Navigation – Plan du site

The Pathological Paradigm of Neuroeconomics

Le paradigme pathologique de la neuro-économie
Nicolas Vallois
p. 525-556


Cet article propose une analyse de la neuroéconomie du comportement motivationnel, également connue sous le nom d’“économie neurocellulaire” (Ross, 2008). L’objectif principal de l’étude consiste à montrer que les neuroéconomistes sont influencés, de manière implicite, par des enjeux et des impératifs médicaux. À partir des thèses avancées par Georges Canguilhem dans le Normal et le Pathologique, j’essaierai d’interpréter la méthodologie de la neuroéconomie comme un paradigme pathologique. En effet, on peut mettre en évidence que, dans les expériences de neuroéconomie, la classification clinique des troubles addictifs est utilisée comme critère de définition normative de l’irrationalité. Une telle approche, fondée sur la pathologie, est négative en ce sens que c’est à partir de l’identification des différentes manières dont les individus ne doivent pas se comporter que l’on peut inférer la manière dont ces mêmes individus doivent se comporter. Cela permet aux modèles dits neuronaux de la prise de décision d’être à la fois descriptifs et normatifs. Les addictions sont pensées dans ce cadre comme des pathologies de la prise de décision. Parce que ces pathologies doivent être traitées, la neurophysiologie a une portée et un intérêt aussi bien pour l’économie positive que pour la réflexion sur les politiques économiques et l’économie du bien-être. La neuroéconomie illustre ainsi la montée en puissance récente d’une sémantique à caractère médical dans l’analyse du bien-être.

Haut de page

Texte intégral

1. Introduction – From Addiction to Public Policy: The Rise of Medical Semantics in Economic Theory

1Over the past few years, neuroscience has gained a growing influence on welfare issues. In particular, in an article in 2005, Bernheim and Rangel advocate a re-foundation of public economics on the basis of contemporary neurophysiology. This paper focuses on these recent propositions to connect neuroscience with welfare economics. My argument is that the neurobiological view of economic policy is implicitly grounded in medical semantics.

  • 1 In this paper, the term “behavioral economics” refers to the experimental studies of economic decis (...)
  • 2 My story excludes hence the research program of “behavioral economics in the scanner” (Ross, 2008) (...)
  • 3 Paul Glimcher is indeed one of its major contributors, as we shall see later.

2Bernheim and Rangel (2005) refer more specifically to “a new economic theory of addiction that bridges the gap between neuroscience and public policy” (Bernheim and Rangel, 2005, 19-20). The theoretical background that Bernheim and Rangel have in mind is associated with “neuroeconomics”, i.e. to the joint collaboration of neuroscience and behavioral economics that emerged at the end of the 1990s.1 It is more precisely the neurophysiology of reward-processing behavior that has implications for welfare economics, as we shall see later in greater detail. This experimental literature is also known as “neurocellular economics” (see Ross, 2008)2 or “Glimcher-style” neuroeconomics.3

3The logic of Bernheim and Rangel’s proposition is roughly the following. Neuroscience offers an improved understanding of addictive behaviors. Addicts are unable to maximize their own subjective utility. Hence, neuronal models of addictions have implications for welfare economics and policy design. Addiction is therefore the key concept that “bridges the gap between neuroscience and public policy”, as Bernheim and Rangel put it. From an economic perspective, addiction represents an empirical violation of expected utility theory (EUT). What is irrational in addictive behavior is its impulsiveness. Addicts often “reverse” their preferences when they face the possibility of consuming an addictive substance.

  • 4 See Bernheim and Rangel, 2007.

4There is a clear proximity between Bernheim and Rangel’s perspective and the well-known notion of “libertarian paternalism” developed by Sunstein and Thaler (2003). According to the latter, experimental evidence coming from behavioral economics also suggests that individuals may sometimes act impulsively. As a result, they propose a new kind of policy intervention, dedicated to improving welfare, through the regulation of individual self-control. Both groups of authors develop the same idea of applying the behavioral analysis of irrational behavior to welfare analysis. In a subsequent article, Bernheim and Rangel precisely defined their approach as a “behavioral welfare analysis” (BWA).4

5There is however a crucial difference between these two approaches. The theoretical justification for libertarian paternalism merely lies in the behavioral evidence of addictions, whereas Bernheim and Rangel’s proposal is grounded in experimental neurophysiology. Bernheim and Rangel start with neurophysiological considerations to end up with policy recommendations. “Glimcher-style” neuroeconomics on the one hand, economic policy on the other hand. Since neurocellular economics merely explains how agents behave, but does not say how they should behave, it is necessary to make additional normative assumptions in order to cross the border between neurocellular and welfare economics. In other words, it is necessary to categorize behavior as rational and irrational, in order to promote welfare-enhancing policies.

6One could argue that Bernheim and Rangel’s logic is precisely that economic theory (associated with expected utility theory) provides a clear normative definition of rationality. The axiom of transitivity is indeed able to rule out certain behavior as “irrational”. If the subject “reverses” his/her preferences and acts impulsively in the lab, he/she will be “classified” as irrational. It is because neuroscience offers a positive theory for addictive behaviors, and because these behaviors are irrational from an economic point of view, that one can cross the line between neurophysiology and welfare economics. According to Bernheim and Rangel, neurocellular economics applied to the specific case of addiction has policy implications.

  • 5 The core of Sunstein and Thaler’s argument is that choice frames greatly affect the choices made by (...)

7I will argue that such a definition of rationality is in fact too weak. It leads in particular to an inflated use of the concept of addiction, as we shall see later in greater detail. Some impulsive behavior may not necessarily be regarded as truly irrational: sexual addiction, shopping addiction, addiction to video games, etc.). The problem concerns the distinction between mere impulsiveness and “true” addictions. A normative criterion needs therefore to be defined, in order to restrict the class of “acceptable” irrational behaviors. What is finally lacking in the libertarian paternalism is a clear normative definition of rational and irrational behavior.5

8The central purpose of this paper is to demonstrate that such additional normative requirements are implicitly provided in Bernheim and Rangels’ framework by medical semantics. Neurobiological models of decision-making do not merely produce additional proof of the existence of “addictions”, but are also heavily influenced by therapeutic concerns. As a result, neuroeconomists take medical definitions of sane and rational behaviors for granted. In other words, the basic fact that some people may be regarded as “addicted” and irrational is taken for granted and frequently not discussed. On the basis of Canguilhem’s The Normal and the Pathological, I will try to demonstrate that neuronal models of addictive behavior are ultimately grounded in a normative distinction between the Normal and the Pathological.

9This is where the present paper intends to be critical. Bernheim and Rangel do not account for the importance of medical semantics in their arguments. According to them, it is rather the economic theory of choice that provides neuroscience with normative standards. I will argue, on the contrary, that neuronal models of rationality are both prescriptive and positive from the beginning, because of the “medical assumption” that addictions are pathologies of decision making and need to be treated. The positive criterion of transitive preference is less fundamental. I will thus try to characterize the methodology of neuroeconomics as a pathological paradigm.

10The paper is structured as follows. The first two sections are dedicated to the “improved neuronal understanding of addictions” that backs up Bernheim and Rangel’s proposal. Section 1 details the emergence of neurobiological research on reward-processing behavior. The main result of these early neuroeconomic experiments is that intransitive behavior is the result of reinforcement learning mechanisms encoded by the “reward system”. Section 2 focuses on the specific case of addiction. Neuroeconomic theory conceptualizes addictive behavior as self-control pathology. This particular view on addictive process has therapeutic implications. Section 3 explains how neuroeconomics could be used as a diagnostic tool in the so-called “neurocomputational psychiatry”. In Section 4, I will finally interpret the methodology of neuroeconomics in the light of Georges Canguilhem’s The Normal and the Pathological. I will suggest that the power of normative neuroscience relies on the objective evidence of disease.

2. The Neuroscience of Reward-Processing Behavior: Are Intransitive Preferences and Hyperbolic Discounting “Rational”?

11This section describes the neurophysiology of reward-processing behavior of the past 20 years. This research program has established that intransitive preference orderings and hyperbolic discounting are “rational” in a biological perspective. These behavioral phenomena are indeed grounded in neurological mechanisms. This may be seen as surprising for an economist, since intransitivity is regarded as a violation of the economic theory of choice, i.e. expected utility theory (EUT).

12What is important here, however, is not the mere observation of these empirical violations. A robust theory is also needed to explain the origins of this anomaly. As Guala points out, “behind a counterexample there must be a rival theory and behind a rival theory there must always be a rationality principle […]. In order to consider a (normative) theory as falsified, we do not need a fully-developed rival theory, a promise is enough. In contrast, an isolated counterexample is not enough: we need at least a “sketch of the new theory, a rough conjecture about the origins of the anomaly” (Guala, 2000, 73). From this perspective, Bernheim and Rangel (2005) do not consider neuroeconomic models of addictions as direct rejections of EUT. Neuroeconomics is rather regarded as a positive theory of a phenomenon that has to be viewed as a counterfactual in economic theory of choice.

  • 6 As for neurobiologists, the term “paradigm” has a restricted sense, which means “experimental proto (...)
  • 7 For a general introduction to the machine learning algorithms mostly used in neuroeconomics, see Su (...)

