NOTE-443: Précis of Breakdown of Will

Contribution

The précis compresses the twelve chapters of Breakdown of Will (LIT-583) into one argument. It starts from a single empirical premise, that vertebrates discount delayed rewards hyperbolically, and derives from it three things usually treated as primitives: a divided self, a will, and the will’s characteristic failures. It also proposes to replace classical conditioning with reward as the selector of “involuntary” processes such as pain, emotion and appetite.

Key insight

If value is roughly divided by delay, a smaller reward that is imminent can outweigh a larger later one, and the same person will prefer each at a different time. Those time-bound preferences grow into “interests” that act strategically against each other. A person who sees that today’s choice predicts tomorrow’s can make each choice stand for a whole series of choices, and summed hyperbolic curves favour the larger, later series. So willpower is a self-enforcing contract among successive selves, made of self-prediction, and it is as fragile as any such contract: one defection lowers the credibility on which the next cooperation depends. “Will, in short, is a bargaining situation, where credibility is power” (§8.1).

Assumptions

• Hyperbolic discounting is basic (§3). Value is inversely proportional to delay plus a small constant (Mazur 1987, one free “impatience” parameter). Fits with an exponent on the denominator are reported, but the exponent is “usually close to 1.0”.
• Rough additivity. The discounted effects of each reward in a series “simply add” (Mazur 1997). The précis says the bargaining model needs “almost no extra assumptions— only rough additiveness” (§6.1).
• Self-awareness. Bundling by perception requires “a degree of self-awareness probably unavailable to nonhumans” (§5). Animals show commitment devices, but not, on this account, will.
• A functional definition of reward: “that which increases the likelihood that the processes it follows will recur”, stripped of its connotation of pleasure (§4.1). The account of pain and emotion depends on this.
• A common currency. Note 1 cites Shizgal and Conover: “the utility of all competing resources must be represented on a single, common dimension”.

Key results

The précis is a theoretical synthesis. Its results are arguments, with cited experiments as support.

• Preference reversal (§3). Hyperbolic curves predict a “strong but temporary” preference for the smaller, sooner (SS) reward just before it is due. The précis cites observations in animals and in people choosing between juice, video-game access, relief from noise and money.
• Commitment in animals (§5). Pigeons choose in advance an option that removes the SS alternative (Ainslie 1974). Rats press a bar committing them to 0.5 s of shock 40 s later instead of 5 s of shock 45 s later, rather than leave the choice open and then, almost always, take the imminent 0.5 s (Deluty et al. 1983).
• Four tactics of self-control (§5): external constraints, attention control, emotion control, and will. Only will is “at once the strongest and most versatile”.
• Bundling (§§5–6). Choosing for a series raises the relative value of larger, later (LL) rewards under hyperbolic but not exponential curves (Figures 3A–3B). Students chose LL money more often when choosing for five weeks at once (Kirby & Guastello 2001). Rats chose LL more when choosing for a block of three trials (Ainslie & Monterosso 2003). Bundling makes summed curves look “more exponential” (Figure 5).
• Exceptions and bright lines (§6.1). Short-range interests find exceptions and long-range interests define principles. A rule of no alcohol is more stable than two drinks a day because “the line between some drinking and no drinking is unique (bright)”.
• Five models of will compared (§8): null (Ryle; Becker & Murphy), organ (Baumeister & Heatherton), resolute choice (Bratman; McClennen), pattern-seeking (Rachlin) and intertemporal bargaining. Bargaining is “the only model that explains both temporary preference and adequate incentive to overcome it from the properties of the rewards involved”.
• Kavka’s toxin (§8.1). Intending is classifying an act as a precedent, staking the value of the series on it. The subject can sincerely intend to drink the toxin because reneging injures the credibility of her later intentions. “Only intertemporal bargaining makes it affirmatively rational.”
• Side effects of will (§9): rules overshadow goods-in-themselves (§9.1); rules magnify lapses into “lapse districts” (§9.2); rules motivate misperception of lapses (§9.3); rules may serve compulsions, since “well-marked criteria may win out simply because they offer more stability” (§9.4). Hence “rationality is elusive” (§9.5).
• Premature satiation (§§10–11). An organism that discounts hyperbolically consumes a self-generated reward too fast. Emotional reward stays strong only when its occasions are uncertain or “mysterious”. This is offered as the reason beliefs feel like facts, why other people move us, and why some goals can only be reached indirectly.

Claims

id	claim	strength	support
C1	Humans and other vertebrates discount delayed rewards hyperbolically, not exponentially	moderate–strong as reported: many cited fits and reversal studies, none re-analysed here	§3
C2	Hyperbolic discounting alone predicts temporary preference for smaller, sooner rewards	strong (follows from the curve shape)	§3, Figure 1B
C3	Bundling a series of choices raises the relative value of the larger, later rewards, under hyperbolic but not exponential discounting	strong as mathematics; two cited experiments	§§5–6, Figures 3–4; Kirby & Guastello 2001; Ainslie & Monterosso 2003
C4	Willpower is the perception of choices as precedents in an intertemporal repeated prisoner’s dilemma	argued; the précis concedes it is “hard to study directly by controlled experiment”	§§6, 8
C5	The bargaining model, unlike null, organ and resolute-choice models, makes drinking Kavka’s toxin rational	argued from a thought experiment	§8.1
C6	Reliance on personal rules predicts legalism, lapse districts, motivated misperception and over-control	argued; fitted to clinical observation, no test reported	§9
C7	Pain and emotions are rapid reward-dependent cycles, not conditioned reflexes	speculative; the précis says “in theory at least”	§4.1
C8	Emotional reward is limited by premature satiation, so its occasions must stay surprising	argued; one reference to reward-centre activity needing surprise	§§10–11

