Economic modellers and methodologists have ascribed a variety of meanings to the notion of utility. Indeed, this term has been employed in so many different senses that one wonders whether we can meaningfully characterize utility as a unified concept.Footnote 1 In the recent literature, three senses of the word stand out as particularly prominent. Firstly, contemporary decision theorists typically regard utility as a mathematical representation of preferences to be inferred from agents’ choices (see e.g. Broome Reference Broome1991: 3; Hausman and McPherson Reference Hausman and McPherson2009: 5).Footnote 2 Secondly, other authors use the term utility to refer to some purported hedonic magnitude reflecting individuals’ experiences of pleasure and pain (see e.g. Kahneman et al., Reference Kahneman, Wakker and Sarin1997: 375; Kahneman, Reference Kahneman, Tversky and Kahneman2000a: 2). Finally, some speak of utility as a desirability signal that can be accurately measured in the activation patterns of specific neural areas (Camerer et al. Reference Camerer, Loewenstein and Prelec2004: 556) and allege that ‘map-like structures in the brain [. . .] are actually the subject of economic theory’ (Glimcher et al. Reference Glimcher, Dorris and Bayer2005: 238). These three notions are often named decision utility, experienced utility and neural utility respectively. Below I refer to the last two notions as ‘true utility’, since they are usually employed to indicate some psycho-physical magnitude as opposed to a mathematical construct such as decision utility.
Over the last decade, several authors (e.g. Camerer et al. Reference Camerer, Loewenstein and Prelec2004; Glimcher et al. Reference Glimcher, Dorris and Bayer2005) have advocated complementing decision utility with some notion of true utility as a central concept of decision theory. Some (e.g. Camerer Reference Camerer2006; Dolan and Kahneman Reference Dolan and Kahneman2008) have gone as far as to urge economists to replace decision utility and modify traditional decision theory in light of neuro-psychological findings. In the words of Camerer, ‘conventional economic language can indeed approximate a lot of neural phenomena’, yet ‘at some point it is more efficient to simply adopt constructs as they are defined and understood in other fields’ (Reference Camerer, Caplin and Schotter2008a: 45). In this article, I examine these calls in favour of true utility and argue that they face substantial difficulties.
The contents are organized as follows. In Section 1, I draw some conceptual distinctions between the three aforementioned notions of utility and consider various methods for measuring experienced utility and neural utility. In Sections 2–4, I critically assess the thesis that decision theoretic analyses should be based on experienced utility or neural utility. In doing so, I provide both empirical and conceptual reasons to doubt that economists should complement or substitute decision utility with some notion of true utility as a central concept of decision theory. My contribution can be seen as a response to those who advocate the replacement of ‘the mathematical ideas used in economics’ with ‘more neurally detailed descriptions’ (Camerer Reference Camerer2005) and speak of ‘substitut[ing] familiar distinctions between categories of economic behaviour’ with the ones adopted in other disciplines (Camerer et al. Reference Camerer, Loewenstein and Prelec2005: 15).
Before proceeding, let me put forward some preliminary caveats regarding the focus of this paper. Firstly, the list of issues I discuss does not include all the difficulties faced by the proponents of true utility measures. Still, it encompasses what I regard as the most pressing challenges that presently hinder their attempts to inform decision theoretic analyses. Secondly, I do not explore the historical interrelations between distinct notions of utility, as the cogency of my remarks does not rest on how one reconstructs them. For the purpose of this enquiry, it suffices to note that the possibility of measuring experienced utility was prefigured well before the advent of standard decision theory (see e.g. Edgeworth Reference Edgeworth1881 [1967]), whereas the notion of neural utility has appeared in the economic literature only during the last decade. Thirdly, the concept of utility is amenable to other interpretations besides the three ones I examine (see e.g. Bradley, forthcoming, ch. 1). I do not expand on those interpretations here, as they are mostly orthogonal to the focus of my investigation.
Finally, decision theory has both positive and normative applications. Sections 2 and 3 focus prevalently on the former, while Section 4 discusses some normative applications of experienced utility and neural utility analyses. I shall devote particular attention to the distinction between positive uses of decision theory to predict and model choices and its application as a normative tool for assessing the welfare implications of decisions. This distinction has direct relevance for my evaluative exercise, as the applicability of different measures of utility may importantly depend on what economists employ decision theory for. As I argue below, however, there are significant interrelations between positive and normative decision theoretic analyses. I shall comment in various places on these interrelations and the implications they can be taken to have for the merits of true utility measures.
1. FROM DECISION UTILITY TO TRUE UTILITY
Traditional decision theory regards choices as rational to the extent that the associated preferences satisfy specific consistency requirements (see e.g. Samuelson Reference Samuelson1938; see also Bhattacharyya et al. Reference Bhattacharyya, Pattanaik and Xu2011, for a review). In such a context, no explicit assumptions are made about the neuro-psychological substrates of agents’ preferences. This notion of rationality concerns whether one's preferences are consistent with each other, and places no constraints on the content of particular preferences, taken individually (see e.g. Sugden Reference Sugden1991: 758–760). Yet, even these basic consistency requirements are occasionally violated (see e.g. Loomes et al. Reference Loomes, Starmer and Sugden1991, on transitivity). Furthermore, observed decisions can vary considerably depending on the way preferences are elicited (see e.g. Lichtenstein and Slovic Reference Lichtenstein and Slovic1971) and how the available options are described (see e.g. Tversky and Kahneman Reference Tversky and Kahneman1981).
