bratman's example of a pianist who throws off his nightly performance by drinking wine beforehand (1999, pp. 35-57). At a distance he prefers to abstain and perform well, but each night at dinnertime the pianist changes his preference to drinking the wine. However, as figure 3A suggests, even at dinnertime he may prefer abstaining all nights to drinking all nights for the foreseeable future. The incentives for choosing between these categories of reward will be the expected values of the series of rewards. The incentives for choosing just for one night will be just the curves from a lone pair, as in figure 2.

FIGURE 3A

Summed hyperbolic curves from a series of larger-later rewards and a series of smaller-earlier alternatives. As later curves are added to the series, the earlier periods of temporary preference for the series of smaller rewards shrink to zero. The curves from the final pair of rewards are the same as in figure 2.

FIGURE 3B

Summed exponential curves from the same series. Summing doesn't change their relative heights. (This would also be true if the curves were so steep that the smaller, earlier rewards were preferred; but in that case summing would add little to their total height, anyway, because the tails of exponential curves are so low.)

In the schooner-like picture of the summed discount curves from series of rewards, the "sails" get gradually lower as the choice point moves later in the series, for they comprise a decreasing number of curves added together. The last pair of sails are the same as a lone pair. If the series has no foreseeable end, which is the case for most real-life categories, the sails may be added forward to a time horizon that stays a constant distance ahead, so that the height of the summed rewards stays roughly constant. In any case, summation of a series of rewards makes the first few sails higher than the sails drawn from a lone pair of rewards would be.

More importantly, the delayed rewards add roughly in proportion to their objective sizes, so that when their aggregate height is added in, the first sail in the series of smaller rewards doesn't protrude as high above the first sail in the series of larger rewards as it would in a solitary pair; with series of some amounts at some delays, the earliest sail doesn't protrude above the larger-later sails at all. That would mean that the pianist would always prefer to abstain, even at dinnertime. The choice of a whole series of rewards will be influenced by the rewards expected after the most immediate pair, and for all the subsequent pairs the discounted value of the larger, later alternatives is greater than that of the smaller, earlier ones. Only the nearest choice in a series is dominated by the smaller, immediate reward-- although the nearest choice will obviously carry more weight than any single one of the later choices. By contrast, exponential discount curves from a single pair stay proportional, and adding the whole series together doesn't change their relative heights (figure 3B). Bundling choices together wouldn't affect the direction of preference if discount curves were exponential rather than hyperbolic.

Most choices in real life aren't between brief moments of different intensity, but between extended experiences-- the pleasure of a binge vs. feeling fit and having intact prospects Monday morning, or a good venting of rage vs. keeping a job and friends. Often the difference isn't between intensities of satisfaction-per-minute, but between different durations of comparable satisfactions. The pleasure of staying up for a couple more hours after midnight may be the same as the differential pleasure of feeling alert the next day, for instance, but the alertness lasts all day. However, if successive rewards are additive, it's easy to convert durations to total amounts. If you value the fun of staying up at one unit per minute and expect to lose one unit per minute of comfort from when you get up at 7:00 the next morning until you leave work at 17:00, your discount curves from a day's aggregation of these rewards will look like those in figure 4A.

FIGURE 4A

Curves that are the aggregate of hyperbolic discount curves from continuing rewards-- staying up from midnight to 2:00 vs. feeling rested from 7:00 to 17:00.

FIGURE 4B

Summed curves from ten pairs of the rewards depicted in 4A. The effect of summation is the same as for the point rewards in figure 3B.

If you face a similar choice nightly and can make your choice for a long series of future nights all at once, your incentives will be described by the curves in figure 4B (simple arithmetic derivations in Ainslie, 1992, pp. 155-162). As with more discrete moments of reward, bundling these experiences into series moves preferability toward the larger, later rewards.

So choosing behaviors in whole categories should lead to less impulsiveness, just as the philosophers noted above have said. Two recent experiments confirm that choosing between whole series of small-early vs. large-late pairs increases preference for the large-late alternatives. Psychologists Kris Kirby and Barbarose Guastello found that undergraduates who preferred the small-early amount of money when choosing between one pair at a time regularly switched to preferring the large-late amounts when choosing between series of five payoffs to be delivered at weekly intervals. The same switch occurred when amounts of pizza were offered rather than money (unpublished manuscript, 2000). Similarly, psychologist John Monterosso and I have observed that rats choosing between squirts of sucrose choose the shorter-earlier of a single pair of squirts, but a series of three longer-later squirts over a series of three shorter-earlier alternatives (unpublished manuscript, 2000). The finding in rats suggests that the bundling effect comes from a basic property of the discount curves, rather than some cultural norm.

This phenomenon implies that you will serve your long range interest if you can make a personal rule to behave alike toward all the members of a category. This is the equivalent of the philosopher Kant's "categorical imperative:" to make all choices as if they were to define universal rules (1993/1960, pp. 15-49). Similarly, it echoes the psychologist Lawrence Kohlberg's sixth and highest principle of moral reasoning, deciding according to principle (1963).

The bundling phenomenon also explains how people with fundamentally hyperbolic discount curves can sometimes learn to choose as if their curves were exponential. Bundling whole series of choices together makes their summed discount curve look more exponential, as shown in figure 5. If you adopt a personal rule to "discount all significant income at 6% per year," summed hyperbolic curves from all the expected amounts might be enough to motivate it. Summed hyperbolic curves from whatever goods accrued from the whole practice of exponential discounting might motivate rates of 3%, or any other rate including 0%; but the lower the rate to be enforced, the less the amount of imminent consumption that could be resisted.

FIGURE 5A

Summed hyperbolic discount curves from eleven rewards, compared with the same single exponential curve and single hyperbolic curve as were shown in figure 1. The summed curves come much closer to exponential discounting than the lone hyperbolic curve does.

Bundling's Achilles heel. So far I've described no reason why a person under the influence of a short range reward should be motivated to choose categorically on the occasion at hand. Awareness of the effect of bundling choices doesn't eliminate the attraction of small, immediate rewards; it offers only a discipline from which a long range interest would benefit from at the expense of short range interests, if you were consistently motivated to follow it. People who have learned a "higher" or "richer" principle of choice won't be freed thereby from temptation. A perverse truth will almost certainly become conspicuous before a person gets very old: If behaving according to categorical principles promises more discounted, expected reward than making isolated choices does, then making an isolated choice now and acting by rule in the future promises still more.

The problem is that there are many possible ways to define global categories. The ice cream at hand may violate one diet but not another; and even if it's so outrageously rich as to violate all conceivable diets, there's apt to be a circumstance that makes the present moment an exception: It's Thanksgiving dinner or my birthday, or a host has taken special trouble to get it, or I have cause to celebrate or to console myself just today, etc. The molar principle that offers an exception just this once will be rewarded more than the one that doesn't, for it produces the aggregation of rewards in figures #3A and #4B for all but the first larger-later reward, and the first early spike of reward.

The tough question is not how molar bookkeeping recruits motivational support for long range interests, but how this process defends itself from short range interests, sometimes unsuccessfully. Acting in my long range interest, how do I keep a short range interest from repeatedly proposing an exception to my rule, "just this once?"

Simple intuition offers an answer: Excuses that are too blatant lower my expectation of following the amended principle. I may be able to go off my diet on Thanksgiving without reducing my belief that I'll stick to it at other times; but if I try proposing other holidays I'll probably notice that I'm starting down a slippery slope. The exceptional status of copious drinking at New Years or similar eating at Thanksgiving provides flexibility to a potentially rigid commitment to a concrete rule; but the same principle that keeps the Thanksgiving feast from setting a precedent might also work for my birthday, and, with decreasing credibility, for the Fourth of July, St. Patrick's Day, Labor Day, Arbor Day, St. Swithin's Day, and Just This Once. This kind of logic can degrade my personal rule without my ever breaking it. Once I expect myself to find an exception whenever the urge is strong, I no longer have the prospect of the whole series of later, larger rewards-- the cumulative benefits of my diet-- available to motivate abstention.