13To understand how neuronal models of the reward system may be related to EUT, the paradigm of these experiments needs to be detailed.6 Their protocols have mainly been inspired by optimal foraging theory and evolutionary biology. Optimal foraging theory (OFT) is based on the idea that animals behave in such way as to maximize their net energy intake per unit of time. In other words, they forage in order to capture and consume food containing the most calories, while expending the least amount of time possible in doing so. To maximize their nutritive rewards, animals develop various strategies (for instance: exploring new areas in search of new prey, or waiting in the most profitable hunting areas). If the problem is specified in mathematical terms, these strategies can be modeled through learning algorithms borrowed from machine learning theory.7

14Reward-processing protocols of early neuroeconomic experiments are a basic application of OFT. Typically, they consist in repetitive visual-motor tasks. Monkeys are trained to perform a visual saccade (i.e. rapid eye movement) in response to light cues. The amount and frequency of rewards (usually, fruit juice) received by the monkey are varied in each trial, depending on the location of the saccade. In the long run, some decisions (the correct saccade for instance) are more rewarding. This allows assessing how a monkey learns to adapt its successive decisions from past rewards. Indeed, with each choice, the monkey has to estimate which action will yield the largest reward. To do so, it must therefore have attributed previously an expected desirability for each available course of action, on the basis of past decisions. This implies that the monkey has adopted a learning strategy, i.e. a feedback mechanism that specifies the correction to be made of future expectations, on the basis of past rewards and losses.

15In OFT, the optimal strategy (the one that yields the largest aggregated reward in the long run) can be specified using reinforcement learning theory (Sutton and Barto, 1998). Similarly, in reward-processing experiments, the best learning algorithm can be deduced from the same method. Reinforcement learning theory considers individuals as hedonistic learning systems, i.e., as Sutton and Barto put it, as a “learning system that wants something”, whose behavior, might be described in terms of “the language of reward and punishment” (Sutton and Barto, 1998, 15).

  • 8 The computation operated by these neurons in visual tasks is not correlated with the amplitude of t (...)

16The great discovery of early neuroeconomic experiments was that the dopaminergic system does indeed work as a hedonistic learning system. Evidence from neuronal firing rates shows that reinforcement learning algorithms are actually encoded and computed in the brain. Using neurophysiological measurements of dopamine neurons in monkeys, Wolfram Schultz established in particular that the activity of these neurons was correlated with the amplitude of a juice reward, when this reward was unexpected. This was interpreted as a “prediction error signal” that encodes the difference between the amplitude of an obtained reward and its prior, expected value (Schultz, 1986).8

17The next decisive step was taken by Paul Glimcher and Michael Platt in the 1990s (see Platt and Glimcher, 1999). Still working with monkeys, Glimcher and Platt showed that the activity of neurons of the lateral intraparietal (LIP) area involved in the similar visual motor task correlated with the expected value of rewards, i.e. with both their magnitude and their probability. These experiments thus gave birth to the idea that the dopaminergic system was a general valuation mechanism working on trial and error processes. These valuation processes could be formalized through reinforcement learning theory, as we have seen before.

  • 9 For instance, in their 1999 foundational article, Michael Platt and Paul Glimcher wrote: “here we d (...)

18However, by mixing reinforcement learning, OFT and the ambiguous notion of “expected reward”, this approach created an important confusion, and initially led to a common misunderstanding about neuroeconomics. The misunderstanding consisted in viewing these experiments as proof that the dopaminergic system works as EUT predicts, since it encodes the expected value of future rewards. Consequently, neuroeconomics was frequently understood as an “economic-style” approach within neurophysiology.9 Still, EUT does not provide any normative standard for the analysis of such learning tasks. It only tells us that in the long-run, the aggregate reward must be maximized, but it does not explain the process of maximization over discrete choices.

  • 10 For instance, Glimcher himself proposed in his 2003 influential book on neuroeconomics that economi (...)
  • 11 Once again, even if the dopaminergic system actually does compute a quantitative desirability for e (...)
  • 12 Glimcher himself admitted his mistake later: “The paper [see Glimcher and Platt, 1999] goes on to d (...)

19Neuroscientists used to associate the loose term of “economic theory” with reinforcement learning mechanisms. The misunderstanding was clearly the result of the loose and vague semantics used by neurobiologists themselves.10 In particular, the expression of “physiological expected utility” used by Glimcher et al. (2005, 253) was probably misleading, because it gave the impression that expected utility theory could be directly translated into experimental neuroscience as a normative ideal.11,12

  • 13 See Herrnstein, 2000.

20This is not to say, as we shall see later, that biological and neural findings have no relevance for economic theory, but there was clearly a confusion about the economic “flavor” of such experiments. Perhaps the confusion was a positive heuristic for neuroeconomic research. It somehow led to the intuition that the study of reward-learning behavior could reveal something about the empirical validity of EUT. One of the main results of foraging theory is indeed the matching law, formulated by Richard Herrnstein.13 It states that the expected utility of an action is perceived by an animal as the expected utility of that action divided by the sum of the expected utilities of all available actions. In a basic task involving successive choices between two actions, the matching law implies:

21NA/NB = E(A) / E(B)

22where A and B represent the number of choices of actions A and B respectively, and E(A) and E(B) represent the expected amount of reward received from actions A and B respectively.

23There is a clear proximity between expected utility maximization and the notion of “matching law” in optimal foraging theory. But, once again, one should carefully distinguish between the kind of theoretical problems involved here and EUT. What interests neuroeconomists and foraging theorists is not the matching law itself, but matching-law behavior, i.e. the way animals learn over time to select the most rewarding actions and to behave according to the matching law.

24Still, there is more a complex version of the matching law, formalized by Rachlin and Green (Rachlin and Green, 1972), that has direct consequences on EUT. Subsequent neuroeconomic experiments in the 2000s extended the basic paradigm of reward-processing tasks to intertemporal choice, in which experimentalists introduce a delay between choices and the receipt of reward. As a result, rewards vary in expected amount, frequency and in delay. As will be shown below in greater detail, it has been established in these experiments that the dopaminergic system also encodes the temporal dimension of expected rewards. The matching law thus becomes:

25NA/NB = [ E(A) / E(B) ] x [ DB / DA ]

26where DA and DB respectively represent the delays between choice of actions A and B and the receipt of rewards.

27This equation brings us to the larger issue of preference reversal and hyperbolic discounting. Indeed, this extended version of the matching law reveals two opposing processes. Optimal foragers display a tendency to prefer larger expected rewards, but the second process leads them to prefer quicker rewards. What is crucial is that the preference between the two actions depends on the temporal distance over which the choice occurs. In particular, when it is immediately available (D = 0), a small reward might be preferred to a larger, later one.

  • 14 See for instance Breiter et al., 2001.

28This experimental result was observed in monkeys. The last and decisive step was thus to extend this experimental paradigm to human subjects. At the start of the 2000s, the rise of neuroimaging allowed similar reward-learning tasks to be applied to humans, whereby participants were placed in a scanner and asked to choose between financial lotteries.14 Monetary rewards replaced gustatory ones (juice), but similar algorithms of hedonistic leaning could be used to characterize the pattern of activation of the dopamine circuit in humans. Correlatively, the same phenomenon of preference reversals is observed.

29Herein lies the explicit connection with EUT. The fact that neuronal firing rates (in microelectrode experiments on monkeys) or BOLD signal (in fMRI experiments on humans) are correlated with the predicted values of rewards does not mean that the brain of monkeys encodes expected utility directly. Yet, consideration of learning process does explain why animals may change their preferences over time and hence behave “irrationally” in EUT sense. As Ross et al. put it, “this research derived from the matching law has revealed that irrationality, in the technical sense of the economist, is in fact normal. This is critically important and worth repeating: matching law research reveals that both people and other animals are naturally disposed to have inconsistent preferences for future rewards as times passes toward the availability of those rewards” (Ross et al., 2008).

3. What is Wrong with Addicts? Preference Reversal and Self Control

30The previous section established that preference inconsistency is “rational” from a biological perspective. Impulsiveness towards immediate rewards is part of our biological constitution. Electrophysiological and neuroimaging techniques show that it is rooted in neuronal mechanisms. It seems intuitively reasonable to consider that biological fitness imposes a minimal degree of impulsiveness, in order to react quickly and take advantage of changes in environmental contingencies. Still, is it always “rational” to reverse preferences? The story is different, however, when individuals show measurable attenuation in prefrontal regions that, in healthy people, inhibit the direct control of consumption choices by the dopamine system. Such pathologies of self control clearly result in irrational choice patterns. This section aims to explain how these neuronal dysfunctions play an essential role in addictive behavior. The main issue here is therefore to distinguish between “rational” phenomenon of preference inconsistencies and addictive/irrational patterns.

  • 15 Counterfactual judgments and “choiceless utility” also explain why it is sometimes rational to reve (...)

31Before examining this question, I shall insist once again on the fact that the existence of a “biologically rational” degree of impulsiveness has not to be taken as a proof against the validity of EUT, as I have argued before. Preference reversals observed in reward-processing experiments merely reflect the fact that leaning is based on trial and error processes. The notion of “prediction error” precisely involves the idea that errors are necessary to improve future predictions. Moreover, the phenomenon of hyperbolic discounting has been extensively studied by economists and theoretical models have been proposed to accommodate preference reversal and EUT maximization.15.

32As a result, it is rational to make occasional mistakes in reward-processing experiments, provided aggregated utility is maximized, in the long run. In experimental tasks involving successive monetary gambles, “pathological” decision makers can therefore be separated from rational ones, on the basis of behavioral performance. The efficiency of the algorithm to maximize the aggregated pay-offs can in principle decide whether individual attitudes toward reward foraging are optimal. As Camerer puts it, “in a game, there is a clear performance metric–who makes the most money?” (Camerer, 2008, 205). In reward-processing games where there is a clear performance metric (who makes the most money?), sub-optimal decision making can hence be understood as a quantitative variation over and under the two key parameters of the optimal learning procedure.