Concepts

• interest: “the processes selected for by a particular kind of reward”; named only where rewards that dominate at different times conflict (§3.1).
• picoeconomics: “a micro-microeconomics” of the internal marketplace (§3.1).
• personal rule: a rule to treat all members of a category alike, enforced by the stake that each choice puts on the whole category (§6).
• bundling: summing the discounted values of a series of rewards by perceiving one choice as predicting the rest (§5).
• lapse district: an area of behaviour where a person no longer dares attempt will, fenced off to protect the rule elsewhere (§9.2).
• premature satiation: harvesting emotional reward “before it’s ripe”, so that appetite is wasted (§§10–11).

Connections

• Schelling. The précis cites Schelling (1960) for “limited warfare” among successive motivational states. Schelling’s own account of self-command, read here as LIT-602 (NOTE-452), describes the same divided chooser from the side of tactics. Schelling’s stark cases (the pain button, the stroke device) are commitment devices of Ainslie’s first kind. His New Year’s resolution, which “raises the stakes” by violating “one’s personal constitution”, is a personal rule.
• Berridge. The précis takes Berridge and Robinson’s “wanted” but not “liked” behaviours (1998) as instances of its itch-like cycle, and rejects their reading of them as classically conditioned. Dayan and Berridge (LIT-584) defend Pavlovian prediction as model-based. They do not reduce it to instrumental reward, so the two disagree about conditioning.
• Dual controllers. Daw, Niv and Dayan (LIT-580) reject the economic picture of an “irrational” limbic system fighting a “rational” prefrontal one, and give the two controllers the same ends. Ainslie rejects the same picture for a different reason: there are many controllers, and their ends differ only by time.
• Shenhav, Botvinick and Cohen (LIT-585) treat patient intertemporal choice as a “default override” needing costly control from a dedicated specification system. That is close to the “organ” model the précis rejects. The two accounts are not tested against each other in either text.

Bearing on the record

• THEORY-056 (regulation is not separate from generation). The précis supports its architecture from outside emotion research. Self-control is the interaction of first-order interests, and there is no governor: “Such interests are not options chosen by an overarching ego, the top down model”. It also addresses the theory’s stated largest gap, which state wins: the one with “the greatest discounted reward at a given moment”, with bundled future rewards added by self-prediction. It differs from the theory in one respect. The extra motive that will recruits is not a concurrent emotion but an expectation about one’s own future choices. Whether that is “one motive state checking another” or a distinct kind of process is a real question for the theory. The précis’s account is filed as THEORY-067, and THEORY-056 says where it stands.
• THEORY-029 (higher-order attitudes cannot confer ownership by order alone). The précis does not discuss ownership. It does reject hierarchical, faculty-based accounts of the will’s force, by a regress of the same shape: an organ of will “has to be guided by some evaluation process outside of motivation”. It also suggests Bratman’s resolute choice reduces to bargaining. If so, a planning agent’s authority would rest on the recursive stake of precedent and not on the order of an attitude. That bears on THEORY-029 without deciding it.
• Self-determination theory (LIT-558, LIT-560; THEORY-047, THEORY-049, THEORY-054). Two points of contact, neither cited by either side. The will’s first side effect, choices “detached from their immediate outcomes” with “an aloof, legalistic quality”, resembles introjected regulation (THEORY-047). The indirection puzzle, in which “Trying to have fun usually spoils the fun”, parallels the undermining of intrinsic motivation (THEORY-049). The mechanisms are different, so this is a convergence of phenomena, not support for either account. SDT also denies that will is “a non-brain-related force” (LIT-560, as read in NOTE-441). Ainslie’s will is likewise not a force but a pattern of reward.
• No ML instruction. Nothing here belongs in the anthology.

Limitations

• A précis, not the argument in full. The précis says it is “necessarily selective”. Derivations are referred to the book and to Ainslie (1992).
• The central mechanism is recursive and hard to test. The précis says so (§8). Its best evidence is that it resolves philosophers’ thought experiments, which shows coherence, not truth.
• Reported, not re-examined. The hyperbolic-discounting literature is cited, not reviewed. Whether hyperbolic fits beat alternatives such as quasi-hyperbolic or two-system models is not discussed.
• The commentaries were not read. BBS printed objections and a response with the précis, and none of them is reflected here.
• The account of emotion (§§4, 10–11) is the most speculative part. Its model of pain as a fast cycle of reward and non-reward is offered “in theory at least” and is not tested.

Open questions

• Does bundling by self-prediction, rather than an external choose-for-the-series manipulation, produce the predicted shift? Kirby and Guastello’s suggestion condition gave “an intermediate degree of self-control”, which is suggestive only.
• Can a lapse district be shown to follow a first failure of a personal rule, as the account predicts, rather than to precede it?
• What do the BBS commentaries object to, and does the response meet them?

Corrections

• Pages. The deposit gives BBS 28, 635–673. That range includes the commentaries and response. Crossref gives the target article as pp. 635–650.
• The book’s figure. The précis notes that its Figure 6B corrects Figure 10B of the book, which was “inaccurately drawn” (note 3). A reader of LIT-583 should use the précis’s version.
• ISBN. The deposit prints “IBSN: 0-521-59694-7” for the book. This is a typo for ISBN.