In this way hyperbolic discount curves make self-control a matter of self-prediction. This effect will be especially noticeable where self-control is tenuous. The hyperbolic discounter who overeats can't simply estimate whether she's better off limiting her food intake or eating spontaneously, and then follow the best course, the way an exponential discounter could. Even if she figures that restricting herself is better from the perspective of distance, she won't know whether or not she'll regularly prefer to eat ad lib when she's hungry. If she expects to eat ad lib, her long range perspective will be useless to her unless she can use one of the first three kinds of precommitments I mentioned at the beginning of this section-- not a rich selection.

But what if she makes a resolution to "decide according to principle"-- to go on a diet-- and starts off expecting to stick to it in the future if she sticks to it now? This condition may be enough to motivate sticking to it, but only insofar as she thinks it will be both necessary and sufficient. If she then violates the diet and loses faith in it, her principle will magically stop being enough. Personal rules are a recursive mechanism; they continually take their own pulse, and if they feel it falter, that very fact will cause further faltering.

Thus deciding according to molar principles is not a matter of making dispassionate judgments, but of defending one way of counting your prospects against alternative ways that are also strongly motivated. The incentive to stick to a principle is not pure a priori reason-- reason is not motivation-- but the saving of your expectation of continuing to stick to it. It's the internal equivalent of the "self-enforcing contracts" made by traders who'll be dealing with each other for a long time, contracts that let them do business on the strength of handshakes (Macaulay, 1963; Klein and Leffler, 1981). This recursive process of staking the credibility of a resolution on each occasion when it's tested gives your resolve momentum over successive times. The ongoing temptation to set a damaging precedent-- and the ever-present anxiety that this may happen-- is probably what makes this strategy of self-control feel effortful. It separates intentions from plain expectations, and force of will from mere force of habit.

Self-enforcing contracts are solutions to the familiar bargaining game, repeated prisoner's dilemma. This is a useful way of modeling the will-- like the "will" of nations not to start a nuclear war, rather than a cognitive hierarchy of some kind-- but it needs to be modified for the intertemporal case. As Bratman has correctly argued (1999, pp. 35-57), a present "person-stage" can't retaliate against the defection of a prior one, a difference that disqualifies the prisoner's dilemma in its classical form as a rationale for consistency.

Picture a lecture audience. I announce that I'll go along every row, starting at the front, and give each member a chance to say "cooperate" or "defect." Each time someone says "cooperate" I'll award ten cents to her and to everyone else in the audience. Each time someone says "defect" I'll award a dollar only to her. And I ask that they play this game solely to maximize their individual total income, without worrying about friendship, politeness, the common good, etc. I say that I will stop at an unpredictable point after at least twenty players have played, at which time each member can collect her earnings.

Like successive motivational states within a person, each successive player has a direct interest in the behavior of each subsequent player; and she'll guess their future choices somewhat by noticing the choices already made. Realizing that her move will be the most salient of these choices right after she makes it, she has an incentive to forego a sure dollar, but only insofar as she thinks that this choice will be both necessary and sufficient to persuade later players to do likewise. If previous players have been choosing dollars she's unlikely to estimate that her single cooperation will be enough to reverse the trend. However, if past choices have mostly been dimes, she has reason to worry that her defection might stop a trend that both she and subsequent players have an incentive to support.

Knowing the other audience members' thoughts and characters-- whether they're greedy, or devious, for instance-- won't help a person choose, as long as she believes them to be playing to maximize their gains. This is so because the main determinant of their choices will be the pattern of previous members' play at the moment of these choices. Retaliation for a defection won't occur-- a current player has no reason to reward or punish a player who won't play again-- but what amounts to retaliation will happen through the effect of this defection on subsequent players' estimates of their prospects. So each player's choice of whether to cooperate or not is still strategic.

These would seem to be the same considerations that bear on successive motivational states within a person, except that in the lecture audience the reward for future cooperations is flat (ten cents per cooperation, discounted negligibly, rather than discounted in a hyperbolic curve depending on each reward's delay). Like an individual person trying to stick to a diet, the audience will either "have faith in" itself or not. It may forgive itself an occasional lapse, but once a run of defections has occurred it's unlikely to recover its collective willpower without some event that gives it a new deal. Furthermore, an audience member who is conspicuously different from the others-- a lone man among women, for instance-- might reasonably expect to get away with choosing a dollar without causing a stampede of defections; he is the logical equivalent of the self who goes off her diet on Thanksgiving.

This model proceeds from hyperbolic discounting with almost no extra assumptions-- only rough additiveness-- and predicts credible weapons for each side in the closely fought contests that seem to occur as people decide about self-control: Long range interests define principles, and short range interests find exceptions.

4. Willpower Brings Harmful Side-Effects.

Unfortunately, a person's perception of the prisoner's dilemma relationship-- and the willpower that results from this perception-- can't simply cure the problem of temporary preference. Willpower may be the best way we know to stabilize choice, but the intertemporal bargaining model predicts that it will also have serious side effects, side effects that have in fact been observed by clinicians. Such bargaining doesn't maximize our prospects as true exponential discounting would. Rather it formalizes internal conflict, making some self-control problems better, but some worse.

These side effects need to be discussed. Where they're noticed at all, they aren't recognized as the consequence of using willpower. In a dangerous split of awareness, we tend to see willpower as an unmixed blessing that bears no relation to such abnormal symptoms as loss of emotional immediacy, abandonment of control in particular areas of behavior, blindness toward one's own motives, or decreased responsiveness to subtle rewards. I will argue that just these four distortions are to be expected to a greater or lesser extent from a reliance on personal rules. They may even go so far as to make a given person's willpower a net liability to her.

1. Rules overshadow goods-in-themselves. The perception of a choice as a precedent often makes it much more important for its effect on future expectations than for the rewards that intrinsically depend on it. When this is true, your choices will become detached from their immediate outcomes and take on an aloof, legalistic quality. I've argued that this legalism underlies the self-control style that clinicians call compulsive (Ainslie, 1992, pp. 205-227). It's a polar opposite from impulsive temporary preferences, despite a common usage that equates them (e.g. "compulsive drinking").

It's often hard to guess how you'll interpret a current choice when looking back on it. Did eating that sandwich violate your diet or not? Where your rules' criteria are ambiguous, cooperation with your future selves will be both rigid and unstable. Unless you can find clear lines to use as boundaries it may be hard to tell whether, facing a choice in the future, you'll look back at your current choice and judge it to have been a lapse. Under the influence of an imminent reward you may claim an exception to a rule, but later think you fooled yourself, that is, see yourself as having had a lapse. Conversely, you may be cautious beyond what your long-range interest requires, for fear that you'll later see your choice as a lapse. This rationale exacerbates compulsiveness. Every lapse reduces your ability to follow a personal rule, and every observance reduces your ability not to. Errors in either direction impose costs that would never result from exponential curves, since those curves wouldn't make choice depend on recursive self-prediction in the first place.

2.Rules magnify lapses. When you violate a personal rule, the cost is a fall in your prospect of getting the long range rewards on which it was based. But this prospect is what you've been using to stake against the relevant impulses; a lapse suggests that your will is weak, a diagnosis that may act recursively to weaken your will.

To save your expectation of controlling yourself generally, you'll be strongly motivated to find a line that excludes from your larger rule the kind of choice where your will failed. This means attributing the lapse to a particular aspect of your present situation, even though it will make self-control much more difficult when that aspect is present in the future. You may decide that you can't resist the urge to panic when speaking in public, or to lose your temper at incompetent clerks, or to stop a doughnut binge once begun. Your discrimination of this special area has a perverse effect, since within it you see only failure predicting further failure. If you no longer have the prospect that your rule will hold here, these urges will seem to command obedience automatically, without an intervening moment of choice. Such an area, where a person doesn't dare attempt efforts of will, could be called a lapse district, by analogy to the vice districts in which Victorian cities encapsulated the vice they couldn't suppress. Where the encapsulated impulses are clinically significant, a lapse district gets called a symptom-- for instance, a phobia, a dyscontrol, or a substance dependence.