  • 16 In particular, these pathological players include patients with prefrontal cortex lesions.

33Such “pathological” decision-making abilities can be revealed in humans in financial games, resembling reward-processing experiments. For instance, in the well-known Iowa Gambling Task (IGT), participants are presented with four decks of cards. They are told that each time they pick a card they will lose or win some money. The goal is to win as much money as possible. Two decks are “bad decks”: they lead to exceptionally large gains but to losses over the long run; “good decks” are associated with smaller but more regular gains and therefore lead to gains over the long run. Most “healthy” participants initially sample cards from each deck, and after 40 or 50 selections are fairly good at sticking to the good decks. However, some “pathological” players continue to persevere with the bad decks, sometimes even though they know that they are losing money overall (Damasio, 1995).16.

34What is exactly “pathological” in the decision-making abilities of such players? In reinforcement learning theory, an optimal algorithm is characterized by two parameters. The first one concerns the error prediction signal: it specifies the way in which rewards experienced in the past and punishments are taken into account in order to improve future predictions. For instance, some agents will radically change their expectations after experiencing an important loss, whereas others will keep using the same strategy. Obviously, optimal learning is located at an intermediate level between these two extremes. “Pathological players” in the IGT obviously belong to the former category. What is striking in IGT tasks is that “pathological individuals” appear to be as sensitive as “healthy” ones to financial gains and losses (see Bechara et al., 2000). But what is problematic is learning: such individuals do not improve their predictions through past experiences of reward and punishment.

  • 17 Such pathological behavioral pattern can be specifically exhibited using the Wisconsin Card Sorting (...)

35Learning inabilities and a greater tendency toward “perseverative” errors are thus a first basic marker of irrational, reward-processing behavior.17 The second parameter of the optimal learning algorithm specifies the adequate degree of “boldness”, i.e. the tolerance that it is necessary to have in face of errors or prediction. In technical terms, this parameter is called the balance between exploration and exploitation. As Sutton and Barto put it, “to obtain a lot of reward, a reinforcement learning agent must prefer actions that it has tried in the past and found to be effective in producing reward. But to discover such actions, it has to try actions that it has not selected before. The agent has to exploit what it already knows in order to obtain reward, but it also has to explore in order to make better action selections in the future. The dilemma is that neither exploration nor exploitation can be pursued exclusively without failing at the task. The agent must try a variety of actions and progressively favor those that appeared to be the best” (Sutton and Barto, 1998, 4).

36The balance between exploitation and exploration characterizes an agent’s ability to elaborate complex cognitive strategies. Indeed, in reward-processing experiments, in order to be an optimal learner, one has to compare not only the direct/immediate consequences of actions, but also to consider the potential value of an action toward the aggregated value of rewards. For instance, a loss might have a positive value since it helps in the long run to improve predictions. Thus, an optimal learning strategy includes a long-term time horizon. In the IGT for instance, the choice between the four decks does not have to be perceived merely as a choice between four immediate or quasi-immediate rewards or losses. A much more complex perspective has to be taken to maximize aggregated rewards. This result is well known in behavioral economics: it is related to the problem of “framing” choice sets.

  • 18 For a full description of PF cortex damaged patients behavior, see Damasio, 1995.

37Hence, optimal learning requires a good memory and self-control, i.e. the ability to resist gratification over immediate rewards. These two issues are related, because the main problem here can be understood as a “temporal” problem in reward-processing. Pathological players in the IGT usually both fail to keep records from outcomes of past actions and to take a long-term perspective on future rewards. Typically, neuroeconomic experiments use patients with prefrontal cortex (PFC) lesions to exhibit such behavioral patterns. Indeed, neuropsychological studies of the PFC damaged subjects in the 1980s established that this brain area is responsible for self-control abilities. PFC damaged patients are unable to have a normal social life and act on purely impulsive grounds, without taking into account complex consequences of their actions in the future or on others.18

38Therefore, neuroeconomic evidence suggests that lack of temporal perspective, both in the past and in the future, in cognitive tasks resembling the IGT can fundamentally be understood as a unitary neuropsychological phenomenon. First, from a psychological perspective, it is related to “reward bundling”. As Ross et al. put it, a rational decision-maker “bundles a whole series of future benefits and puts them at stake in the present moment” (Ross et al., 2008, 71). But “to be effective a personal rule must be remembered, and the series of individually non-impulsive choices it describes must promise more reward than the series of impulsive choices starting with the current threat to the rule. This, like much executive cognition, requires reconstructing and maintaining representations of goal states including considerations that are independent from the immediate context” (Ross et al., 2008, 111). What is also worth noticing is that this psychological ability has a clear neural signature: impulsivity results from the attenuation of frontal inhibitory circuits that is measurable (Ibid., 17).

39The IGT framework is an abstract formalization of impulsiveness and can be used, as we shall see in the next section, as a diagnostic tool in neuropsychiatry. But its scope is limited to monetary rewards: IGT players face only financial gains and losses. The neurobiological concept of rewards has an extensive use: it extends to basic rewards (food) and to symbolic ones (money for instance), but also to long-term gains, and to “social” rewards (pleasure to help, to give, generosity, altruism, etc.). The fact that the dopaminergic system encodes various forms of reward is known in the literature as the “common neuronal currency” (Montague and Berns, 2002).

40There is consequently no reason to consider only impulsiveness toward financial rewards. Self control pathologies may arise in any type of valuation. More importantly, detecting impulsiveness and lack of self-control is only a part of the story. It does not explain why some individuals behave in such a pathological way: why are these people unable to learn from previous failures and to consider long-term consequences of their actions? Is it the result of purely neurological pathologies, as PFC-damaged patients suggest?

41This is where the neuronal understanding of addictions provides interesting explanations. First, it appears that addicts also exhibit impulsiveness and lack of self-control. But these characteristics can be understood as particular consequences of a general model of the addictive process. Moreover, the neuroeconomic concept of addiction is restricted neither to “substance abuse”, nor to financial reward-processing behavior. As a result, it has been proposed to merge various neuroeconomic studies on impulsiveness and self-control in a larger theory of addictive processes (see Ross et al., 2008).

42The neuroeconomic view on addiction remains embedded in reinforcement learning theory. From this perspective, addiction is related to a “self-stimulation” of the reward system. A key element to understanding what goes wrong with such individuals is that the dopaminergic system, as learning proceeds, progressively reacts not to the receipt of rewards, but to their predictors. This is the way the reward system becomes predictive. To start with, dopamine release is the result of an unexpected reward (see Schultz, 1986). But once a signal becomes associated with the prospect of a reward, one can observe activations both in the “prospect phase” (prediction of reward) and in the “outcome phase” (error prediction: difference between the prediction and actual amount of reward received) (see Breiter et al., 2001). In reward-processing games, such cues are typically the ones that are used to initiate a task: luminary cues in experiments with monkeys, visuals of financial lottery on a computer screen in humans.

43The prospect of reward has nevertheless to remain minimally uncertain, in order to generate a dopamine release. If the reward is delivered quasi-immediately with a 100% probability, there is no need for a prediction at all. The subject has to represent the reward merely as potential, either in probability or/and delay. This is the reason why there is a clear separation, at the neuronal level, between reward prediction and feeling of hedonic pleasure. The dopaminergic circuit encodes the expected amount of reward that will be received, and fMRI experiments have proved that actual or effective pleasure is encoded in a separate part of the brain (see Berns et al., 2001). In the neurobiological literature, this is known as the difference between the “liking” and the “wanting” systems (Berridge, 2007).

44This explains why the main pleasure of a gambling addict lies not in the pleasure of winning money, but rather in the mere excitement of seeing the dice roll. The problem with such individuals is that they find “reliable cues for surprise”, as Ross et al. put it. They face signals that reliably announce the prospect of uncertain rewards (and probable punishment). This is the necessary condition to induce a dopamine release. For instance, the reward system of gamblers will be activated when they see the dice rolling, as long as they assume that the probability of winning is above zero. The process of choosing becomes pleasant in itself: “what leads people to have a problem with gambling is that it is rewarding in itself” (Ross et al., 2008, 165).

45Addicts are hooked on surprise: “what gambling fundamentally consists in is paying for the possibility of a surprise…gambling simply is direct stimulation of the midbrain reward system” (Ross et al., 2008, 165). The main characteristic of addiction is thus the existence of “perpetual positive prediction errors” (Ross et al., 2008). In gambling addiction, the regular administration of surprise “self-stimulates” the dopaminergic system. Surprise is the equivalent of addictive substance in drug addictions: heroin, cocaine, etc. The chemical properties of such drugs have indeed the power to generate a similar “artificial” dopamine release.

46This is not to say, however, that addiction involves any malfunctioning in the dopamine system itself. The reward system in addicts still exploits consumption opportunities and computes reward valuation in a way that is coherent with the reward-structure of the environment. The problem comes precisely from the internal perception of the environment. Focus on reward-predictors signals allows addicts to see their environment as particularly reward-rich, because they experience recurrent surprise (that boosts phasic dopamine signals) regardless of what actually happens. The brain responds to this by reducing inhibitory prefrontal synapses that prevent the dopamine circuit from directly controlling motor responses.

47Gamblers and drug addicts therefore exhibit the same dependence towards the predictors of reward. Ross et al. propose viewing “the intrinsic chemical properties of drugs as special-case distractors from the fundamental structure of addiction” (Ross et al., 2008, 164). To sum up, the basic structure of addiction is the following:

    • 19 See Ross, forthcoming, 39: “The neuropsychological account of the way in which the striatal reward (...)