When these findings were first presented, many economists reacted by doubting the reliability and the robustness of the collected evidence. In particular, several authors argued that the documented anomalies were more likely to reflect peculiar features of the examined choice settings than widespread behavioural tendencies. This line of defence, however, came under increasing pressure when the systematic character of anomalies was documented (see e.g. Kahneman et al. Reference Kahneman, Slovic and Tversky1982). At this stage, various economists attempted to modify traditional models so as to reconcile them with the available evidence (see e.g. Machina Reference Machina1982, on generalized expected utility theory). More recently, some authors have advocated complementing or even replacing the notion of decision utility with specific neuro-psychological constructs. The idea is to view agents as ranking outcomes in terms of some ‘objective measure’ (Bruni and Sugden Reference Bruni and Sugden2007: 170) and ground decision theoretic analyses on such measure for positive or normative purposes (see e.g. Glimcher and Rustichini Reference Glimcher and Rustichini2004: 452; Kahneman and Sugden Reference Kahneman and Sugden2005: 161).Footnote 3
Various measures of utility have been proposed in the literature. Below I focus on experienced utility and neural utility in turn, explicating some major conceptual distinctions between these notions and decision utility. The expression experienced utility is usually employed to indicate a purported magnitude reflecting agents’ hedonic experiences to be measured with psycho-physical methods (Read Reference Read2007: 58).Footnote 4 Measurements of experienced utility and decision utility diverge in a wide range of circumstances. For example, for there to be a general correspondence between decision utility and experienced utility, agents must be able to make accurate predictions of the hedonic consequences of their decisions (Kahneman and Sugden Reference Kahneman and Sugden2005: 168). Nonetheless, individuals frequently fail to anticipate the effects various choices have on their future preferences and hedonic states (see e.g. Kahneman and Snell Reference Kahneman, Snell and Hogarth1990). In fact, some are aware of their tendency to mispredict experienced utility, but keep mispredicting it on particular occasions (Dolan and Kahneman Reference Dolan and Kahneman2008: 227).Footnote 5
Three notions of experienced utility can be usefully distinguished, namely: instant utility, a moment-based measure which reflects the valence and the intensity of agents’ ongoing hedonic experiences (Kahneman Reference Kahneman, Tversky and Kahneman2000a: 2); remembered utility, a memory-based measure ‘inferred from a subject's retrospective reports of the total pleasure or displeasure associated with past outcomes’ (Kahneman et al. Reference Kahneman, Wakker and Sarin1997: 376); and anticipated utility, a forward-looking measure which targets ‘a person's ex ante beliefs about the hedonic quality of future experiences’ (Kahneman and Sugden Reference Kahneman and Sugden2005: 174). In such a context, measurements of so-called total utility can be obtained by taking the temporal integral of instant utility data (Kahneman et al. Reference Kahneman, Wakker and Sarin1997: 389; Kahneman Reference Kahneman, Tversky and Kahneman2000a: 7). This possibility was allegedly anticipated by Edgeworth (Reference Edgeworth1881 [1967]), who suggested that an imaginary instrument – the hedonimeter – could measure individuals’ hedonic experiences over temporally extended intervals.Footnote 6
The notion of neural utility relates to the activation patterns of particular areas of the neural architecture. The idea is that desirability ‘is realized as a concrete object, a neural signal in the human brain, rather than as a purely theoretical construction’ (Glimcher and Rustichini Reference Glimcher and Rustichini2004: 452). The proponents of neural utility often argue that standard decision theory can be employed to model not just consistent decisions, but also specific areas’ activation patterns. In their view, ‘the utility calculations that people were assumed to do really happen in the brain’ (Park and Zak Reference Park and Zak2007: 50) and traditional economic models of choice constitute a ‘limiting case of a neural model with multiple utility types’ (Camerer et al. Reference Camerer, Loewenstein and Prelec2004: 564). Over the last few years, several authors have suggested that anatomically delimited neural populations track rewards’ subjective values in various experimental settings (see e.g. Glimcher Reference Glimcher, Glimcher, Camerer, Fehr and Poldrack2009; Kable and Glimcher Reference Kable and Glimcher2009). In particular, there is growing consensus that sub-regions of the ventromedial prefrontal cortex/orbitofrontal cortex (henceforth, vmPFC/OFC) encode subjective values across multiple reward types (see e.g. Levy and Glimcher Reference Levy and Glimcher2012, for a meta-analysis of recent fMRI studies).
According to some authors, the collected findings do not directly support standard decision theory, but demonstrate that the subjects who behave in accordance with such a theory ‘do so because they neurally represent something having the properties of utility – a neural activation that encodes the desirability of an outcome in a continuous monotonic fashion’ (Glimcher Reference Glimcher2011: 133–134). For their part, others (e.g. Camerer Reference Camerer2007: C26; Quartz Reference Quartz2008: 463) take the available neuro-physiological evidence to provide direct support to standard decision theory. Let us focus on this latter interpretation. As noted by Vromen (Reference Vromen2010: 33), it is interesting that standard decision theory can be applied to entities that few economists had previously regarded as the target of their predictions. Yet, the mere fact that such a theory can be used to model the activations of specific neural areas falls short of implying that it accurately predicts agents’ choice behaviour (see Ross Reference Ross2008a: 481, for a similar remark).