Thus the perception of repeated prisoner's dilemmas stabilizes not only long range plans but lapses as well (Discussed further in Ainslie, 1992, pp. 193-197). Alternative models of self-control failure based on exhaustion of "strength" (Baumeister and Heatherton, 1996) or an opponent process (Polivy, 1998), do not account for regular failure that is specific to a particular circumstance.

3. Rules motivate misperception. Personal rules depend heavily on perception-- noticing and remembering your choices, the circumstances in which you made them, and their similarity to the circumstances of other choices. And since personal rules organize great amounts of motivation, they naturally create temptations for you to suborn the perception process. When a lapse is occurring or has occurred, it will often be in both your long and short range interests not to recognize that fact: Your short range interest is to keep the lapse from being detected so as not to invite attempts to stop it. Your long range interest is also at least partially to keep the lapse from being detected, because acknowledging that a lapse has occurred would lower the expectation of self-control that you need to stake against future impulses.

After a lapse, the long range interest is in the awkward position of a country which has threatened to go to war in a particular circumstance that has then occurred. The country wants to avoid war without destroying the credibility of its threat, and may therefore look for ways to be seen as not having detected the circumstance. Your long range interest will suffer if you catch yourself ignoring a lapse, but perhaps not if you can arrange to ignore it without catching yourself. This arrangement, too, must go undetected, which means that a successful process of ignoring must be among the many mental expedients that arise by trial and error-- the ones you keep simply because they make you feel better without your realizing why. As a result, money disappears despite a strict budget, and people who "eat like a bird" mysteriously gain weight.

4. Rules may serve compulsions. The fact that a decision comes to be worth more as a precedent than it is in its own right doesn't necessarily imply that it's the wrong decision. On the contrary, you'd think from the logic of summing discount curves that judging choices in whole categories rather than by themselves would have to improve your overall rate of reward (figures 3A, 4B). Cooperation in a repetitive prisoner's dilemma would have to serve the players' long range interests, or else they'd abandon it. How, then, can self-enforcing rules for intertemporal cooperation ever become prisons? Why should anyone ever conclude that she was trapped by her rules, and even hire a psychotherapist to free her from a "punitive superego?"

The likeliest answer is that in everyday life a person can discern many possible prisoner's dilemmas in a given situation; and the way of grouping choices that finally inspires intertemporal cooperation need not be the most productive, because of two factors-- the selective effect of distinctness, and the properties of aggregated hyperbolic discount curves themselves:

Personal rules operate most effectively on distinct, countable goals. Thus the ease of comparing all financial transactions lets the value of a sum of money fluctuate much less over time than, say, the value of an angry outburst, or of a night's sleep. The motivational impact of a series of moods has to be much less than that of an equally long series of cash purchases.

The effect of having rewards marked by discrete stimuli can be seen in "melioration" experiments, where subjects successfully learn not to make bad choices when the consequence is less money per turn, but fail to learn this when the consequence is longer delays between turns, which cost just as much in a game of fixed duration (Herrnstein et.al., 1993). Amounts are eminently countable; delays are a matter of intuition unless someone specifically measures them. Accordingly, subjects achieve "rational" behavior toward amounts much more readily than to delays. By the same logic, when compulsions are based on well-marked contingencies, and their richer, longer range alternatives are harder to specify, the compulsions may win out because they offer clearer criteria for personal rules. The personal rules of anorectics or misers are too strict to promise the greatest satisfaction in the long run, but their exactness makes them more enforceable than subtler rules that depend on judgment calls.

FIGURE 6

How personal rules can serve mid range interests: Summed, discounted values of three alternative rewards (13:70:100), such that in a single choice (the last triplet), the first temporarily dominates the second; with bundling, e.g. at the beginning of a series of four choices, the first never dominates the second, but the second repeatedly dominates the third for a time. If the largest-latest rewards couldn't be summed, the dominance of the middle rewards would be even greater.

Furthermore, a look at series of discount curves themselves tells us that relatively long range temporary preferences can combine against interests of both shorter and longer duration. Figure 6 shows that hyperbolic discount curves can create three-way conflicts, and that summation of these curves can make a series of middle rewards dominant over not only the smallest, earliest rewards, but also over the largest, last ones. This dominance is temporary: From a distance the series of largest, last rewards is always preferable; but if the rewards are constituted so that there's often the opportunity for one of the middle ones to be chosen in the relatively near future, the middle series may dominate choice most or all of the time.

So cooperation among successive motivational states doesn't necessarily bring the most reward in the long run. The mechanics of policing this cooperation may produce the intrapsychic equivalent of regimentation, which will increase your efficiency at reward-getting in the categories you've defined, but reduce your sensitivity to less well-marked kinds of reward.

Both hyperbolic discounting and the personal rules that compensate for it have distorting effects. Therefore, there can be no hard and fast principle that people should follow to maximize their prospective reward. Thus "rationality" becomes an elusive concept. Insofar as it depends on personal rules demanding consistent valuation, rationality means being systematic, though only up to the point where the system seems to go too far and we look compulsive. Even short of frank compulsiveness, the systemization that lets rules recruit motivation most efficiently may undermine our longest range interests:

The attempt to optimize our prospects with personal rules confronts us with the paradox of definition-- that to define a concept is to alter it, in this case toward something more mechanical. If you conclude that you should maximize money you become a miser; if you rule that you should minimize your vulnerability to emotional influence, you'll develop the numbing insensitivity that clinicians have named alexithymia (Nemiah, 1977); if you conclude that you should minimize risk, you become obsessively careful; and so forth. The logic of rules may come to so overshadow your responsiveness to experience that your behavior becomes formal and inefficient. A miser is too rigid to optimize her chances in a competitive market, and even a daring financier undermines the productiveness of her capital if she rules that she must maximize each year's profit (Malekzadeh & Nahavandi, 1987). Similarly, strict autonomy means shielding yourself against exploitation by others' ability to invoke your passions; but alexithymics can't use the richest strategy available for maximizing emotional reward, the cultivation of human relationships (Ainslie, 1995). Likewise, avoidance of danger at any cost is poor risk management.

In this way people who depend on willpower for impulse control are in danger of being coerced by logic that doesn't serve what they themselves regard as their best interests. Concrete rules dominate subtle intuitions; and even though you have a sense that you'll regret having sold out to them, you face the immediate danger of succumbing to short range urges like addictions if you don't. If you haven't learned ways of categorizing long range rewards that permit them to dominate systematic series of mid-range rewards, your mid-range interests will make you compulsive.

5. Phenomena That Limit the Usefulness of the Will

Hyperbolic discounting suggests explanations for several other phenomena that are puzzling from the conventional utility viewpoint. However, these experiences are so pervasive that behavioral science has treated them as givens, as "just the way things are." Utility theory habitually sidesteps the questions that they raise, sometimes with the excuse that there's no accounting for tastes. However, they are amenable to analysis with a hyperbolic model. Ultimately they're what limit the scope of the will:

1. The limitation of emotion puzzle. The experiences that people value most are usually emotional ones. But it isn't hard to generate emotions voluntarily. People don't usually do this, because daydreamed emotions are less intense than the kind that have proper occasions, but that fact in itself lacks an explanation. How do emotions come to function like the limited goods of commerce, that you have to get from the outside world?

2. The construction of fact puzzle. Writers since Plato have noticed that people's beliefs about the world don't correspond entirely to the measurable facts, and sometimes barely resemble them at all. Yet we experience beliefs, like passions, as obligatory, as things that our observations force us to have. It has lately been fashionable to assert the contrary, that there's no objective basis for beliefs. Beliefs and fictions are said to share the status of "texts," strings of possible interpretations that people choose because of the rules of language games.