    Addiction has a clear neural signature: inhibition of prefrontal regions that usually control dopamine system responses. That neuronal phenomenon is measurable. It is a necessary (but not sufficient) condition for the distinctive pattern of addictions.19

  1. At the behavioral level, addiction is mainly the result of the interaction with an “addictive environment”. The reward-structure of such environments is the following. First, easily-accessible rewards are associated with “reliable” predictor cues. Second, the prospect of rewards has to remain uncertain and/or at least ambiguous. As a result, addicts are obsessed with a particular type of predictor cues that is “reliably” associated with a potential reward.

  2. (i) and (ii) have two psychological effects that addicts experience as strongly aversive, i.e. that cause objective suffering in addicts:

    1. addicts have difficulties concentrating on any reward cue that does not predict consumption of the addictive substance. Their behavior becomes obsessive.

    2. because of motor system stimulation caused by dopamine, addicts suffer from cravings when consumption does not or cannot occur.

4. Neurocomputational Psychiatry: Treating Addictions as Pathologies of the Reward System

48My central claim is that neuroeconomics models addiction in a manner that is both normative and descriptive from the outset, because it is driven by medical concerns. Still, Bernheim and Rangel consider the neurobiological approach as purely descriptive. According to them, the “normative part” of neuroeconomics is provided by reinforcement learning and behavioral economics. Before coming to my main argument in the last section, I shall here take the perspective of Bernheim and Rangel to explain how economic theory might be applied to therapeutic problems arising in neuropsychiatry. This has not to be taken as a mere illustration. The purpose is to follow the full logic of Bernheim and Rangel’s argument, before assessing its validity.

49From the point of view of neurobiologists, the introduction of reinforcement learning theory was as an important medical innovation. It allowed the traditional verbal description of pathologies used in neurophysiology to be substituted by quantitative models. This reinforcement learning approach in neuropsychiatry has been called “computational neuropsychiatry” by Rangel, Camerer, and Montague. The basic idea behind computational neuropsychiatry is to use the reward-processing paradigm in order to produce an objective diagnosis of mental-disease, and in particular of impulsive and addictive behaviors: “Computational models of reinforcement learning provide a new language for understanding mental illness and a starting point for connecting detailed neural substrates to behavioral outcomes. For example, reinforcement-learning models predict the existence of valuation malfunctions, in which a drug, a disease or a developmental event perturbs the brain’s capacity to assign appropriate value to behavioral acts or mental states […]. This presents the possibility of generating a new quantifiable taxonomy of mental-disease states.” (Rangel et al., 2008, 554)

50What is important here is to understand how the reward-processing paradigm could be used to assess human decision-making abilities. The first and more basic method consists in using financial protocols resembling the IGT. Developed in the 1980s by Antonio Damasio, the IGT was specifically conceived to provide a neutral psychometric evaluation of a neuronal deficit: according to Damasio, it is “the first laboratory diagnostic process for patients with ventromedial PF cortex damage–a rather useful advance, given that these patients generally passed all other neuropsychological tests and only exhibited their defects in real life and real time” (Damasio, 2008, 211).

51Behavioral performance in the IGT allows “pathological” ways of behaving when faced with uncertain rewards and punishments to be detected. Here, it seems that the existence of neuronal deficits can be inferred from a behavioral criterion: the tendency to perseverative errors. For instance, Daniel Houser et al. have observed patients with damaged pre-frontal (PF) cortexes and sane subjects playing the IGT in order to formalize their respective learning procedures. The authors intended to explain clearly what was wrong in the gambling behavior of these patients, without having to understand their pathology by a traditional medical diagnosis: “The idea behind the procedure is to group subjects according to similarities in their decision-making behavior, regardless of any known physical abnormalities they might possess.” (Houser et al., 2005, 375)

  • 20 For a detailed account of those criticisms, see Ross et al., 2008, 97.

52The use of IGT-like experiments seems however to be limited. First, interpretations of behavioral performance with the IGT have been largely criticized.20 Parts of the protocol are unjustified: why for instance use four decks of cards and not two? Such arbitrariness is not legitimate, since behavioral performance is radically modified, for example, when the frequency of rewards is altered in good decks. More importantly, as shown in the last section, the IGT is restricted to impulsive behavior concerning monetary rewards.

53In contrast, the broader scope of the neuroeconomic theory of addictions extends to all kinds of reward-processing troubles. As seen in the last section, the main problems of addictive behaviors are that they yield perpetual positive prediction error signals. The addicts learn to be “reliably surprised” and react impulsively to the predictors of a very specific class of rewards. The neuroeconomic diagnosis of addictions raises two related issues. First, a specific type of behavior (drug consumption, gambling, eating behavior, etc.) has to be declared as potentially addictive. The environment has to offer explicit and reliable associations between perceptual incentives and prospects of easily accessible rewards. This is the “conditioning effect” of an addictive environment.

54Still, the prospect of rewards has to remain uncertain, or at least ambiguous (in the sense of involving a future loss). In gambling addiction, there is a possibility of punishment. The prospect of financial loss yields perpetual surprise when an effective reward is obtained. This type of behavior is known as “loss chasing”. Another way to do this is to associate the administration of reward with a sure (long-term) loss. For instance, in food addiction, unhealthy and appetizing food is both thought of as a reward and as a long-term loss. Items of tempting food can be pictured as a mix between immediate self-gratification and long-term loss. Anorexics “know that it is wrong” to eat such a food, and this is probably the reason why they are so strongly attracted to it. Similarly, an obese person who follows a diet knows that gustatory rewards involve a long-term cost when he/she indulges himself/herself with unhealthy but appetizing food. This is where the neuroeconomic criterion for addiction becomes hazy, as we shall see later in greater detail. The environment is not merely the sphere in which Pavlovian reflexes are developed, but it is also a convenient location for “loss chasing behavior”.

55Second, for a given addictive type of behavior, addicted individuals can be separated from “rational” decision-makers on the basis of their impulsiveness. When they face “addictive cues”, addicts radically overvalue addictive courses of action relatively to other types of rewards, in particular long-term ones. Here, both behavioral and neuronal data are necessary to assess impulsiveness. In particular, behavioral results on the “liking system” can be contrasted to neuronal activations associated to the wanting system. When data display a clear disconnection between what the participant “likes” and “wants”, the subject will probably be unable to resist his/her impulsiveness to immediate/addictive behavior. The ability to take a long-term perspective on rewards and punishment (“reward-bundling”) is also a useful indicator of self-control.

56Eating behavior is, for instance, frequently regarded as potentially addictive. Experimental tasks involving successive choices between healthy/unhealthy foods can be used to diagnose food addictions. A recent study on self-control conducted by Hare et al. is a good illustration. The participants were 37 obese persons who followed a diet. In the first phase, they had to rate 50 food items for taste and health separately. On the basis of these ratings, an item was selected for being neutral in both taste and health. Then the subjects had to choose between each of the food items and the neutral item. Participants also had to indicate the strength of their decision by using a five-point scale, in which respondents had to indicate the relative value for eating a particular type of food, instead of the reference item. Behavioral results classified subjects who made their decision on both health and taste as “self-controllers” and the participants who made their decision on taste only as “non-self-controllers”. fMRI data correlated with behavioral results. The relative value of food items appeared to be encoded in the ventromedial PFC. Moreover, self-control choices (i.e., choosing the neutral item instead of an unhealthy but tasty one) involves the modulation of that value signal by the dorsolateral PFC. Such a modulation was not observed on non-self-controlling subjects (Hare et al., 2009). As a result, in potentially addictive behavior, the diagnosis of impulsiveness could be based on quantitative approaches using such correlations between behavioral and neuronal data.

5. The Methodology of Experimental Physiology: the Role of the Pathological

  • 21 Canguilhem, [1966] 2005.
  • 22 For the sake of clarity, I will not provide a complete review of the works of Canguilhem here, but (...)

57Computational neuropsychiatry applies quantitative-algorithmic approaches to neurophysiological issues that were usually conceived in “pathological terms”. It aims to evacuate the need for the verbal description of clinical pathologies, on the basis of an objective definition of rationality. Rational decision-making is indeed referred to in each protocol as EUT-consistent behaviour. For instance, in the IGT, rational decision-makers can be easily distinguished by their ability to maximize aggregate rewards; in potentially addictive environments, rational individuals are the ones capable of “reward-bundling” and taking long-term perspectives. My claim, however, is that such a definition of rationality is not as neutral and objective as neuroeconomists usually think. The concept of rationality in neuroeconomics is implicitly normative, in the sense that it states how decision-makers should behave in order to be rational. More precisely, neuroeconomics can be interpreted in the terms of Georges Canguilhem’s The Normal and the Pathological as a “pathological” paradigm in behavioral science.21,22 Using Canguilhem’s perspective on experimental physiology, I suggest that the evidence of pathologies plays a founding role in the neuroeconomic analysis of behavior, which is, as a result, heavily influenced by medical categories and semantics.

  • 23 The terms “implicit” and “explicit” methodologies do not appear in Canguilhem’s vocabulary. Once ag (...)

58Computational neuropsychiatry can be understood as the application of quantitative models in experimental physiology. This approach is what Canguilhem calls the “Broussais principle”, which is associated with Claude Bernard’s methodology. According to Canguilhem, experimental physiology has two methodologies, one that is explicit, and the other implicit.23 The Broussais principle belongs to the assumed/explicit methodology. It states that there is a quantitative homogeneity of pathological and normal phenomena: diseases are considered as quantitative variations outside (below and above) the range of healthy states. To this end, normality is reduced to average, regular, mean characteristics: physiological variables are “qualified as normal because they refer to average characters” (Canguilhem, 1966, 75). For instance, physiologists may define “normal” body temperature as the average: 37° C.