Before concluding this section, let me emphasize one important respect in which the notions of experienced utility and neural utility differ. Neural utility is usually taken to be measurable accurately in specific areas’ activation patterns. The proponents of experienced utility, instead, often leave it ambiguous whether this notion is measurable solely by means of third-person data or whether measuring it requires one to consider also agents’ first-person evaluations. To be sure, some authors speak of experienced utility as a magnitude amenable to objective measurements and allege that ‘in spite of the immense diversity in the occasions that evoke pleasure [. . .] the hedonic attribute that they share is salient and readily recognized’ (Kahneman et al. Reference Kahneman, Wakker and Sarin1997: 380). Still, these authors fail to specify what exactly the hedonic attribute purportedly shared by heterogeneous experiences consists in. In particular, they put forward dissimilar views as to whether experienced utility ultimately reflects just individuals’ feelings or also their evaluative attitudes regarding their own hedonic experiences (see e.g. Alexandrova Reference Alexandrova2005: 311, on Kahneman Reference Kahneman, Tversky and Kahneman2000a; see also Kelman Reference Kelman2005: 406–408).Footnote 7 As I argue in the next section, these ambiguities severely complicate the assessment of the measurability, integrability and interpersonal comparability of experienced utility.
2. EXPERIENCED UTILITY: CONCERNS
In this and the next section, I examine several empirical and conceptual concerns regarding measures of experienced utility (Section 2) and neural utility (Section 3). In doing so, I argue that the available neuro-psychological evidence does not license the claim that economists should base decision theoretic analyses on these notions of true utility. With regard to experienced utility I discuss three issues, which respectively relate to its measurability, integrability and interpersonal comparability. Some of these issues point to familiar debates concerning alleged difficulties for traditional decision theory. As we shall see below, true utility analyses are especially vulnerable to these difficulties.
(i) The proponents of experienced utility often take this notion to indicate some hedonic magnitude that can be measured in reliable and accurate terms. Unfortunately, distinct authors hold heterogeneous positions regarding what methods should be used to measure it, and different methods frequently yield dissimilar results. To illustrate this, consider the remarks that Kahneman et al. (Reference Kahneman, Wakker and Sarin1997) put forward concerning instant utility. In their view, instant utility ‘can be derived from immediate reports of current subjective experience or from physiological indices’ (1997: 376). Regrettably, these measures can diverge significantly, and it is unclear which of them should be regarded as more accurate and reliable. Related concerns apply to other measures of experienced utility. Let me expand on this issue.
Over the last few decades, several methods for measuring agents’ hedonic experiences have been advocated. Prominent proposals include the experience sampling method (see e.g. Hektner et al. Reference Hektner, Schmidt and Csikszentmihalyi2006, parts I–II) and the day reconstruction method (see e.g. Kahneman et al. Reference Kahneman, Kreuger, Schkade, Schwarz and Stone2004). The former consists in asking subjects at random times during the day to answer questions about their current affective state. In the latter, each participant is asked to reconstruct her previous day into short episodes and then rate each episode in terms of various feelings. These methods seem less liable than others to biases and confounding factors (see e.g. Frederick and Loewenstein Reference Frederick, Loewenstein, Kahneman, Diener and Schwarz1999, on happiness reports, and Kahneman and Krueger Reference Kahneman and Krueger2006, on life satisfaction measures). Even so, subjects’ responses can noticeably vary depending on the framing of the examined questions and how the provided response scales are interpreted. This variability, in turn, constrains both the interpersonal comparability and the informativeness of experienced utility reports. By way of illustration, consider the following Garden of Eden example (see also point iii below).
Imagine that Adam and Eve have to provide a report regarding the hedonic experience of eating the fruit of the forbidden tree. Suppose that both of them are sincere and have adequate access to their inner hedonic states. Assume further that Adam states that he ‘really liked’ eating the apple, while Eve just says that she ‘liked’ eating it. These reports do not license the claim that Adam had more hedonic pleasure when eating the apple than Eve did. For Adam and Eve may be rating their experiences in terms of dissimilar response scales, i.e. non-equivalent mappings between linguistic expressions of satisfaction and inner hedonic states. In real-life experiments, averaging responses across individuals can partly mitigate the effects of interpersonal variability in the use of response scales. Yet, researchers’ attempts to link response categories to terms having a uniform meaning across respondents usually provide ‘no guarantee that respondents use the scales comparably’ (Kahneman and Krueger Reference Kahneman and Krueger2006: 18–19; see also Angner Reference Angner2011, for a discussion).