It has become clear that belief is at least partially a goal-directed activity, and thus classifiable as a behavior, rather than a passive consequence of incoming stimuli. However, propositions that are overtly goal-directed are experienced as different from beliefs; they're experienced as "make-believe." Insofar as belief is a behavior, what constraints make it different from make-believe?

3. The vicarious experience puzzle. Other people are especially valuable as sources of emotional experience. Conventional utility theory calls this a simple putting-yourself-in-the-other's-place, and regards it as natural whenever "social distance" is short. This idea, first elaborated by Adam Smith (1759/1976) has been put into terms of utility by Julian Simon (1995). But moving social experiences don't precisely depend on distance, or even on the existence of a real other person as opposed to a fictional character; and in many cases the experience that one person gets is obviously different from that of her vicarious object-- at the extreme, for the sadist and for the victim. How do other people move us, and what are the constraints on that process?

4. The indirection puzzle. Some goal-directed activities can't effectively approach their goals by direct routes. Trying to have fun usually spoils the fun, and trying to laugh inhibits laughter (Elster, 1981; Wegner, 1994). At first glance, this problem seems to strike at the heart of any motivational model, not just one that assumes exponential discounting. How can any goal-directed activity be undermined by striving toward its goal? How can a reward-dependent activity not be strengthened by reward?

I'll argue that these four puzzles are aspects of the same phenomenon: the fourth side effect of willpower just discussed. The greatest limitation of the will comes from the same process as its greatest strength: its relentless systemization of experience through attention to precedent, which braces it against temporary preferences but also makes it unable to follow subtle strategies to keep the reward mechanism productive.

Let's look at the four puzzles more closely:

1. The limitation of emotion puzzle.

Emotional rewards of one kind or another seem to be a large part of most people's incentives. We may decide to climb mountains, or become an object of envy, or achieve moral purity, or perform any number of other feats that aren't necessary for our physical comfort. We could ignore these tasks without any obvious penalty; but we somehow become committed to them, occasionally to the point of dying for them.

However, emotional reward is physically independent of any particular turnkey in the environment, an inconsistency with conventional utility theory. To function as a reward according to that theory, a good has to be limited in supply or accessibility; if it's available unconditionally it will never induce significant motivation to obtain it. As Adam Smith originally observed, this is just the reasoning that makes air have less market value than diamonds, although air is more necessary. To let emotional rewards be seen as economic goods, utility theory has had to assume that they are unmotivated reflexes that must be released by turnkey stimuli; but as any actor knows, emotions can be cultivated to appear, if not exactly on demand, at least as a reliable result of deliberate effort (Strasberg, 1987).

The most stirring emotions do seem to require a sense of necessity, so that we experience them not as choices but responses to an external provocation. Although emotions are physically available, something makes them less intense in proportion as the occasion for them is arbitrary. To the extent that someone learns to access them at will, doing so makes them pale, mere daydreams. Even an actor needs to focus on appropriate occasions to bring them out with force. But what properties does an event have to have in order to serve as an occasion for emotion? The fact that there's no physical barrier opposing free access to emotions raises the question of how emotional experiences come to behave like economic goods that are in limited supply. That is, how do you come to feel as if you have them passively, as implied by their synonym, "passions?"

Avoiding premature satiation. The basic question is, how does your own behavior become scarce? I'll divide it into two parts: Why would you want a behavior of yours to become scarce, that is, to limit your free access to it? And given that this is your wish, how can you make it scarce without making it physically unavailable?

A hyperbola-based hypothesis about the first part can provide a basis for resolving all four of the seeming paradoxes I've just described: All kinds of reward depend on a readiness for it that's used up as reward occurs and can't be deliberately renewed. I'll call this readiness appetite; although the term may also refer to an arousal for consuming the reward (as in "stimulating your appetite"), the arousal isn't possible without this underlying readiness. The properties of appetites are often such that rapid consumption brings an earlier peak of reward but reduces the total amount of reward that the appetite makes possible, so that we have an amount-vs.-delay problem of the kind that was described in figure 2. Where people-- or, presumably, any reward-governed organisms-- have free access to a reward that's more intense the faster it's consumed, they'll tend to consume it faster than they should if they were going to get the most reward over time from that appetite. In a conflict of consumption patterns between the long and pleasant versus the brief but even slightly more intense, an organism that discounts the future hyperbolically is primed to choose brief but intense.

This problem makes no sense in a world of exponential discounting. In an exponential world, an adept consumer should simply gauge what the most productive way to exploit an appetite will be, and pace her consumption accordingly. People could sit in armchairs and entertain themselves optimally by generating just enough emotional appetite and then satisfying it. By contrast, common experience teaches that emotional reward, indulged in ad lib, becomes unsatisfactory for that reason itself. To get the most out of any kind of reward, we have to have-- or develop-- limited access to it.

Limiting access should be easiest for physical rewards: You can make a personal rule to consume them only in the presence of adequately rare criteria; but with emotional rewards, the only way to stop your mind from rushing ahead is to avoid approaches that can be too well learned. Thus the most valuable occasions will be those that are either 1. uncertain to occur or 2. mysterious-- too complex or subtle to be fully anticipated. To get the most out of emotional reward, you have to either gamble on uncertainty or find routes that are certain but that won't become too slick. In short, your occasions have to stay surprising.

To restate this pivotal hypothesis: In the realm of emotional reward-- the great preponderance of the reward that well-off people pursue-- possible behaviors must compete on the basis of how well they can maintain your appetite. The processes that are rewarded by emotion compete for adoption on the basis of the extent to which their occasions defy willful control. Direct paths to reward become progressively less productive, because insofar as they become efficient they waste your readiness for reward. Conversely, if there's factor that delays consumption from the moment at which the consumption could, if immediate, compete with available alternatives-- the moment it reaches what could be called the market level of reward-- that factor may substantially increase the product of [value x duration] before the appetite satiates (figure 7).

FIGURE 7A

Cycles of growing reward potential (rising straight lines) and actual consumption (gray areas) leading to satiety. Consumption begins when total value of expected consumption reaches the competitive market level. Hyperbolic discount curves of the total value of each act of consumption decline with delay from its anticipated onset.

FIGURE 7B

Increased reward (stripes) resulting from increased appetite when there is an obligatory delay in consumption from the moment of choice ("{" brackets); the choice to consume occurs when the discounted value of the delayed consumption reaches the market level.

To repeat satisfactions that were once intense, you have to at least structure them as fantasies involving obstacles in order to achieve a modicum of suspense; but as a fantasy becomes familiar and your mind jumps ahead to the high points, the fantasy collapses further into being just a cursory thought-- an irritant if it retains any attractiveness at all, and a disregarded, empty option if it doesn't. Durable occasions for emotion have to be surprises, so that you don't have to restrain your attention from jumping ahead. Thus it's usually more rewarding to read a well-paced story than to improvise a fantasy. Accordingly, surprise is sometimes said to be the basis of aesthetic value (Berlyne, 1974; Scitovsky, 1976). And the mystical quality of existentialists' pronouncements can be seen as a way of recognizing the premature satiation problem: "The world is ambiguous... [this] is the reward for being human because it adds challenge, variety and opportunity to existence (Herzberg, 1965, p. 62)." Furthermore, "as long as man is an ambiguous creature he can never banish anxiety; what he can do instead is to use anxiety as an eternal spring for growth into new dimensions of thought and trust (Becker, 1973, p. 92)." In modalities where an organism can mentally reward itself, surprise is the only commodity that can be scarce.