59From this perspective, physiological variables are manipulated as statistical variables. This approach raises two related issues:

  • First, what is lacking in a purely quantitative analysis is the explanation of why the Normal should be referred to as the average: what if most people were actually anormal? If “normal” body temperature were 38° C, then average individuals would be “too cold”.

  • More importantly, the second issue concerns the definition of the standard deviation. Even if one could justify the first assumption –normality is defined by the average– one would then need to establish the threshold beyond which a variable could be characterized as pathological. What is the amount of variation over and under the normal average state that can be tolerated? Is it still “normal” to have a body temperature of 37.1°, 37.5°, or 38° C?

  • 24 Once again, the term “ethics” means the ability to form a normative judgment about what is a good, (...)

60Canguilhem, however, is not saying that experimental physiology has no legitimacy as a science. The answers to these questions require taking a normative perspective on life: “From his/her objective point of view, the scientist only sees anomalies in statistical differences, ignoring the fact that the interest of the scientist first was aroused by a normative difference.” (Canguilhem, 1966, 85) The Normal is always a prescriptive concept because it plays the role of an implicit “ideal in the normative activity of therapeutics”. It is associated with ethics: “A human characteristic is not normal because it is common, but common because it is normal, i.e. normative in a given type of life.24” (Canguilhem, 1966, 102)

61Therefore, the validity of quantitative analysis of the behaviour of living organisms only depends on a normative pre-definition of normality. According to Canguilhem, this definition relies on the objective experience of pathologies: it is because some individuals suffer from pathologies and need to be treated that a normal state can then be defined as the average state of “sane” individuals. The experience of fever legitimates for instance the diagnosis of a pathological body temperature. It is hence the experience of disease by a patient that elicits the need for a medical knowledge: “There is nothing in [medical] science that did not first appear in the consciousness of a patient.” (Canguilhem, 1966, 56) Pathologies can be explained with reference to a “normal state” taken as the average, only if one is first of all capable of understanding how the whole personality of the patient is altered qualitatively: “In order to establish a diagnosis, one needs to observe the behavior of the diseased.” (Canguilhem, 1966, 149)

62The definition of normal states in experimental physiology can be summarized in the following way:

    • 25 The term “ethical” here has therefore to be understood as a particular form of behaviour and conduc (...)

    First, this definition is implicit, in the sense that experimental physiology is explicitly based on a purely quantitative methodology. However, as we have seen above, this approach blurs the qualitative and ethical distinction between the Normal and the Pathological: “being sick for humans is really living another life” (Canguilhem, 1966, 49).25

  • Secondly, the definition of normality is negative, in the sense that it relies on a prior identification of possible pathologies. It is because people with a body temperature below 36.5° are objectively “too cold” that one can say that the normal temperature should be above 36.5°; but there is no direct definition of normal body temperature. As Canguilhem wrote, health, i.e. normal life, is “living in the silence of organs”: the Normal depends on possible violations of the norms. Therefore, the Pathological has the privilege of precedence over the Normal. In experimental physiology, medical techniques and clinical concerns have an anticipatory role in orienting “pure science” toward problems to be solved.

63These issues regarding normal and pathological states can be illustrated through so-called “neurocomputational psychiatry”. Neuroeconomists subscribe to what Canguilhem calls the “Broussais principle”: in computational neuropsychiatry, disease (addiction) only differs from health in intensity. This allows physiologists to consider illness as a quantitative variation outside (above or below) a pre-defined rational range. Nevertheless, the adoption of the Broussais principle – i.e. measuring personal rational abilities in the laboratory instead of describing pathologies verbally – requires a clear performance metric to be established, allowing for a quantitative definition of a rational or optimal behavior. The two main approaches discussed in the last section have to be examined separately.

  • 26 Still, one need to keep in mind that strictly speaking, economic rationality is not equivalent to n (...)

64In financial games such as the IGT, the formal structure of the protocol design provides such a quantitative criterion for individual assessment, based on who makes the most money. Hence, one could argue that, in neuroeconomics, the normative definition of rationality is not necessarily the same as normality in Canguilhem’s sense, i.e. the average. In these gambling tasks, rational behavior is defined as maximizing aggregated pay-off; while normality refers merely to average behavior, which is usually not optimal. However, the semantics used in the neuroeconomic literature are very ambiguous. Most interpretations of the IGT actually consider “rational” decision-makers to manifest average behavior (see for instance Bechara et al., 1994). These issues are not frequently debated by neuroeconomists, and the distinction between rationality, normality and average behavior is not very clear in the literature26.

65More importantly, in a quantitative approach, it is not sufficient to say that rational behavior is the behavior that maximizes the aggregated-pay off, because it is necessary to specify the amount of variation around this optimum in which individuals can still be considered as “rational”. In other words, the behavioral-performance metric does not provide the standard deviation that separates rational and irrational choice patterns.

66These issues are particularly important when assessing the efficiency of a decision-making rule. Rather than being purely quantitative, the neuroeconomic approach relies in fact on the prior acceptance of medical categories. For instance, the study quoted above (Daniel Houser et al., 2005) on PFC damaged patients playing the IGT took for granted the fact that their subjects were diagnosed as “PF cortex damaged patients”, according to neurological classification. The authors also knew that these patients usually display pathological gambling behavior. On these assumptions, they were able to formalize a “pathological” heuristic in the IGT. The deviation of this decision-making rule from the optimal one could thus in principle define the standard deviation that separates the Rational and the Pathological. The IGT can therefore be used subsequently as an algorithmic tool for diagnosis, but only because people whose behavior has been induced to formalize the pathological learning procedure were previously declared as such by clinical neurologists.

67The neuroeconomic theory of addictions also provides at first sight such an explicit objective criterion for rationality. For a given potential addiction, there are clear markers at the neuronal, psychological and behavioral level of impulsiveness and excessive delay discounting. The neural characteristic (attenuation of prefrontal regions activity) is not, however, a sufficient condition in itself for the distinctive pattern of addiction (see supra, footnote 19). It has to be supported with behavioral evidence, i.e. the interaction with a potentially addictive environment. The problem here lies in the definition of “potentially addictive”. As we have seen before, the main characteristic of such environments is that the prospect of rewards is “reliably” uncertain. It can be understood as a basic condition so that the pattern of choices becomes rewarding in itself. Risk becomes rewarding in itself through “loss chasing” behaviors. This behavioral condition is supposed to impose strong constraints on the neuroeconomic definition of addictions. As a result, as Ross et al. rightly point out, impulsive behavior is not necessarily “addictive” in the neuroeconomic sense. For instance, it seems very difficult to consider sexual behavior as addictive, because the prospects of rewards are either too uncertain (in the case of flirting for instance) or too certain (for masturbation) (see Ross et al., 2008, 213).

68My claim however is that “loss chasing” is actually a too weak a condition to distinguish between impulsive and non-addictive behaviors and “truly addictive” behaviors. Let us consider the case of eating behavior, which is nowadays unanimously regarded (and treated) as potentially addictive. Why should eating behavior lead to “loss chasing”? If an individual is offered the possibility to indulge himself/herself with a large hamburger for free, the prospect of reward appears certain and immediate.

69Perhaps eating behavior becomes potentially addictive when items of food are perceived as involving both an immediate gratification and a long-term loss. For this reason, an obese person who follows a diet might be rewarded for the mere decision to take the hamburger (in addition to the pleasure of eating it). In this view, the uncertainty is about the rewarding property of food itself. An anorexic might perceive food as both appetizing (rewarding) and feel at the same time some disgust towards it. This is probably the reason why sufferers of anorexia and bulimia spend so much time cooking, or choosing their meals in restaurant, etc. In doing so, such individuals become obsessed and totally absorbed by predictors of purchase; they usually build their own codified ritual that makes the prospect of food delivery reliable. In contrast, an obese person who does not follow a diet – or just a “regular” individual – will not envisage food consumption as problematic. He/she cannot take pleasure in the process of choosing. “Food addicts” probably have an ambivalent relationship towards food: they love eating and at the same time they consciously know that it is wrong. In the study quoted in the last section about self-control and eating behavior (see Hare et al., 2009), the results would have been probably a lot different if the subjects were obese persons who do not follow a diet.

70It is probably because addicts perceive their consumption as problematic that the prospect of reward becomes controversial. Hence, one needs to go through introspection to check the addictiveness of a given behavior. The point here is not to investigate the psychological profile of food addicts. I am merely suggesting that the notion of “potentially addictive environment” is actually a psychological construct, and not a clear behavioral characteristic. It is a lot more ambiguous than the codified tasks of reward-processing experiments reveal. It can be interpreted in various extended manners and is not so “objective”. For instance, impulsive shopping could be pictured as an addiction. Impulsive shoppers know that spending too much money shopping is “wrong”, in a similar manner to food addicts. They probably experience pleasure not only (and mainly) in the act of buying a pair of shoes, but rather in the process of choosing.

71The behavioral criterion of addiction is therefore not sufficient to distinguish between mere impulsiveness and addictions. The problem is that common sense views of addictions are too vague and general. As Ross et al. put it “addiction in the popular sense of the term is a familiar, culturally tagged syndrome that some people adopt as a norm around which they and others can structure expectations […]; this norm can in principle be constructed around any perceptually and socially salient consumption pattern” (Ross et al., 2008, 161).

