Contribution
The paper brings Bentham’s hedonic utility back into economics as a measurable quantity distinct from choice-revealed utility. It sums up the Peak-End and duration-neglect findings of Kahneman’s group as evidence that the two utilities come apart in normal people, and it gives a normative axiomatization of the total experienced utility of an extended episode.
Key insight
Choice is governed by memory of experience, not by experience: “In many other situations the sovereign masters that determine what people do are not pleasure and pain, but memories of pleasure and pain” (p. 385). Because memory summarizes episodes by a “representative moment” (p. 384) that leaves out duration, preference can be shown to violate dominance, and utility maximization becomes an empirical claim.
Assumptions
- Hedonic states are measurable as psychophysical magnitudes are, and “the variance they share can serve to operationalize the concept of instant experienced utility” (p. 380). Here “they” are verbal reports, physiological indices and facial measures.
- A distinctive neutral point exists and is stable as an experience while the stimulus that produces it shifts with context, as in the two pails of warm water (p. 380).
- Instant utility is a sufficient statistic. Any effect of past or anticipated consumption must already be in the moment’s rating. This is what makes separability (Axiom 3) reasonable (pp. 390–391).
- No endorsement of hedonism. The authors disclaim it: “no endorsement of Bentham’s view of pleasure and pain as sovereign masters of human action is intended” (p. 377).
Key results
- The distinction. ““Decision utility” is the weight of an outcome in a decision. “Experienced utility” is hedonic quality, as in Bentham’s usage” (p. 375). The amnesic toaster: “The patient’s decision utility for using the two toasters is equal, but his experienced utilities are quite different” (p. 376).
- Instant and remembered. “instant utility is the pleasure or distress of the moment; remembered utility is the retrospective evaluation of a temporally extended outcome” (p. 379). Instant utility is “a measure of hedonic and affective experience, which can be derived from immediate reports of current subjective experience or from physiological indices” (p. 376).
- Functions. “Pleasure is evidently a “go” signal” and “Pain is a “stop” signal, which interrupts activities that are causing harm”. “The common characteristic of the basic forms of pleasure and distress is that they regulate the response to the current situation” (p. 379). Hedonic value tracks need. Sweet food’s hedonic value falls during a meal (citing Cabanac 1971), which is alliesthesia under another name (p. 379).
- Peak-End and duration neglect. “(1) Peak-End evaluation: the remembered utility of pleasant or unpleasant episodes is accurately predicted by averaging the Peak (most intense value) of instant utility (or disutility) recorded during an episode and the instant utility recorded near the end of the experience”. “(1) Duration neglect: the duration of experiences has little or no independent effect on their remembered utility” (p. 381). In the colonoscopy study the Peak-End correlation with global rating was .67 and duration’s .03 (p. 383).
- Decisions by remembered utility. In the cold-water study, subjects “simply determined which of the two memories they disliked less”, and “the choice is made by comparing remembered utilities, without performing the explicit comparison that could reveal the Long trial to be a dominated option” (p. 387). Shown the two profiles side by side, subjects obey dominance (p. 387). Hence “Their decisions did not maximize experienced utility (#9), although they may have maximized remembered utility (#10)” (p. 377).
- The normative theory. “Ordinal measurement of instant utility implies that any two moments of experience can be compared, to establish which of them carries the higher hedonic value” (p. 389). The three axioms hold if and only if total utility orders profiles by the integral of a nondecreasing value transform of instant utility that assigns 0 to 0, and that transform is a ratio scale (Theorem 1, p. 391). Under cardinal instant utility, the integral of instant utility itself (Theorem 2, p. 392).
- Measurement. Real-time measures “avoid the biases of memory and evaluation that affect retrospective judgments” (p. 394), and “We believe that experienced utility can only be understood when it is directly measured” (p. 395).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Decision and experienced utility are distinct and separately measurable | strong | conceptual; amnesic example; experiments |
| C2 | Remembered utility follows Peak-End with duration neglect | strong | several studies, film, colonoscopy, cold water, sound |
| C3 | Choices between episodes follow remembered, not total experienced, utility, and so can violate dominance | moderate | cold-water (N = 32, replication 37), sound pairs |
| C4 | Total experienced utility is the time integral of a value transform of ordinal instant utility with a neutral zero | proved | Theorem 1 |
| C5 | Instant utility can be measured, with shared variance across report and physiology | moderate | psychophysics; interpersonal agreement in colonoscopy study |
Concepts
- decision utility — the weight of an outcome in choice, inferred from choices.