Although there are wide variations in the equilibria people find between gratification at will and strict dependence on external occasions-- the fantasy-prone seem to have emotions that are more robust than other people's despite equally free access (Rhue & Lynn, 1987), while sociopaths can usually imagine very little-- everyone learns limits to her self-induction of emotions. Most people probably develop intuitions about how to foster sources of surprise, e.g. a rule not to read ahead, without ever making an explicit theory. People-- and presumably lower animals-- wind up experiencing as emotion only those patterns that have escaped the habituation of voluntary access, by a selective process analogous to that described by Robert Frank for the social recognition of "authentic" emotions (1988): Expressions that are known to be intentionally controllable are disregarded, as with the false smile of the hypocrite. By this process of selection emotion is left with its familiar guise as passion, something that has to come over you.

It's undoubtedly adaptive for vivid rewards to fade away into habit as you get efficient at obtaining them; this process may keep you motivated to explore your environment, both when you're young and inept and when you've become a master problem-solver. If internal reward were strictly proportional to how much of some external stimulus you could get, then a reward rate that was sufficient to shape your behavior when you were a beginner would lead you to rest on your laurels once you'd solved that particular problem. But instead, as you become increasingly skilled in an activity, the reward it generates increases only at first, and then decreases again because your appetite doesn't last as long.

For simplicity I've been talking mainly about "positive," or pleasurable emotions. Elsewhere I've argued that negative emotions aren't the opposite, but that their attraction differs from that of positive emotions in being transient and committing you to aversive experiences. Negative emotions are thus avoided by your longer latency response processes, the ones that you experience as intentional or deliberate. A full explanation of this hypothesis about negative emotions requires an analysis of the short-term rewarding elements in aversive stimuli, to which I refer the reader (Ainslie, 1992 pp. 101-114, 1999b, 2001, pp. 51-61, 67-69, 173-174).

2. The construction of fact puzzle.

It's now common knowledge that people's beliefs are heavily influenced by their own tacit choices. Decisions about attending to or ignoring information shape perception so much that some "social constructivists" have put fact and fiction on a par, under the name "text" (e.g. Gergen, 1985; see Harland, 1987).

To a great extent belief does seem to be a goal-seeking activity. However, it can't be based simply on rewardingness and still be experienced as belief. Belief differs from make-believe in depending on the ruling of some external arbiter, some test that's beyond your direct influence, rather than simply being chosen.

Beliefs could be viewed as dispositions to choose one particular text over its alternatives. Often these dispositions are shaped by environmental contingencies-- My belief that a dropped object will fall rather than rise comes easily and consistently; any urge to reexamine it has long since extinguished. Such instrumental beliefs-- those that are differentially rewarded by their practical effects-- have little room for social construction. But for other beliefs there's no such shaping, or it's delayed, leaving the present motivational impact of the belief to depend on the way it occasions emotion. Such beliefs may concern the past or future (was there a conspiracy to assassinate JFK? will my pension be big enough?), or refer to facts that can't be discovered with certainty (does my spouse really love me?), or are meant to be assertions rather than descriptions (is abortion murder?). Instrumental beliefs may also occasion emotions in major ways, of course (are my brakes failing?), but in the instrumental realm self-deception is identifiable and significantly punished by experience.

Where the consequences of beliefs are emotional rather than instrumental, the constraints on these beliefs have not been explored. However, the pervasive urge for premature satiation discussed in the foregoing section is a likely limiting factor. The existence of this urge implies a selective process favoring emotions that are occasioned by adequately inaccessible texts, thereby promoting those texts to a status more significant than fantasy. There must be an incentive to cue emotions by facts in order to optimize available appetite. Emotions tied to beliefs that can shift as convenience dictates will become daydreams, just like emotions that aren't tied to beliefs at all. The texts that get selected as noninstrumental beliefs will be those interpretations of reality that serve as effective occasions for emotions. Those texts should have the feel of facts, and the recognition of their significance should have the feel of belief.

According to this hypothesis, accuracy per se is only one selective factor for belief in a fact, and not an indispensable one at that. The property that makes an interpretation a belief is that it selects decisively among the available texts. To do so reliably it must not only be unique, but occasion adequate emotion. Beliefs that foster suspense will reward us more than beliefs that merely have internal consistency or match observable stimulus patterns without regard to what appetite is available for them, and will thus divert emotional investment from those beliefs.

People probably discover the value of hard-to-get occasions by trial and error. It's more rewarding to solve puzzles, conquer mountains, discover historical data, or gamble on other people's behavior than just to daydream. The payoffs for these activities depend on the pace at which they invite emotion, and that pace depends in turn on my beliefs about them: Does my interpretation make this puzzle, or mountain, or historical problem, or human relationship, tough enough without being impossible? My interpretation may be constrained by instrumental needs, of course, but it will also be shaped by how inaccessible it makes reward.

The events that best occasion emotional reward are:

a. outside of your control. Even fiction, if written by someone else, has more emotional impact than your own daydream with the same content.

b. rare. A situation in a sports event that comes up only once in a decade is more moving than one that happens about every week, which is in turn more moving than one that happens several times in a game. This factor makes cues better pacers in proportion as they are:

b(1). current rather than historic-- since current events of a given kind are rarer than those that may have happened in the past generally. There may have been only one major earthquake this year, but six if you scan the past decade, and scores before that.

b(2). germane, for instance affecting nearer relatives rather than more distant ones. You can have only one spouse and two parents, but any number of cousins.

b(3). verifiably true. Events seen as facts have more impact than those seen as fictions. This, too, is because of their relative rarity; and veridical truth isn't necessary, as long as your belief stands up to some well-disciplined test for truth. An entrenched historical myth or urban legend may serve as well as a fact.

b(4). Consistent. Potential beliefs that are consistent with beliefs already held are fewer than those that could be formed ad hoc from current information.

c. surprising, as described in the precious section. Rarity, which implies externality, is necessary but not sufficient to maintain occasions for emotion.

However, we should expect spontaneous learning based on these properties to find only moderately restrictive occasions for reward. People will naturally try to interpret events as calling for reward sooner rather than later. It's in my longest range interest to restrict my own freedom of interpretation, so as to commit myself to tolerate delay and suspense. Without rationales for keeping the facts I believe in unique, these texts would be too fluid, too obedient to my wishes. This consideration often makes the beliefs shaped by instrumental needs the best occasioners of emotion. The insight that hyperbolic discounting theory supplies is that beliefs may often do their pacing job without being accurate; the role of accuracy for noninstrumental beliefs is only to supply uniqueness.

I'm hypothesizing that noninstrumental beliefs function in effect as personal rules to have emotions according to particular kinds of occasions. An awareness of taking arbitrary control of such beliefs comes to predict loss of emotional force. The choice involved is a lot like that of how much importance to invest in a movie. You're usually not conscious of this choice, but it becomes evident when a movie becomes too punishing and you withdraw importance from it. You say to yourself, "This is only a movie," and yet you aren't discovering new information-- You've never ducked when guns pointed to the audience. Rather you're announcing your disinvestment, an end to your "suspension of disbelief": "This movie shall no longer be important to me. I won't reward myself any longer according to its plot."

Where there's no instrumental need, the penalty for badly placed beliefs is not some practical failure, but a failure to occasion emotional reward as effectively as possible. The person who withdraws her investment during the scary part of the movie loses her chance to be rewarded by the parts that follow; to some extent she will lose her ability to keep her investment in subsequent movies when they tempt her to disinvest.

In this way texts that are adequately rare and surprising become facts or close relatives of facts, goods in limited supply that reward us in ways not necessarily connected with their instrumental value.

3. The vicarious experience puzzle.

Especially puzzling for utility theory is the way we take on other people's experiences, both pleasant and unpleasant, as our own. There has been a lively debate between authors who believe that altruism is a primary motive (e.g. Batson & Shaw, 1991) and those who think it reduces to selfish pleasure (Piliavin et.al., 1982; Sen 1977). Economic man is supposed to maximize his own prospects, and help others only insofar as doing so will accomplish this. However, you find counterexamples all the time, from transients who leave tips for waiters they'll never see again to heroes who give their lives to save strangers in fires and accidents. People also have the potential to derive satisfaction from others' pain-- even, in the extreme, from their death throes (e.g. Davies, 1981, pp. 78-82). Instrumentality again aside, what makes this range of perceived experiences in other people valuable to us?