72It may therefore be asked how an objective difference can be established between rational and irrational (i.e. addictive) impulsiveness. This issue is tackled by Ross et al., 2008. The authors support the following view: “We do not endorse moralization of people’s preferences that other people avoid dependence on alcohol, cocaine, heroin, or nicotine (let alone marijuana). We can at least agree, however, that these “bossy” preferences themselves have a strong basis in objective fact: the drugs in question almost always cause serious harm to those who become dependent on them, including by reference to the normative standards of the dependent themselves. By contrast, there is no social consensus, and no prospect of a rationally justifiable social consensus, on how much shopping is too much, on how many hours a person ought to spend surfing the internet or playing the online games, or on how much effort, time, and emotional opportunity cost people should devote to various forms of sexual stimulation and gratification.” (Ross et al., 2008, 212-213)

73Two arguments are presented here. The first is that there is an “objective fact”, which is the high “costs” of addictions “by reference to the normative standards of the dependent themselves”. Ross et al. take the example of an individual who likes swimming with crocodiles; such a preference can objectively be said to be “wrong” because this behavior can be potentially harmful for the agent. Concerning addictions, the internal struggle that is experienced at the psychological level causes objective sufferings in addicts: cravings, attention deficits, etc. The fact that this conflicted state is experienced as strongly aversive is the reason why addictions can be counted as counter-normative: “So far as consumer theory is concerned, addictive consumption is not unusual except with respect to the intensity of the addict’s internal struggle between maintenance and change.(Ross, forthcoming, 26)

  • 27 Hence, the argument that “there is no social consensus” on “how much shopping is too much” is not l (...)

74Second, the objective and undeniable nature of that internal struggle results in a large social acknowledgment. Hence, addictive impulsiveness has to be grounded in a “social consensus” (or at least the possibility of building one). Shopping addiction for instance does not appear to be a legitimate candidate for behavioral addiction since there is “no prospect of a rationally justifiable social consensus, on how much shopping is too much”. Still, the psychosomatic experience of craving and attention disorders is logically anterior and independent of its social recognition. It is not directly drawn from a consensus of “how much is too much” that a given impulsive behavior can be regarded as addictive. It is rather based on direct evidence of impulsive individuals who experience internal struggle and are consequently regarded as disordered individuals by others:27 “Almost all drug addicts find their conflicted state aversive in itself, so their friends and relatives—and indeed the state—can readily justify taking sides in their internal bargaining.” (Ross, forthcoming, 36)

75According to Ross et al., the objective proof for addiction relies in the mere psychological fact that addicts endure discomfort and internal struggle. How could such a subjective experience have an objective (scientific) value? Neuroeconomists could argue that the psychological condition (aversive experience) has to be supported with neural evidence, i.e. measurable attenuation in the prefrontal circuit. For instance, shopping addiction could be recognized as a true addiction if and only if the distinctive neuronal patterns of addiction were found in people who complain that they experience obsessive shopping urges while trying to do other things. The shutting down of prefrontal inhibitory regions does not reflect, however, objective suffering, but rather impulsivity toward short-term rewards and self control failures. Hence, one can expect to find measurable attenuation in the prefrontal circuit in every impulsive behavior.

  • 28 The same remarks could be addressed to video game or Internet addictions. The fact that a growing n (...)

76Since it is not a sufficient condition for the distinctive pattern of addiction (see supra), the neuronal proof looks like a mere confirmation of what is previously anticipated by most clinicians (and most people). Neuronal evidence for shopping addiction for instance has not so far been found for the very simple reason that it has not been rigorously looked for. Some impulsive shoppers (and some psychologists) truly think nonetheless that shopping can be an addiction. Hence, even if the idea that impulsive shoppers are truly sick sounds nowadays ludicrous to most of us, maybe in the future impulsive shoppers will be regarded as pathological decision-makers.28 This is where the neuroeconomic identification of pathologies is circularly related to moral judgments. If the sufficient condition for addiction depends only on the personal experience of an internal struggle, as Ross et al. (2008) argue, then the definition depends on the purely subjective feeling of the addict himself. Additional neuronal evidence is not in itself an objective proof. It merely points out the importance of a prior social concern that is necessary for addicts to be recognized as such.

77The subtle tricks of neuroeconomics consist in considering the subjective and intuitive experience of psychosomatic discomfort as an objective fact. This assertion depends nonetheless on its prior clinical acceptance. For instance, in the study quoted above (Hare et al., 2009), the authors take as granted that a “rational” obese person who suffers from feeding behavior disorders should follow a diet and stick to it, because he or she would otherwise feel much worse: “dieters should be concerned with the healthiness of the foods, and not only with their taste, and that optimal decision-making requires integrating these two separate concerns” (Hare et al., 2009, 646). This assertion implicitly assumes that feeding behavior is classified by clinicians as a potentially addictive one. The neuroeconomic definition of addiction is therefore relative to medical practices.

  • 29 On the confusion made by neuroeconomists between normality and rationality, see footnote 26, supra.
  • 30 One could object indeed that my interpretation of a pathological paradigm was originally about comp (...)

78The same argument applies also to the definition of rationality. Rational behavior in such experiments is indeed characterized as the average behavior followed by normal subjects, i.e. individuals who meet social norms.29 And specifically, normal individuals are considered as sane or healthy subjects in the vocabulary of neurobiologists. How are healthy participants selected for neuroeconomic experiments? A simplistic but true answer is to say that healthy subjects are those presenting no neurological pathology. It is therefore because neuroeconomists take clinical definitions of neuronal deficits as given, that they can select a population of participants considered negatively as healthy, i.e. as not presenting any kind of neuronal deficit. In doing so, neuroeconomists can evaluate quantitatively what the “rational” decision should be.30

  • 31 The actual tendency seems to be toward an inflated use of the notion, since the next edition of the (...)

79Such a view backs up Canguihem’s thesis. According to Caguihem, the ontology of biological sciences is not made of atoms, molecules cells or organisms; their ultimate evidence lies rather in the subjective experience of disease and suffering. As a scientific concept, life is therefore not entirely objective and measurable. I am not saying that one should encourage “masochists” and “suicidal” tendencies, but medical assumptions are frequently drawn on as evidence and is not debated. The problem is that the neuroeconomic definition of addiction could lead to an inflated use, depending on its more or less restricted interpretation by clinicians. Shopping might be for instance addictive, since shopping addicts seem to truly experience the same kind of aversive internal conflict. The neuroeconomic view on addiction is actually an entirely psychological story, which is familiar to most of us: it is the story of people who become obsessed over the delivery of a given type of reward, in a given environment. Self control failures caused by this obsession result in guilt and suffering. A lot of different behaviors could fit such a story: internet addiction, porn addiction, etc.31

80On the other hand, from an economic perspective, the pathological paradigm in neuroeconomics does not produce a generic and axiomatic theory of rationality. Neuroeconomic models of addictive behaviors have a very limited scope. They are usually applied to eating behavior, disordered gambling, and to the issue of long-term savings. Moreover, their normative scope is negative: it is possible to identify pathological behavior in these experiments, but it is not possible to deduce a logical and self-consistent definition of rational behavior. However, my claim is that the introduction of pathological vocabulary in the decision-making analysis is fundamentally value-based: i.e., it is rooted in a normative judgment about what a normal life is. And it is precisely because this definition of rationality is both quantitative, very specific and at the same time normative that it is has potential for welfare economics and policy design.

6. Summary and Concluding Remarks

81This paper has tried to establish that neuroeconomics is embedded in a medical framework. This pathological paradigm legitimates the need for welfare enhancing policies. It thus involves a radical departure from libertarian paternalism (Sunstein and Thaler, 2003). Yet, Sunstein and Thaler’s proposals present a major weakness. These two authors refer to EUT to regulate addictive behaviors but they lack of an accurate distinction between mere impulsiveness and addiction. It is necessary to have a clear normative criterion of optimal decision-making to restore economic efficiency.

  • 32 For a complete review of the methodological issues concerning neuroimaging, see Hardcastle and Stew (...)

82Neuroeconomists benefit on the contrary from their close relationship to neuropathologists. This allows them to take the characterization of pathological forms of decision-making as granted, and to say how people should not behave. The most innovative experiments in neuroeconomics are perhaps the ones in which the performance of a group of healthy subjects is compared with patients exhibiting various forms of reward-processing pathologies. This protocol probably reflects the best neuroeconomic methodology, which is based on the combination of quantitative measurement and functional neurology. It should be noted that these experiments do not necessarily use neuroimaging techniques to be relevant. This is important because methodological discussions about neuroeconomics often overestimate the role of fMRI in this new discipline, and focus exclusively on the limits of neuroimaging instruments (see for instance Rubinstein, 2008; Harrison, 2008)32.

  • 33 I have therefore only defended here the potential of neuroeconomics for theoretical welfare analysi (...)

83Many neuroeconomists will not agree with the idea that their work is implicitly normative and could be used for economic policy design. However, my claim is not that the neurobiology of decision-making could directly be applied to “brain manipulation”, in order to enhance and improve decision-making abilities. What this paper has tried to demonstrate is rather that neuroeconomics offers the a priori theoretical justification for the design of such policies. These medical techniques borrowed from neurology can solve the theoretical problem of the legitimacy of libertarian paternalism. Their use in experimental economics does not necessarily serve the purpose of applying the principles of behavioral welfare analysis to the “real world”, but offers the authority of medical expertise to diagnose various pathologies of decision-making. In any case, neuroeconomists rarely explain how rationality could be “enhanced”, they merely suggest that some people should exert self-control in a potentially addictive environment.33

Acknowledgements are due above all to Don Ross and Jean-Sébastien Lenfant, who have offered very helpful and valuable comments on the manuscript. I am grateful to the members of the A. O. Hirschman seminar at the CES (University Paris 1 Panthéon- Sorbonne), who read the article with a critical eye, including Annie L. Cot, Jérôme Lallement, Cléo Chassonnery-Zaïgouche, Judith Favereau, Nadeera Rajapakse, and Sophie Pellé. I am also indebted to the two anonymous reviewers. Finally, I thank Harro Maas, Alan Hutton, Philippe Fontaine, and the participants of the 13th Summer School on Economic History, Philosophy, and History of Economic Thought in Acqui Terme.