Similar remarks can be made regarding the informativeness of experienced utility measures. To see this, let us focus on remembered utility and consider the following First Date example. When remembered shortly thereafter, a first date can give one an unsurpassable pleasure, awakening her noblest and most elevated sentiments. However, the remembered utility associated with such an episode can vary dramatically depending on how the subsequent relationship unfolds. Moreover, one may continue to derive utility (or disutility) from that episode long after it ends by remembering it, by experiencing how her inner life has been affected by it, etc. To be sure, agents’ decisions occasionally correlate with remembered utility reports, and researchers identified some systematic patterns in remembered utility evaluations (see e.g. Fredrickson and Kahneman Reference Fredrickson and Kahneman1993, on duration neglect). Still, remembered utility measurements can significantly fluctuate depending on factors such as what memories one happens to recall, the valence she temporarily ascribes to those memories, etc. In this respect, it remains unclear how informative it would be to know what the remembered utility associated with an episode at a given moment is.
(ii) The limitations associated with retrospective and anticipatory hedonic reports prompted various authors to develop moment-based measures of experienced utility. These measures are frequently claimed (see e.g. Kahneman Reference Kahneman, Kahneman, Diener and Schwarz1999: 20) to be more accurate and reliable than retrospective and anticipatory hedonic reports. Nonetheless, even moment-based measures involve individuals’ evaluations of their own hedonic states and are vulnerable to serious criticisms. To illustrate this point, let us examine Kahneman's prominent index of what he calls ‘objective happiness’.
Kahneman (Reference Kahneman, Tversky and Kahneman2000a: 6–8) imposes various requirements on how to construct total utility from profiles of instant utility as a measure of objective happiness. Among these requirements, we find: separability, according to which ‘the order in which moment-utilities are experienced does not affect total utility’; time neutrality, which states that ‘all moments are weighted alike in total utility’; and so-called ordinality, which requires that ‘any two moments of experience can be compared, to establish which of them carries the higher hedonic value’ (Kahneman et al. Reference Kahneman, Wakker and Sarin1997: 389). As argued by Kahneman (Reference Kahneman, Tversky and Kahneman2000a), separability and time neutrality are necessary – and, together with inclusiveness (see below) and ordinality, sufficient – for representing utility profiles as a function of the time an agent spends at various levels of pleasure and pain. Unfortunately, each of these conditions is vulnerable to severe objections. Let us consider those conditions in order.
Separability requires that ‘the contribution of an element to the global utility of [a] sequence is independent of the elements that preceded and followed it’ (Kahneman Reference Kahneman, Tversky and Kahneman2000a: 6). As noted by Kahneman et al. (1997: 391), separability in terms of outcomes is liable to obvious counterexamples. For instance, the hedonic correlates associated with taking a nap and reading Wittgenstein's Tractatus tonight depend on whether you spent the day sleeping or compulsively computing Hamiltonians. At the same time, Kahneman et al. contend that separability of instant utilities is justifiable provided that one's measure of instant utility ‘incorporates all order effects and interactions between outcomes’ (1997: 391; see also Kahneman Reference Kahneman, Tversky and Kahneman2000a, on inclusiveness). Regrettably, these authors leave it unclear on what grounds we are to ascertain whether this condition is satisfied. Moreover, it seems doubtful that our measures of instant utility reflect all the affective consequences of previous experiences and the anticipation of future events. To give one example, one's past experiences may adaptively influence her current hedonic states in ways that elude her conscious evaluations or existing physiological indicators of instant utility.
Regarding time neutrality, total utility measures differ markedly from decision utility and remembered utility ones, which typically assign more weight to outcomes that respectively occur early and late in a sequence. According to Kahneman et al., a time-neutral perspective is ‘appealing, both as a rule of personal prudence and as a principle of social planning’ (1997: 393). Yet, much controversy surrounds this issue. For instance, one may maintain that the decreasing psychological connectedness between our present self and our future selves gives us a reason to discount the utility of those selves. More generally, the inherent temporal situatedness of the position from which agents evaluate outcomes makes it dubious that the temporal distance between an outcome and the moment in which such outcome is evaluated is irrelevant to its evaluation. I am not concerned here with settling the controversy over the rationality of discounting the future. Still, the proponents of total utility cannot gloss over this issue if they are to provide decision theorists with a convincing case for relying on such a measure.
As to ordinality, Kahneman (Reference Kahneman, Tversky and Kahneman2000a: 10) himself acknowledges that whether agents can meaningfully compare distinct hedonic states is an open empirical issue. For the purpose of this enquiry, it suffices to observe that people are often unable to express definite preferences between heterogeneous experiences (e.g. suppose you had to choose between undergoing some excruciating physical pain and suffering from a profound depression). Moreover, even when they are able to do so, their stated preferences might not rest on considered comparisons of underlying hedonic magnitudes. This, in turn, could happen both because agents may fail to engage in careful reflection about their own hedonic states and because their evaluations may not refer to any definite hedonic magnitude in the first place.