The premature satiation hypothesis predicts that vicarious experience ought to be a good criterion for occasioning emotional reward, but should become less valuable to the extent that you can bring it under your control, because your control will inevitably undermine your appetite. The greatest rewards from other people come through gambles. Gambles that are rigged--interactions that are predictable, people you can boss around, relationships you're poised to leave if they turn disappointing-- push your emotional experiences in the direction of daydreams. These hedges are tantamount to exchanging a mutual game of cards for a game of solitaire, and perhaps even to cheating at solitaire; such an impulse is punished by a loss of suspense, and hence of all but the shortest range reward.

Given adequate challenge, the emotional payoff comes when the other person gives you a rewarding occasion. Predicting other people becomes a highly rewarded activity for its emotion-occasioning value, quite aside from how it may help you influence them.

Empathy as modeling. However, this is only part of the story. So far, there's no reason to think that gambling on other people's behavior would be any more rewarding than gambling on a horse race, or on your ability to solve a puzzle. The fact that this puzzle responds strategically to your choices might make it more challenging, but wouldn't qualitatively change the experience of succeeding or failing. But because this kind of puzzle is built like the person solving it-- that is, because it's another person-- it may foster what is likely to be a much richer strategy for occasioning emotions:

First of all, this similarity supplies a different way of solving the puzzle. Since other people's choices depend more on their interaction with you than on anything you know about them in advance, you soon learn that the best way to predict them is to use your own experience to model theirs. You say, "if I were her, and were angry at me but amused by me and hoped for a job with me, how would I feel if I (the real me) were to say X?" You entertain the other's likely emotions and notice where they pull you. In effect you create a model of the other person using your own emotional equipment. It's a familiar experience to hold conversations with such a model-- "If I say X I can just hear her say Y..." If the model isn't arbitrary-- if it's disciplined by observation-- it's apt to behave much more like the actual other person than a non-empathic model would, for instance one made like the model of an economy from statistical data. Even infants can be shown to predict a person's behavior by a "theory of mind," that is, by empathic modeling, which in some situations makes them much better predictors of an adult human's behavior than are adult apes using trial-and-error (Povinelli, 1999). The best way to predict people is to put yourself in their shoes.

However, this empathic modeling process yields more than just prediction. Putting yourself in the other person's shoes means adopting the criteria that you think she's using to occasion emotion. For the time being you entertain what you think would be her emotions. But of course, they are hers only in the sense that you're having them according to a theory about her. You are the person through whose brain they're percolating. This means that you can use such a model to occasion emotions just as you use your own prospects.

Since emotions don't need a turnkey, just appetite and adequately rare occasions to preserve this appetite, you can sometimes experience the emotions you're modeling in the other person as substantially as the ones you have as yourself. To model the other people is to have their expected feelings; and nothing makes these "vicarious" feelings differ in kind from "real" ones. However, the impact of this phenomenon will be limited by the uniqueness of your relationship with the other person, just as the impact of texts a la constructivism is limited by their factuality; your vicarious experiences from strangers picked for the purpose will be little more than daydreams.

To the extent that we've gambled on another person's discernable feelings, these feelings become a commodity that we'll work for. Information for refreshing our models of other people becomes the limited good that constrains the otherwise too-available resource of emotion. This, I argue, is how other people come to compete for our interest on the same footing as the goods of commerce.

4. The indirection puzzle.

I've described how the will can’t stop the premature satiation of suspense. I'll now argue that will may actually make premature satiation worse. The will needs conspicuous, discrete criteria of success or failure to maintain the incentive to cooperate with future selves at each choice-point. If the criteria are subtle-- "eat what you'll be glad of in retrospect","follow your true heart", or even the self-referential "do as much as you have to in order to maintain intertemporal trust"-- then there will be too much room for impulses to dominate individual choices without getting caught. Remember our hypothesis of what the will needs in order to recruit motivation: You must expect to obey it in the future if and only if you obey it currently. Subtle criteria make this contingency escapable and thus take away its power. This is the very reasoning that leads the will often to serve compulsions because of their better definability, as I argued earlier.

But systematically following well-defined criteria is exactly what makes your behavior predictable, by other people as well as yourself. It's a great way to achieve a goal as efficiently as possible, so that you can go on and do something else. It's a terrible way to enjoy an activity for its own sake, because it kills appetite. You inevitably learn to anticipate every step of the activity, so that it eventually becomes "second nature," making it so uninteresting that people used to think that ingrained habits were run by the spinal cord. You can't use will to prevent this anticipation, because clear criteria for rules directing attention aren't available, and even if they were, attention probably moves too quickly to be made contingent on testing for them.

So a too-powerful will tends to undermine its own motivational basis, creating a growing incentive to find evasions. The awkwardness of getting reward in a well-off society is that the creation of appetite often requires undoing the work of satisfying appetite. Physiological appetites like those for food and sex regenerate as a function of time, but appetites for safety, or wealth, or comfort don't. If you're comfortable, you have it made, and staying comfortable may not take much effort. Once your emotional appetite for comfort has been satiated, it won't be a source of much further reward unless something restores your need for it. You have to face a challenge, some kind of doubt or delay.

However, you can't simply try to get satisfaction slowly, or inefficiently-- always to be reining in your own impatience would take more effort than it repaid. Besides, the circumstance that made overcoming the challenge more than a daydream-like activity to begin with must have involved seeing it as a need, as something more than a mere game; deliberately going slowly contradicts that belief.

Since willpower is based on systemization, it's the self-control tactic that least accommodates the complementary needs for satisfaction and deprivation: The more you make satisfaction depend on a condition outside of your control-- the more you believe in the objective value of wealth, safety, comfort, etc.-- the more you'll perceive your need to refresh your appetite for it as impulsive. That is, belief in the desirability of these conditions themselves commits you to guard them from any tendency you have to put them at risk. Thus the will usually comes to superintend the getting and maintaining of the objects of appetite; the creating of appetites themselves has to come from some evasion of will.

Where we haven't committed ourselves to maximizing the goods that satisfy an appetite, we can recognize available appetite as a resource: People not uncommonly work up an appetite for dinner, boast of an appetite for sex, complain of a jaded appetite for entertainment, and so on. However, where appetites have to be restored by repudiating hard-won accomplishments, such recognition would often strike us as irrational. There's clearly an appetite for hoarding, for instance-- for collecting things, or getting rich-- but to whet the appetite by gambling and intermittently losing your hoard isn't seen as maximizing your utility. Gamblers are called foolish because we see value as inhering in their goods, not in how well they exploit their appetite.

To deal with this contradiction, you usually have to believe in some larger quest that requires you to put your satisfaction at risk. To climb mountains or jump out of airplanes as a test of manhood, to stay with an abusive lover to prove your loyalty, to join a religion that demands self-abasement, to play the stock market or the horses as a way to get rich, even to bet your dignity on staying in the forefront of fashion, leads to repeated losses or at least the credible threat of losses. You get your appetite back while struggling not to.

According to this hypothesis, the old psychoanalytic cliché, that gamblers unconsciously want to lose is partially correct: They're moved by a need to restore their appetites by losing, but must serve this need without contradicting their wholehearted attempts to win. This is just the intrapsychic version of a familiar phenomenon: If one sports team wins too regularly it diminishes the excitement of the game, and eventually attendance and income. It's in the winning team's interest as well as its competitors' to have a governing body impose some kind of handicap, like last pick in next season's player draft. What the team can't do, without both impairing its morale and removing the spark from the game, is try less hard to win.