- experienced utility — hedonic quality, in Bentham’s sense.
- instant utility — the pleasure or distress of a moment, by real-time report or physiology.
- remembered utility — the retrospective evaluation of an episode.
- total utility — normative: the integral over time of transformed instant utility.
- predicted utility — beliefs about future experienced utility.
Connections
Berridge and Kringelbach (LIT-1459) make the neural version of the same split, with ‘wanting’ on the decision side and ‘liking’ on the experience side. Their note that ‘liking’ tracks satiety and KWS’s citation of Cabanac 1971 both point to alliesthesia (LIT-1467, unread). Thinking, Fast and Slow (LIT-1209, Deferred) restates the experiencing and remembering selves for a general audience.
Bearing on the record
Is decision against experienced utility a counterexample to valence is potential? Not directly, but it constrains how potential may be measured. The cold-water subjects pull their hands out “as soon as they were allowed to” (p. 387). In the moment, pain wins its conflict, as the “stop” signal that “interrupts activities that are causing harm”. That is control precedence, and KWS’s functional gloss on instant utility is close to the record’s reading of potential (TERM-076, THEORY-055). The dissociation is elsewhere: between instant utility and choices made later between remembered episodes. Two things follow for CLAIM-440:
- If potential is read from deferred choices between episodes, it measures remembered utility, and remembered utility diverges from instant utility. So layer (1) must be measured on conflicts won during the episode, not inferred from later preference. QUESTION-268 should say so.
- The dissociation is not a case of strong potential with weak valence, as Berridge’s is. It is a case of one ordering (the moment’s) being replaced by another (memory’s) across time. On the account that is a change of which ordering prevails, and it can be described without contradiction.
For layer (2), instant utility is the record’s hedonic marker in self-report form. It is measured in real time, apart from choice, and it agrees across patient, observer and physician in the colonoscopy study (pp. 383–384). Its weakness for the record is that it is report, which makes it hard to keep separate from layer (3). KWS’s own proposal to add physiological indices is the better route.
For layer (3) and the felt-as-self-model candidate, remembered utility is a warning. Remembered utility is a system’s model of its own past hedonic states, and KWS show that this model is systematically wrong: it neglects duration and weights peak and end. If “felt” means the system models its own potential, and the model is read off by asking the system retrospectively, it will inherit these biases. So layer (3) cannot be validated against retrospective report. It needs a concurrent measure, or a comparison like Hesp et al.’s (LIT-1468), where the self-modelling level is removed and behaviour compared.
On ordinality. The record has insisted that an ordering does a lot without a measure (CLAIM-347, CLAIM-436). Theorem 1 gives a precise case. Ordinal instant utility, with a distinguished neutral point and the three axioms, yields a ratio-scale value whose integral is total utility. So a neutral zero plus separability turns an ordering into a currency over time. The sign clause of TERM-076 needs a neutral point too, and the two neutrals differ. KWS’s is a constant experience (“neither pleasant nor unpleasant”). Carver and Scheier’s (LIT-1441) is a reference rate of progress. KWS’s pails example, where the neutral stimulus shifts with adaptation, is compatible with a reference course. It does not decide between them.
Limitations
The experiments are short, passive and hedonically homogeneous episodes, as the authors say (p. 388). The cold-water sample is small. The normative theory assumes constant calibration of the instant scale across compared episodes, which the authors grant fails for radically different situations (p. 390). The paper does not address non-human animals or organizations.
Open questions
- Is there a behavioural, non-report measure of instant utility, a within-episode analogue of taste reactivity, that could serve layer (2) in humans?
- Does a system’s remembered utility count as a model of its own potential? If so, is the felt-as-self-model candidate a claim about remembered or about instant valence?