Haut de page


Bechara, Antoine et al. 1995. Insensitivity to Future Consequences Following Damage to Human Prefrontal Cortex. Cognition, 50(1-3): 7-15.

Bechara, Antoine, Daniel Tranel, and Hanna Damasio. 2000. Characterization of the Decision-Making Deficit of Patients with Ventromedial Prefrontal Cortex Lesions. Brain, 123(11): 2189-2202.

Becker, Gary S. and Kevin M. Murphy. 1988. A Theory of Rational Addiction. The Journal of Political Economy, 96(4): 675-700.

Bernheim, Douglas B. and Antonio Rangel. 2005. From Neuroscience to Public Policy: A New Economic View of Addiction. Swedish Economic Policy Review, 12: 11-46.

Bernheim, Douglas B. and Antonio Rangel. 2007. Toward Choice-Theoretic Foundations for Behavioral Welfare Economics. The American Economic Review, 97(2): 464-470.

Berns, Gregory S. et al. 2001. Predictability Modulates Human Brain Response to Reward. Journal of Neuroscience, 21(8): 2793-2798.  

Berridge, Kenneth C. 2007. The Debate over Dopamine’s Role in Reward: The Case for Incentive Salience. Psychopharmacology, 191(3): 391-431.

Breiter, Hans C. et al. 2001. Functional Imaging of Neural Responses to Expectancy and Experience of Monetary Gains and Losses. Neuron, 30(2): 619-639.

Camerer, Colin F. 2008. Behavioral Game Theory and the Neural Basis of Strategic Choices. In Paul W.  Glimcher et al. Neuroeconomics: Decision Making and the Brain. London: Academic Press.

Canguilhem, Georges. [1966] 2005. Le normal et le pathologique. Paris: PUF.

Damasio, Antonio R. 1995. Descartes’ Error: Emotion, Reason, and the Human Brain. New York: Harper Perennial.

Damasio, Antonio R. 2008. Neuroscience and the Emergence of Neuroeconomics. In Paul W. Glimcher et al.. Neuroeconomics: Decision Making and the Brain. London: Academic Press.

Glimcher, Paul W. 2003. Decisions, Uncertainty, and the Brain: The Science of Neuroeconomics. 1st ed. Cambridge: MIT Press.

Glimcher, Paul W., Michael C. Dorris, and Hanna M. Bayer. 2005. Physiological Utility Theory and the Neuroeconomics of Choice. Games and Economic Behavior, 52(2): 213-256.

Glimcher, Paul W., Joseph Kable, and Kenway Louie. 2007. Neuroeconomic Studies of Impulsivity: Now or Just as Soon as Possible? American Economic Review, 97(2): 142-147.

Glimcher, Paul W. et al. 2008. Introduction: a Brief History of Neuroeconomics. In Paul W. Glimcher et al. Neuroeconomics: Decision Making and the Brain. London: Academic Press.

Guala, Francesco 2000. The Logic of Normative Falsification: Rationality and Experiments in Decision Theory. Journal of Economic Methodology, 7(1): 59-93.

Guala, Francesco. 2005. The Methodology of Experimental Economics. New York: Cambridge University Press.

Hands, D. Wade. 2009. The Positive-Normative Dichotomy and Economics. Paper prepared for Uskali Mäki. Philosophy of Economics. Vol. 13 of Dov M. Gabbay, Paul Thagard, and John Woods. 2011 Handbook of the Philosophy of Science. Amsterdam: Elsevier.

Hardcastle, Valerie G., and C. Mathew Stewart. 2002. What Do Brain Data Really Show? Philosophy of Science, 69(3 Supplement): 72-82.

Hare, Todd A., Colin F. Camerer, and Antonio Rangel. 2009. Self-Control in Decision-Making Involves Modulation of the vmPFC Valuation System. Science, 324(5927): 646-648.

Harrison, Glenn W. 2008. Neuroeconomics: A Critical Reconsideration. Economics and Philosophy, 24(Special Issue 03): 303-344.

Herrnstein, Richard J. 2000. The Matching Law: Papers in Psychology and Economics. Cambridge: Harvard University Press.

Houser, Daniel et al. 2005. Identifying Individual Differences: An Algorithm with Application to Phineas Gage. Games and Economic Behavior, 52(2): 373-385.

Lichtenstein, Sarah and Paul Slovic. 1971. Reversals of Preference Between Bids and Choices in Gambling Decisions. Journal of Experimental Psychology, 89(1): 46-55.

Loomes, Graham and Robert Sugden. 1982. Regret Theory: An Alternative Theory of Rational Choice under Uncertainty. The Economic Journal, 92(368): 805-824.

Kahneman, Daniel and Amos Tversky. 1979. Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47(2): 263-291.

Marschak, Jakob. 1950. Rational Behavior, Uncertain Prospects, and Measurable Utility. Econometrica, 18(2): 111-141.

Milner, Brenda. 1963. Effects of Different Brain Lesions on Card Sorting: The Role of the Frontal Lobes. Archives of Neurology, 9(1): 90-100.

Mitchell, Gregory. 2005. Libertarian Paternalism is an Oxymoron. Northwestern University Law Review, 99(3): 1245-1277.

Montague, P. Read. 2002. Neural Economics and the Biological Substrates of Valuation. Neuron, 36(2): 265-284.

Palminteri, Stefano et al. 2009. Pharmacological Modulation of Subliminal Learning in Parkinson’s and Tourette’s Syndromes. Proceedings of the National Academy of Sciences, 106(45): 19179-19184.

Platt, Michael L. and Paul W. Glimcher. 1999. Neural Correlates of Decision Variables in Parietal Cortex. Nature, 400(6741): 233-238.

Rachlin, Howard and Leonard Green. 1972. Commitment, Choice and Self-Control. Journal of the Experimental Analysis of Behavior, 17(1): 15-22.

Rangel, Antonio, Colin F. Camerer, and P. Read Montague. 2008. A Framework for Studying the Neurobiology of Value-Based Decision Making. Nature Reviews Neuroscience, 9(7): 545-556.

Ross, Don. 2008. Two Styles of Neuroeconomics. Economics and Philosophy, 24(Special Issue 03): 473-483.

Ross, Don et al. 2008. Midbrain Mutiny: The Picoeconomics and Neuroeconomics of Disordered Gambling: Economic Theory and Cognitive Science. Cambridge: MIT Press.

Ross, Don, forthcoming. Addictive, Impulsive and other Counter-Normative Consumption. Forthcoming in Victoria Wells and Gordon Foxall. New developments in Consumer Behaviour. Edward Elgar.

Rubinstein, Ariel. 2008. Comments on Neuroeconomics. Economics and Philosophy, 24(Special Issue 03): 485-494.

Schultz, Wolfram. 1986. Responses of Midbrain Dopamine Neurons to Behavioral Trigger Stimuli in the Monkey. Journal of Neurophysiology, 56(5): 1439-1461.  

Sokol-Hessner, Peter et al. 2009. Thinking Like a Trader Selectively Reduces Individuals’ Loss Aversion. Proceedings of the National Academy of Sciences, 106(13): 5035-5040.

Schotter, Andrew. 2006. Strong and Wrong. Journal of Theoretical Politics, 18(4): 498-511.

Sutton, Richard S. and Andrew G. Barto. 1998. Reinforcement Learning: An Introduction. Cambridge: MIT Press.  

Thaler, Richard H. and Cass R. Sunstein. 2003. Libertarian Paternalism. The American Economic Review, 93(2): 175-179.  

Tversky, Amos, Peter Slovic, and Daniel Kahneman. 1990. The Causes of Preference Reversal. The American Economic Review, 80(1): 204-217.

Haut de page


1 In this paper, the term “behavioral economics” refers to the experimental studies of economic decision-making. It is closely associated with experimental psychology. In a broader sense, “experimental economics” includes all kinds of experimental research in economics. Hence, it is not necessarily focused on decision-making analysis; for instance, the experimental studies of market institutions do not investigate individual psychology. This distinction is important because some experimental economists do not consequently consider themselves as “behavioral economists” (e.g., Vernon Smith).

2 My story excludes hence the research program of “behavioral economics in the scanner” (Ross, 2008) and the neuroeconomic literature dedicated to “social-decision making”, which is sometimes associated with the works of Ernst Fehr. For the sake of clarity, I will rather focus on the problem of non-transitive preferences that is studied through reward-processing experiments. My argument about the pathological meaning of neuroeconomics appears indeed much more convincing when restricted to a particular sub-class of experiments.

3 Paul Glimcher is indeed one of its major contributors, as we shall see later.

4 See Bernheim and Rangel, 2007.

5 The core of Sunstein and Thaler’s argument is that choice frames greatly affect the choices made by individual. As a result of this preference endogeneity, public authorities have to design choice environments in order to “nudge” individuals to select the welfare-advancing option. Sunstein and Thaler thus assume that some choice settings favor rational behavior, and some do not. But what is fundamentally lacking in libertarian paternalism is precisely “a theory of when choice frames will control choice and when they will not” (Mitchell, 2005, 11). Hence, there is no way to determine the true “rational” preferences of individual. Libertarian paternalism is not therefore welfare-advancing unless an objective measure of welfare is adopted. Since “Sunstein and Thaler do not endorse any particular measure of welfare to guide the planner” and “do not choose one particular conception of the good that the planner should try to achieve” (Mitchell, 2005, 29), libertarian paternalism inevitably becomes coercive because it is necessary to identify what people’s best interests are.

6 As for neurobiologists, the term “paradigm” has a restricted sense, which means “experimental protocol”. Still, the design of the protocol also reveals a theoretical choice: a particular experimental task is thought to exhibit a relevant behavioral regularity. For instance, reward-processing experiments are thought to be adequate to investigate the properties of the dopaminergic system.

7 For a general introduction to the machine learning algorithms mostly used in neuroeconomics, see Sutton and Barto, 1998.

8 The computation operated by these neurons in visual tasks is not correlated with the amplitude of the reward itself but with its expected value, or with any characteristic of the motor task to perform (amplitude, direction, speed of the visual saccade). Reward processing has therefore to be distinguished both from hedonistic experience (pleasure to consume a reward) and motor control; however, it may explain these two functions in terms of decision processes (see Bernheim and Rangel, 2005).

9 For instance, in their 1999 foundational article, Michael Platt and Paul Glimcher wrote: “here we describe a formal economic-mathematical approach for the physiological study of the sensory-motor process, or decision-making” (Platt and Glimcher, 1999, 233).

10 For instance, Glimcher himself proposed in his 2003 influential book on neuroeconomics that economics could serve as a source of normative models in cognitive neuroscience (Glimcher, 2003).

11 Once again, even if the dopaminergic system actually does compute a quantitative desirability for each available course of action, these valuations work on trial and error processes and are always optimal with regard to the individual’s past history: as Glimcher points out, “valuations are always rational with regard to these internal [acquired, learned] representations of desirability” (Glimcher, 2005, 220). What are rational or irrational are not the valuations themselves, but the evaluating process, i.e. the algorithm that allows the agent to learn from previous decisions. But EUT does not specify such an optimal feedback mechanism and is not a theory of rational learning. Paul Glimcher was in fact quite clear about these issues in his 2005 article, and a careful reading of his paper would not have resulted in any misunderstanding. I am merely arguing here that the semantics—in particular, the expression of “physiological expected utility”—were not very appropriate for economist readers.

12 Glimcher himself admitted his mistake later: “The paper [see Glimcher and Platt, 1999] goes on to demonstrate that the activity of single neurons in the posterior parietal cortex is a lawful function of both the probability and the magnitude of expected rewards. This was significant, because standard expected utility theory predicates choice on lawful functions of these same two variables. The paper, however, makes a critical mis-step in its examination of actual choice behavior. The authors go on to examine a matching-law type behavior which they interpret in terms of normative expected utility theory. This is problematic, because there is no normative standard for the analysis of matching law behaviors.” (Glimcher et al., 2008, 8)

13 See Herrnstein, 2000.

14 See for instance Breiter et al., 2001.

15 Counterfactual judgments and “choiceless utility” also explain why it is sometimes rational to reverse preferences; see for instance the famous “regret theory” of Loomes and Sugden (Loomes and Sugden, 1982).

16 In particular, these pathological players include patients with prefrontal cortex lesions.

17 Such pathological behavioral pattern can be specifically exhibited using the Wisconsin Card Sorting Task (see Milner, 1963).

18 For a full description of PF cortex damaged patients behavior, see Damasio, 1995.

19 See Ross, forthcoming, 39: “The neuropsychological account of the way in which the striatal reward circuit comes to be captured by a dominant network of attention-arresting cues converging on one or two targets (e.g., alcohol, cocaine, gambling, tobacco, or two or more rarely three of these), and then comes to gain increasingly direct control of motor response...identifies the necessary (but not sufficient) conditions for the distinctive patterns characteristic of addiction.”

20 For a detailed account of those criticisms, see Ross et al., 2008, 97.

21 Canguilhem, [1966] 2005.

22 For the sake of clarity, I will not provide a complete review of the works of Canguilhem here, but I will try to summarize and explain his arguments about biological and medical sciences that are relevant to this paper.

23 The terms “implicit” and “explicit” methodologies do not appear in Canguilhem’s vocabulary. Once again, I introduce this distinction for purposes of clarity.

24 Once again, the term “ethics” means the ability to form a normative judgment about what is a good, “normal” life, and to distinguish it from pathological ways of living.

25 The term “ethical” here has therefore to be understood as a particular form of behaviour and conducting of one’s own life.

26 Still, one need to keep in mind that strictly speaking, economic rationality is not equivalent to normality in Canguihem’s sense, i.e. to the average. It is precisely on the basis of such a confusion that most neuroeconomists are able to make the critical step from a clinical view on normality to an “economic” one on rationality.

27 Hence, the argument that “there is no social consensus” on “how much shopping is too much” is not logically sufficient to a priori reject shopping as a potentially addictive behavior. The exact quantification of “too much shopping” might be difficult (as, for instance, defined as a percentage of the agent’s salary). Still, impulsive shopping might be considered as an addiction if most people consider as an objective fact that some individuals spend too much money on shopping and behave irrationally in a shopping mall. As a result, there can be a social consensus on pathologies without having a clear definition of normality or socially accepted behavior.

28 The same remarks could be addressed to video game or Internet addictions. The fact that a growing number of online players consider themselves as addicts is probably the first step to “official” recognition of on-line activity as a potential addiction (see for instance the website of “On-line Gamers Anonymous”: The idea of on-line game addiction has nowadays a growing influence in the audience and will probably be regarded in the future as a serious disease by medical authorities.

29 On the confusion made by neuroeconomists between normality and rationality, see footnote 26, supra.

30 One could object indeed that my interpretation of a pathological paradigm was originally about computational neuropsychiatry and is therefore restricted to the neuroeconomic experiences on addictive behaviors. In other words, the fact that brain-imaging and brain-stimulation technology can be applied to investigate neuro-psychological pathologies does not imply necessarily that neuroeconomics on the whole rests on medical semantics. Still, my claim also applies to rational behavior, because, as I have said before, rational decision-makers are implicitly characterized in these experiments as sane subjects. Hence, the only proof that my argument requires is the explicit acknowledgment that “normal” subjects are recruited according to medical advice. It is in this sense that all neuroeconomists explicitly endorse medical categories and presuppositions. In every neuroeconomic experiment involving human participants, one can read such statements as “each participant was screened to exclude medication and conditions including psychological or physical illness or history of head injury” (Coricelli et al., 2005, 1260). This quote is taken for instance from an experience on the role of regret in rational decision-making, It can be understood therefore as an illustration of the pathological paradigm in the field of rational behavior.

31 The actual tendency seems to be toward an inflated use of the notion, since the next edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) made by the American Psychiatric Association will integrate behavioral addiction in its next classification. It will be restricted to gambling addiction but it is expected to include internet addiction in the following years: “There is some support in the DSM client community for immediate additional inclusion of Internet surfing addiction. The conservative policy of recognizing, for the present, only gambling addiction simply reflects the severely incomplete current state of rigorously scrutinized empirical evidence.” (Ross, forthcoming, 39) Once again, the definition of addiction will be clearly influenced by both social concerns and medical authorities, through the orientation of empirical research.

32 For a complete review of the methodological issues concerning neuroimaging, see Hardcastle and Stewart, 2002.

33 I have therefore only defended here the potential of neuroeconomics for theoretical welfare analysis, but I have not discussed the effective means to restore the efficiency to pathological decision-making. Reward learning abilities appear to be nevertheless manipulated at will, through pharmacological techniques (Palminteri et al., 2009, 1042). However, it is probably too early to evaluate the possible uses of neuroeconomic research in real life and therapeutic applications (of the so-called neurocomputational psychiatry) seem for now quite uncertain. More basically, practical lessons that can be learned from neuroeconomic experiments include most of the times very simple psychological recommendations. “Emotional reappraisal” for instance might enhance personal decision-making abilities: asking subjects to re-evaluate consciously their expectations is an easy way in order to make them act more rationally. This is the conclusion of a recent paper whose title is rather explicit “Thinking like a trader selectively reduces individual loss aversion” (Sokol-Hessner et al., 2009). Individuals have to think rationally about their process of choosing; this issue is related to “meta-cognition” and “new behavioral therapies” (see Ross et al., 2008).

Haut de page

Pour citer cet article

Référence papier

Nicolas Vallois, « The Pathological Paradigm of Neuroeconomics », Œconomia, 1-4 | 2011, 525-556.

Référence électronique

Nicolas Vallois, « The Pathological Paradigm of Neuroeconomics », Œconomia [En ligne], 1-4 | 2011, mis en ligne le 01 décembre 2011, consulté le 29 avril 2017. URL : ; DOI : 10.4000/oeconomia.1442

Haut de page


Nicolas Vallois

Centre d’Economie de la Sorbonne (CES), Université Paris I Panthéon-Sorbonne,

Articles du même auteur

Haut de page

Droits d’auteur

Licence Creative Commons
Les contenus d’Œconomia sont mis à disposition selon les termes de la Licence Creative Commons Attribution - Pas d'Utilisation Commerciale - Pas de Modification 4.0 International.

Haut de page
  • Logo Association Œconomia
  • Logo CNRS
  • Les cahiers de