Refreshing your emotional appetite without having to contradict what you've willed often requires believing in some seemingly rational, or arguably necessary, activity that makes the direct routes to a reward less of a sure thing. That is, you need to find indirect routes to success: dummy activities that aren't actually maintained by their ostensible purpose, but stay desirable insofar as they maintain appetite by creating good gambles, or by directing your efforts toward tasks whose mastery won't give you arbitrary control over their reward. There are also other incentives for indirection besides avoidance of premature satiation, like avoiding occasions for self-consciousness, competitiveness, panic, and performance anxiety; but these don't arise because of efficiency at reward-getting per se, and thus aren't consequences of will.

The strategy of indirection goes a step beyond the harnessing of emotions to external facts; it finds beliefs that specifically divert effort away from the efficient satisfaction of the relevant appetites. Indirect tasks are a form of self-deception and thus vulnerable to exposure, but are maintained by some relatively long range reward that is spoiled by short cuts. The necessary blind spots may seem preposterous to people who don't share that particular indirection strategy; a vulnerability to satire is frequently a sign that an activity is indirect.

Activities that are spoiled by counting them, or counting on them, have to be undertaken through indirection if they are to stay valuable. For instance, romance undertaken for sex or even "to be loved" is thought of as crass, as are some of the most lucrative professions if undertaken for money, or performance art if done for effect. Too great an awareness of the motivational contingencies for sex, affection, money, or applause spoils the effort, and not only because it undeceives the other people involved. Beliefs about the intrinsic worth of these activities are valued beyond whatever accuracy they might have, because they promote the needed indirection. Similarly, the specific tasks that various schools of psychotherapy believe in turn out to be unnecessary when the effectiveness of seemingly contradictory teachings is compared; perhaps the empathic engagement that has been shown to make the difference is awkward unless attention is directed away from it.

A flaw of the indirection strategy is that you can't acknowledge the possibility of following it too far. Indirections are supposed to be goods in their own right, and awareness of an optimal point that you shouldn't pass has to be tacit. Failure of this delicate perception may be responsible for the literal-mindedness and consequent boorishness of nerds-- the kind that leads them to answer inquiries about their health in full detail, and give true opinions of others' wearing apparel. Perhaps from a lack of talent in evaluating emotional occasions, they cling to the limited purpose of smalltalk in pacing the exchange of emotional occasions. On the other hand, it's risky to acknowledge the underlying source of reward. Someone who plays bridge "for fun" may look on someone who plays "to win" as playing compulsively; but if the former player directly tries to maximize the fun, she may spoil it by a laxity that makes the game trivial.

People have always had trouble conceiving a pressure to behave which is at the same time a resource. A bottle of wine that "demands to be drunk" or money that "burns a hole in your pocket" are intuitive enough, but an appetite per se for drink or spending isn't usually seen as a good. Plato saw something of this problem in the emotion of love. In his Symposium Socrates says that Love is the child of Plenty (Poros) and Poverty (Penia), and has some of the qualities of each-- Love doesn't possess plenty, beauty, wisdom, etc., for then he would be satisfied and wouldn't be Love; but, unlike Poverty, he appreciates these things. "That which is always flowing in is always flowing out, and so he is never in want and never in wealth," but represents a "mean" between the two (Plato, 1892, p. 329;Symposium 203-204). Many writers since then have said that love is a blessing but desire a curse; they haven't been able to specify what makes the difference.

Even in intellectual endeavors, society can't seem to find a stable balance between creating and overcoming challenges. There is a pendulum of favor that swings between "getting to the bottom of things" in one direction and mystical holisms, rejections of crass reductionism in the other. When it swings in the former direction it fosters science, organization, a love of clear, explicit thought, simple form, and comprehensive theory, in short, classicism. When it swings the other way, it motivates subtlety, an ambivalence toward science, and a love of the intense and incomprehensible emotions evoked when the world is mysterious-- that is, romanticism. The ultimate breakdown of the will seems to occur when efficiency has pushed the pendulum further in the direction of systemization-- and the attenuation of appetite-- than people can stand. The will is no good at pushing the other way. It can only accede to being bypassed by indirections that have more vigor, until the pendulum has again swung well into unsystematic territory.

The foregoing predictions illustrate the strategic patterns that hyperbolic discounting may produce. This kind of analysis, which I've called picoeconomics (micromicroeconomics; Ainslie, 1992) can't be verified directly by controlled experiment because it deals with recursive phenomena. Rather it requires pattern fitting, which will always depend to some extent on what strikes individual observers as parsimonious. Patterns different from the ones I've proposed may eventually be seen to fit experience better; but I would argue that some such intertemporal bargaining model will be necessary to accommodate the robust empirical finding of hyperbolic discounting.

EFERENCES

Ainslie, George (1992) Picoeconomics: The Strategic Interaction of Successive Motivational States within the Person. Cambridge U.

Ainslie, George (1995) A utility-maximizing mechanism for vicarious reward: Comments on Julian Simons "Interpersonal allocation continuous with intertemporal allocation" Rationality and Society 7, 393-403.

Ainslie, George (1999) The intuitive explanation of passionate mistakes, and why its not adequate. In Jon Elster, ed., Addiction. Sage, 209-238.

Ainslie, George (in press) Breakdown of Will. Cambridge U.

Ainslie, George and Haendel, V. (1983) The motives of the will In E. Gottheil, K. Druley, T. Skodola, H. Waxman (eds.), Etiology Aspects of Alcohol and Drug Abuse. Charles C. Thomas, pp. 119-140.

Batson, C. Daniel and Shaw, Laura L. (1991) Evidence for altruism: Toward a pluralism or prosocial motives. Psychological Inquiry 2, 159-168.

Baumeister, R.F. (1984) Choking under pressure: Self-consciousness and paradoxical effects of incentives on skillful performance. Journal of Personality and Social Psychology 46, 610-620.

Baumeister, Roy F. and Heatherton, Todd (1996) Self-regulation failure: An overview. Psychological Inquiry 7, 1-15.

Becker, G. and Murphy, K. (1988) A theory of rational addiction. Journal of Political Economy 96, 675-700.

Berger, Arthur A. (1987) Humor: An introduction. Special issue: Humor, the psyche, and society. American Behavioral Scientist 30, 6-15.

Berlyne, D.E. (1974) Studies in the New Experimental Aesthetics. Hemisphere.

Bratman, Michael E. (1999) Faces of Intention: Selected Essays on Intention and Agency. Cambridge U.

Carlson, A.J. (1916) The relation of hunger to appetite. The Control of Hunger in Health and Disease. University of Chicago Press.

Case, David (1997) Why the delay-of-reinforcement gradient is hyperbolic. Paper presented at the 20th Annual Conference of the Society for the quantitative Analyses of Behavior. Chicago, 5/22.

Davies, Nigel (1981) Human Sacrifice in History and Today. Morrow.

Douglas, M. (1966) Purity and Danger: An Analysis of Concepts of Pollution and Taboo. Routledge and Kegan Paul.

Elster, J. (1981) States that are essentially by-products. Social Science Information 20, 431-73. Reprinted in Elster, J. (1983) Sour Grapes: Studies in the Subversion of Rationality. Cambridge U., pp. 43-108.

Empson, W. (1930) Seven Types of Ambiguity. New Directions.

Erdelyi, Matthew Hugh (1990) Repression, reconstruction and defense: History and integration of the psychoanalytic and experimental frameworks in Singer, Jerome L., ed., Repression and Dissociation: Implications for Personality Theory, Psychopathology, and Health. U. Chicago.

Frank, Robert H. (1988) Passions Within Reason. W.W. Norton and Company.

Galen (1963) (Harkins, P.W., trans.) Galen on the Passions and Errors of the Soul. Ohio State University Press.

Gergen, K. J. (1985) The social constructionist movement in

modern psychology. American Psychologist 40, 266-275.

Gibbon, John. (1977) Scalar expectancy theory and Weber's law in animal timing. Psychological Review 84, 279-325.

Glantz, K. and Pearce, J. (1989) Exiles from Eden: Psychotherapy from an Evolutionary Perspective. Norton.