(iii) The proponents of experienced utility frequently take their measures to be interpersonally comparable across contexts. In their view, ‘a distinctive neutral point [neither pleasant nor unpleasant] permits comparisons across situations and persons’ (Kahneman Reference Kahneman, Tversky and Kahneman2000a: 5). The thought is that although different stimuli may give rise to neutral experiences, ‘the neutral experience itself is constant’ and can be used to ground meaningful interpersonal comparisons of experienced utility across contexts (Kahneman et al. Reference Kahneman, Wakker and Sarin1997: 380; see also Kahneman and Varey Reference Kahneman, Varey, Elster and Roemer1991). Unfortunately, it is questionable whether the so-called neutral point aptly serves this purpose. To be sure, various authors insist that we can often make interpersonal comparisons on the alleged basis that the functions relating hedonic states to observable psycho-physical variables are ‘qualitatively similar’ across individuals (see e.g. Kahneman et al. Reference Kahneman, Wakker and Sarin1997: 380). Yet, appealing to this vague notion of qualitative similarity does not license wide-ranging conclusions regarding the interpersonal comparability of experienced utility reports.Footnote 8
In such a context, the further question arises as to whether experienced utility can be measured and compared interpersonally in purely physical terms. This question relates to an issue discussed extensively by philosophers of mind, namely whether having complete knowledge of an agent's physical states is sufficient for acquiring complete knowledge of her mental states (see e.g. Nagel Reference Nagel1974; Jackson Reference Jackson1982). Establishing the measurability and interpersonal comparability of experienced utility in purely physical terms would require one to show that, once all the relevant physical data have been collected, there is nothing more to know regarding what it is like to enjoy a particular level of experienced utility. This constitutes a controversial issue in philosophical and psycho-physical research (see e.g. Papineau Reference Papineau2003; Chalmers Reference Chalmers and Gazzaniga2004; see also the end of Section 1 on the ambiguities inherent in several authors’ conceptualizations of experienced utility). In particular, it remains to be seen whether our increasing ability to measure neuro-physiological activity will enable us to ‘objectively compare mental state between individuals’ (Glimcher Reference Glimcher2011: 425).
3. NEURAL UTILITY: CONCERNS
Let us focus on neural utility. In this section, I examine some concerns related to the informativeness of neural utility measures, the purported reducibility of measures of experienced utility to measures of neural utility, and the underdetermination of accounts of the neural substrates of value representation by the available evidence. These three issues bear on both positive and normative applications of true utility analyses, ranging from attempts to predict choices to the development of putative measures of well-being. As we shall see below, rapid progress is being made in neuro-physiological research, yet many central questions regarding those three issues remain unanswered.
(i) My first concern relates to the informativeness of neural utility measures. In order to identify which areas compute and integrate value-related signals, researchers have to employ correlations with data such as observed choices and reported expressions of satisfaction. In this respect, several authors worry that the dependency of neuro-physiological measures on behavioural and psychological data may inherently constrain these measures’ potential to challenge or correct those data (see e.g. Bernheim Reference Bernheim2009: 31; see also Rubinstein Reference Rubinstein2008). To be sure, researchers have identified some correspondences linking choices and hedonic states to neural states in specific experimental settings. These correspondences may arguably be employed to build more predictive and explanatory models (see e.g. Fumagalli Reference Fumagalli2011, for a discussion). Yet, the point remains that establishing which areas compute and integrate value-related signals requires one to rely on behavioural and/or psychological evidence. This, in turn, limits the extent to which current neuro-physiological findings can inform decision utility and experienced utility analyses.
In recent years, various authors have argued that isolating some neuro-biological magnitude that is systematically related to decision utility will provide ‘fundamental economic insights’ (Caplin and Dean Reference Caplin and Dean2008: 669) and ‘a neurobiological foundation for standard economic theory’ (Levy and Glimcher Reference Levy and Glimcher2012: 1). Now, it would be implausible to deny that neuro-biological evidence may be used to evaluate and constrain specific models of choice. For example, knowledge of neuro-physiological activity can restrict the set of neural mechanisms that are taken to underlie observed decisions (see e.g. Camerer Reference Camerer2008b; Caplin et al. Reference Caplin, Dean, Glimcher and Rutledge2010). This, pace some authors (e.g. Gul and Pesendorfer Reference Gul, Pesendorfer, Caplin and Schotter2008), could in principle lead some economists to discard particular models of choice as neuro-biologically implausible.
That said, it remains unclear how exactly neuro-biological findings bear on the conceptual foundations of standard economic theory. To see this, suppose – following Glimcher (Reference Glimcher2011: ch. 6) – that we are able to identify some anatomically delimited neural population whose activations are systematically related to decision utility in specific choice settings. This finding will be of interest to many neuroscientists, yet does not per se provide ‘fundamental economic insights’ or ‘a neurobiological foundation for standard economic theory’. For in primis, profound conceptual differences remain between a variable such as the firing rate of some neurons and an abstract construct like decision utility (e.g. only the former is a cardinal object). And secondly, severe complications plague current attempts to derive economically significant implications from observed correlations between specific areas’ activations and choices. For instance, as illustrated in point iii below, a variety of neural populations seem to be involved in computing and integrating value-related signals, with many such populations playing other functional roles besides value representation.
Unfortunately, the proponents of neural utility do not always acknowledge these issues and often tend to overstate the implications of their findings for standard decision theoretic analyses. To give one example, recent studies have begun to investigate the neural underpinnings of value representation across reward types (see e.g. FitzGerald et al. Reference FitzGerald, Seymour and Dolan2009, on money and some consumer goods) and choice situations (see e.g. Platt and Huettel Reference Platt and Huettel2008, on choices under uncertainty). However, previous investigations predominantly focused on highly constrained laboratory settings where non-human primates engage in experimental tasks (e.g. repetitive choices between primary rewards) that are far less sophisticated and computationally demanding than decision problems commonly faced by human individuals (e.g. choices between multi-stage courses of actions involving abstract rewards). In such a context, several researchers took non-human primates to ‘provide a good model for human decision makers’ on the basis of some ‘similarity in behavioral data across species’ (Dorris and Glimcher Reference Dorris and Glimcher2004: 376). Now, animal studies certainly provide valuable insights and background evidence for the study of the human neural architecture. Nonetheless, the limited complexity of the examined tasks, coupled with the functional dissimilarities between humans’ and other primates’ neural architectures (see e.g. Passingham Reference Passingham2009), severely constrains the generalizability of experimental results in non-human primates to the more sophisticated decision problems faced by human individuals.
(ii) My second concern targets the purported reducibility of measures of experienced utility to measures of neural utility. The proponents of true utility often prefigure the development of measures that ‘will be sensitive to the many kinds of pleasure and anguish in people's lives’ (Kahneman Reference Kahneman, Tversky and Kahneman2000a: 12). In particular, various authors maintain that a partial reduction of psychological accounts of choice to neural ones is on the way (see e.g. Glimcher Reference Glimcher2011: ch. 2 and 7; see also Vromen Reference Vromen2012, for a discussion). In recent years, several insights regarding the neuro-physiological correlates of experienced utility have been provided.Footnote 9 The provision of increasingly accurate characterizations of the neuro-physiological correlates of experienced utility may foster significant explanatory and predictive advances in neuro-psychological research. Still, it does not per se allow for a neuro-physiological reduction of our accounts of hedonic experiences.
My point is not just that the neural underpinnings of some hedonic states are topographically and functionally too complex to be captured by current neuro-physiological investigations. Rather, my main concern is that any claim to have disclosed the neural constituents of individuals’ hedonic experiences would be philosophically naïve. In this respect, one might well reiterate that agents’ hedonic states supervene on their neural activation patterns, with every change in experienced utility reflecting some modification in neural utility. Yet, we do not have to be Cartesian dualists to doubt that experienced utility measures, which relate to individuals’ subjective experiences, can be reduced to third-person neural utility data. Furthermore, many of the concepts we employ to characterize hedonic experiences lack definite material referents at the neuro-physiological level. This does not preclude us from attempting to precisify these concepts so as to associate them with more specific neuro-physiological referents. However, the vagueness of those concepts considerably limits the susceptibility of our accounts of hedonic experiences to neuro-physiological reduction (see e.g. Papineau Reference Papineau2002: ch. 7).
(iii) The practitioners and the philosophers of distinct decision sciences often debate about the underdetermination of theories by the evidence.Footnote 10 Recent research on neural utility appears to face particularly severe instances of this problem. The thought is that the collected data do not generally impose stringent constraints on our accounts of the neural substrates of value representation in the human neural architecture. That is to say, different accounts of how value-related neural signals are computed and integrated have been proposed, and the available evidence does not enable us to effectively discriminate between them. Let me expand on this issue.
In exploring the neural underpinnings of choice, several authors advocated the so-called common currency hypothesis (see e.g. Landreth and Bickle Reference Landreth and Bickle2008). The idea can be summarized as follows. Economic agents often have to decide between complex alternatives. For principled choice to be possible, the value of the available options must be computed and integrated in terms of some common neuro-physiological scale of desirability. In the words of Montague and Berns, ‘to decide on an appropriate behavior, the nervous system must estimate the value of [. . .] potential actions, convert it to a common scale, and use this scale to determine a course of action’ (Reference Montague and Berns2002: 276). Over the last few years, growing evidence for a common currency representation in the human brain has been collected (see e.g. Smith et al. Reference Smith, Hayden, Truong, Song, Platt and Huettel2010). In particular, various authors have modelled decision making as a two-stage process whereby value-related neural signals are first computed and then integrated on a common desirability scale for choice implementation (see e.g. Glimcher Reference Glimcher, Glimcher, Camerer, Fehr and Poldrack2009; Kable and Glimcher Reference Kable and Glimcher2009). However, as I argue below, severe difficulties hinder the ongoing attempts to identify the neural substrates of value representation.
Researchers have localized anatomically delimited sub-regions of the vmPFC/OFC that seem to encode subjective values across some reward types (see Section 1). Even so, the neural substrates of value representation appear to be highly distributed, with several neural populations being involved in computing and integrating value-related signals (see e.g. Hare et al. Reference Hare, O'Doherty, Camerer, Schultz and Rangel2008; Rushworth et al. Reference Rushworth, Kolling, Sallet and Mars2012). This does not exclude that specific sub-regions of the vmPFC/OFC play an important role in value representation. Still, establishing that some of these sub-regions compute or integrate value-related signals falls short of demonstrating that no other areas contribute to such operations (see e.g. Hunt et al. Reference Hunt, Kolling, Soltani, Woolrich, Rushworth and Behrens2012). This, in turn, calls into question the alleged economic significance of reported correlations between vmPFC/OFC activations and choices. Moreover, it is questionable whether activations in the vmPFC/OFC can be robustly associated with value representation. For activations in this region (especially the vmPFC) have been ascribed a wide array of functional roles related to decision making, ranging from emotion regulation (see e.g. Mitchell Reference Mitchell2011) to the facilitation of theory of mind inferences (see e.g. Sebastian et al. Reference Sebastian, Fontaine, Bird, Blakemore, Brito, McCrory and Viding2012).
At this stage, one may well conjecture (see e.g. Levy and Glimcher Reference Levy and Glimcher2012: 10) that distinct sub-regions of the vmPFC/OFC provide dissimilar functional contributions to choice. Yet, researchers presently lack a clear understanding of how many kinds of value-related signals are computed in the human brain and how exactly these signals are integrated (see e.g. Cisek Reference Cisek2012). Indeed, it is still debated under what circumstances decisions are made on the basis of action-related values as opposed to values computed and compared in the space of goods irrespective of sensorimotor contingencies (see e.g. Padoa-Schioppa Reference Padoa-Schioppa2011, contra Rangel and Hare Reference Rangel and Hare2010). Additional complications arise when we consider what neuromodulatory interactions allegedly underlie decisions. To give one example, dopaminergic activity is often claimed to play a central role in reward valuations (see e.g. Schultz Reference Schultz2000). In particular, several authors take it to encode deviations between anticipated and obtained rewards (see e.g. Bayer and Glimcher Reference Bayer and Glimcher2005), with some going as far as to allege that dopamine is ‘a key input into the construction of utility’ (Caplin and Dean Reference Caplin and Dean2008: 669). However, other theories of dopaminergic function have been developed (see e.g. Redgrave and Gurney Reference Redgrave and Gurney2006), and various neurotransmitters besides dopamine can contribute to reward valuations (see e.g. Dayan and Yu Reference Dayan and Yu2006, on noradrenaline).
4. A NORMATIVE CASE FOR TRUE UTILITY?
As we have seen in the previous sections, a number of concerns arise regarding various authors’ proposals to base decision theoretic analyses on some notion of true utility. Now, let us suppose – for the sake of argument – that we could measure experienced utility and neural utility accurately, compare these measurements across subjects and choice contexts, etc. Even so, the question would remain as to whether decision theoretic analyses should be based on these notions. In recent years, several researchers (e.g. Kahneman Reference Kahneman, Tversky and Kahneman2000a; Kahneman and Krueger Reference Kahneman and Krueger2006) have advocated the adoption of true utility measures on normative grounds. In particular, some have suggested that experienced utility or neural utility provide a more suitable basis for welfare analyses than agents’ own preferences. The idea is that recent neuro-psychological research contributes to ‘the measurement of welfare and the design of economic and social systems that maximize welfare’ (Loewenstein and Haisley Reference Loewenstein, Haisley, Caplin and Schotter2008: 238). Let us assess the cogency of these assertions.
Suppose we are facing a situation where there is some fact of the matter as to what an agent's well-being is. A proponent of true utility might claim that well-being is somehow reducible to some measure of true utility, in the sense that well-being ultimately consists of particular hedonic experiences or neural activation patterns. This claim, however, fails to withstand scrutiny. For instance, we do not have to embrace a radical form of eudaimonism to doubt that experiences enhance well-being solely in virtue of their liability to generate pleasurable neuro-psychological states. Moreover, we can think of several cases where individuals having pleasant hedonic experiences do not appear to be well off (see e.g. Nagel Reference Nagel1979: ch. 1, on pleasures based on mistaken beliefs or inaccurate information).
The proponents of true utility occasionally concede that well-being depends not just on hedonic experiences or their neuro-physiological correlates, but also on ‘other aspects of life, such as autonomy, freedom, achievement’ (Kahneman and Sugden Reference Kahneman and Sugden2005: 176). At the same time, various authors insist that pleasurable neuro-psychological states constitute fundamental or necessary components of well-being (see e.g. Kahneman Reference Kahneman, Tversky and Kahneman2000a: 10). However, this claim also has counterintuitive implications. In the words of Bernheim, we often deem ourselves to be better off when ‘we have actual autonomy, liberty, and a firm grasp on reality even if, as a consequence, we must relinquish [. . .] pleasurable neurobiological sensations’ (2009: 30; see also Nozick Reference Nozick1974: 42–45). Regrettably, the proponents of true utility do not engage with these considerations and the implications they may be taken to have for the normative significance of their findings (see e.g. Hausman Reference Hausman2010: 332, contra Kahneman Reference Kahneman, Tversky and Kahneman2000a).
A proponent of true utility may further contend that some measure of experienced utility or neural utility constitutes a more accurate and reliable indicator of well-being than decision utility (see e.g. Kahneman Reference Kahneman, Kahneman, Diener and Schwarz1999: 7). Our intuitions concerning this issue might vary depending on what conception of well-being we endorse. Still, it is doubtful that, in general, experienced utility and neural utility provide more accurate and reliable indicators of well-being than decision utility. After all, there are various reasons why an agent may make choices that fail to promote her well-being (e.g. think of self-control problems). Yet, we can easily envision situations where choices yielding high experienced utility and/or neural utility face the same difficulty. To give one example, consider drug addicts and pathological gamblers. As noted by Ross (Reference Ross2008b: 132), a person can experience very pleasant sensations when her dopaminergic circuits suppress competing serotonergic connections. Regrettably, this process often results in addictive forms of behaviour that have a detrimental impact on individuals’ health. That is to say, neuro-psychological data may be employed to complement and refine choice-based indicators of well-being for evaluating specific decisions and policy interventions. Nonetheless, one would have to adopt a rather idiosyncratic conception of well-being to identify agents’ optimal decisions with the ones that maximize their experienced utility or neural utility.Footnote 11
In such a context, the additional question arises as to why economists should commit themselves to some particular conception of well-being. This concern can be explicated as follows. Economists often associate well-being with preference satisfaction and occasionally use the term ‘utility’ ambiguously to indicate both a substantive psychological state and a mathematical representation of preferences (Hausman and McPherson Reference Hausman and McPherson2009: 9–10). Even so, traditional decision theory does not rest on explicit presuppositions as to whether agents’ well-being consists in experiencing pleasure and avoiding pain, satisfying specific preferences, etc. Let us assume – for the sake of argument – that decision theory did rest on the assumption that well-being consists in maximizing the value of some measure of utility. One may wonder which notion of utility economists should favour and why agents should choose actions that maximize the value of this – rather than some other – measure.
By way of illustration, suppose that well-being consisted in maximizing the value of some measure of experienced utility. Still, several questions would remain such as: qua rational agent, should you maximize the value of instant, anticipated or remembered utility? Over what time spans? Or maybe should you compute a weighted average of distinct kinds of experienced utility, each measured over various temporal intervals? If so, how should the inputs of such a function be weighed? Providing principled grounds to identify the maximandum of this optimization exercise appears to be disconcertingly difficult. Moreover, the proponents of true utility have hitherto failed to specify precise criteria for ascertaining under what circumstances particular neuro-psychological measures should take primacy over agents’ considered and informed preferences as an indicator of well-being. To be sure, some allege that their analyses apply to ‘situations where a separate value judgment designates experienced utility as a relevant criterion for evaluating outcomes’ (Kahneman et al. Reference Kahneman, Wakker and Sarin1997: 377). Yet, the issue at stake is precisely when this is the case, and by means of what criteria we are supposed to identify these situations.
Even more severe difficulties affect putative attempts to build neural indicators of well-being. Let me comment briefly on two such difficulties in turn. Firstly, there is what we may call the identification problem of establishing what kinds of welfare-relevant operations (if any) are reflected by observed neural activity (e.g. aggregations of value-related signals at different times, responses prompted by anticipated or remembered outcomes, etc.). And secondly, one faces the so-called integration problem of ascertaining how exactly welfare-relevant signals are aggregated over time (see Bernheim Reference Bernheim2009: 30–31, for similar remarks). Both problems appear to be quite challenging. For instance, resolving the identification problem is complicated by the fact that even basic operations such as anticipating or reminiscing specific experiences can foster neural activations besides those associated with the hedonic correlates of these experiences. As to the integration problem, the variability that value-related neural activations exhibit across subjects and choice settings (see e.g. Padoa-Schioppa Reference Padoa-Schioppa2011) hinders the identification of robust generalizations about the intertemporal aggregation of welfare-relevant signals.
At this point, one might speculate that agents would often think that increasing experienced utility or neural utility enhances their own well-being. Unfortunately, this conjecture leaves us in the dark in cases where individuals have different intuitions regarding what constitutes or enhances their well-being. In this respect, it is hardly surprising that distinct proponents of true utility hold dissimilar views as to whether experienced utility and neural utility provide a compelling basis for policy evaluations.Footnote 12 More generally, the point remains that most of the current calls to ground economists’ normative analyses on some measure of true utility involve presuppositions which transcend both the scope of traditional decision theory and the evidential reach of neuro-psychological investigations (e.g. who is entitled to establish what agents’ well-being consists in? By means of what criteria we are to adjudicate disagreements regarding this issue?).
CONCLUDING REMARKS
In the recent literature at the boundary between economics, psychology and neuroscience, various measures of experienced utility and neural utility have been developed. As I argued in the previous sections, these measures differ profoundly both from each other and from the notion of decision utility at the core of standard decision theory. These dissimilarities make it highly advisable to keep the available notions of utility distinct ‘rather than presume to measure a single, unifying concept that motivates all human choices and registers all relevant feelings and experiences’ (Kahneman and Krueger Reference Kahneman and Krueger2006: 4).
As to the prospects of true utility investigations, the considerations put forward in this article do not exclude that advances in neuro-psychological research can inform positive and normative economic analyses. At the same time, they provide both empirical and conceptual reasons to doubt that economists should complement or substitute decision utility as a central concept of decision theory. In this respect, one might well advocate using measures of experienced utility and neural utility to improve specific models of decision making. Even so, additional argument is needed to license the claim that decision theoretic analyses should be based on these measures. Indeed, there is one sense in which the search for true utility appears to be not just incomplete, but also futile. After all, we may expect to acquire increasingly accurate information on the neuro-psychological substrates of choice across decision settings. Yet, it is hard to see how exactly these empirical investigations are supposed to foster the replacement of a mathematical construct such as decision utility.