Grace, Randolph C. (1994) A contextual model of concurrent chains choice. Journal of the Experimental Analysis of Behavior 61, 113-129.

Green, Leonard, and Myerson, Joel (1993) Alternative frameworks for the analysis of self-control. Behavior and Philosophy, 21, 37-47.

Green, Leonard, Fry, Astrid, and Myerson, Joel (1994) Discounting of delayed rewards: A life-span comparison. Psychological Science 5, 33-36.

Green, Leonard, and Myerson, Joel (1993) Alternative frameworks for the analysis of self-control. Behavior and Philosophy, 21, 37-47.

Harland, Richard (1987) Superstructuralism: The Philosophy of Structuralism and Post-Structuralism. Methuen.

Harris, Christopher and Laibson, David (1999) Dynamic choices of hyperbolic consumers. Presented at the GREMAQ/CEPR Conference on Economics and Psychology, Toulouse, France, June 19.

Harvey, Charles M. (1994) The reasonableness of non-constant discounting Journal of Public Economics 31-51.

Herrnstein, R.J., Loewenstein, G., Prelec, D., and Vaughan, W.,Jr. (1993) Utility maximization and melioration: Internalities in individual choice. Journal of Behavioral Decision Making 6, 149-185.

Herzberg, F. (1965) Work and the Nature of Man. The World Publishing Company.

Heyman, Gene M. (1996) Resolving the contradictions of addiction. Behavioral and Brain Sciences 19, 561-610.

Hollerman, Jeffrey R., Tremblay, Leon, and Schultz, Wolfram (1998) Influence of reward expectation on behavior-related neuronal activity in primate striatum. Journal of Neurophysiology 80, 947-963.

James, W. (1890) Principles of Psychology. Holt.

Kant, I. (1793/1960) Religion Within the Limits of Reason Alone (T. Green and H. Hucken, trans.). Harper and Row, pp. 15-49.

Kirby, Kris N. and Herrnstein, Richard J. (1995) Preference reversals due to myopic discounting of delayed reward. Psychological Science 6, 83-89.

Klein, B. and Leffler, K.B. (1981) The role of market forces in assuring contractual performance. Journal of Political Economy 89, 615-640.

Kohlberg, L. (1963) The development of childrens’ orientations toward a moral order: I. sequence in the development of moral thought. Vita Humana 6, 11-33.

Laibson, David (1997) Golden eggs and hyperbolic discounting. Quarterly Journal of Economics, 62, 443-479.

Loewenstein, George (1996) Out of control: Visceral influences on behavior. Organizational Behavior and Human Decision Processes 35, 272-292.

Logue, A. W. (1988) Research on self-control: An integrating framework. Behavioral and Brain Sciences 11, 665-709.

Macaulay, S. (1963) Non-contractual relations in business: A preliminary study. American Sociological Review 28, 55-67.

McDiarmid, C.G. and Rilling, M.E. (1965) Reinforcement delay and reinforcement rate as determinants of schedule preference. Psychonomic Science 2, 195-196.

Malekzadeh, Ali, R. and Nahavandi, Afsaneh (1987) Merger mania: Who wins? Who loses? Journal of Business Strategy8, 76-79.

Martin, J. (1991) A shortage of shock sparks plea for more. The Philadelphia Inquirer, Feb. 3, p. 3-I.

Mazur, James E. (1984) Tests of an equivalence rule for fixed and variable reinforcer delays. Journal of Experimental Psychology: Animal Behavior Processes 10, 426-436.

Mazur, J.E. (1986) Choice between single and multiple delayed reinforcers. Journal of the Experimental Analysis of Behavior 46, 67-77.

Mazur, James E. (1997) Choice, delay, probability, and conditioned reinforcement. Animal Learning and Behavior25, 131-147.

Millar, A. and Navarick, D.J. (1984) Self-control and choice in humans: effects of video game playing as a positive reinforcer. Learning and Motivation 15, 203-218.

Myerson, Joel, and Green, Leonard (1995) Discounting of delayed rewards: Models of individual choice. Journal of the Experimental Analysis of Behavior 64, 263-276.

Nemiah, John. C. (1977) Alexithymia: Theoretical considerations Psychotherapy and Psychosomatics, 28, 199-206.

O'Brien, Charles P., Ehrman, R.N., and Ternes, J.W. (1986) Classical conditioning in human dependence. In S.R. Goldberg and I.P. Stolerman, Eds., Behavioral Analyses of Drug Dependence. Orlando, FL: Academic, pp. 329-356.

Piaget, J. (1932/1965) The Moral Judgment of the Child. Free Press.

Piliavin, Jane A., Callero, Peter L., and Evans, Dorcas E. (1982) Addiction to altruism? Opponent-process theory and habitual blood donation. Journal of Personality and Social Psychology 43, 1200-1213.

Plato (1892) The Dialogues of Plato, B. Jowett, Trans. Random House, v. 1.

Polivy, Janet (1998) The effects of behavioral inhibition: Integrating internal cues, cognition, behavior, and affect. Psychological Inquiry 9, 181-204.

Povinelli, Daniel J. (1999) Paper presented at the annual meeting of the Association for the Study of Animal Behavior, London.

Putnam, F.W. (1989) Diagnosis and Treatment of Multiple Personality Disorder. Guilford.

Rachlin, Howard (1995) Self-control: Beyond commitment Behavioral and Brain Sciences 18, 109-159.

Ragotzy, S.P., Blakely, E., and Poling, A. (1988) Self-control in mentally retarded adolescents: Choice as a function of amount and delay of reinforcement. Journal of the Experimental Analysis of Behavior 49, 191-199.

Rhue, Judith W. and Lynn, Steven Jay (1987) Fantasy proneness: The ability to hallucinate "as real as real" British Journal of Experimental and Clinical Hypnosis 4, 173-180. Russell, J.Michael (1978) Saying, feeling, and self-deception. Behaviorism 6, 27-43.

Schultz, Wolfram, Dayan, Peter, and Montague, P. Read (1997) A neural substrate of prediction and reward. Science 275, 1593-1599.

Schultz, Wolfram, Tremblay, Leon, and Hollerman, Jeffrey R. (1998) Reward prediction in primate basal ganglia and frontal cortex. Neuropharmacology 37, 421-429.

Scitovsky, T. (1976) The Joyless Economy: An Inquiry into Human Satisfaction and Consumer Dissatisfaction. Oxford.

Sen, Amartya K. (1977) Rational fools: A critique of the behavioral foundations of economic theory Philosophy and Public Affairs 6, 317-344.

Simon, Julian L. (1995) Interpersonal allocation continuous with intertemporal allocation: Binding commitments, pledges, and bequests. Rationality and Society 7, 367-430.

Skog, Ole-Jorgen (1999) Rationality, irrationality, and addiction in Jon Elster and Ole-Jorgen Skog, eds. Getting Hooked: Rationality and Addiction. Cambridge U.

Smith, Adam (1759/1976) The Theory of Moral Sentiments. Oxford U.

Solnick, J., Kannenberg, C., Eckerman, D. and Waller, M. (1980) An experimental analysis of impulsivity and impulse control in humans. Learning and Motivation 2, 61-77. Review, 217-225.

Strasberg, Lee (1987) A Dream of Passion: The Development ofthe Method. Plume.

Strotz, R.H. (1956) Myopia and inconsistency in dynamic utility maximization. Review of Economic Studies 23, 166-180.

Sully, J. (1884) Outlines of psychology. Appleton.

Thaler, R. (1991) Quasi Rational Economics. Russell Sage.

Thomas, Elizabeth Marshall (1989) The Harmless People. Vintage.

Tomkins, S. S. (1978) Script theory: Differential magnification of affects. Nebraska Symposium on Motivation 26, 201-236.

Wegner, Daniel M. (1994) Ironic processes of mental control. Psychological Review 101, 34-52.

NOTES: