Reviewed by
Cristiano Castelfranchi
ISTC-CNR, Rome
This book will be an unavoidable reference in the next decade, since it deserves to be read, or even better carefully studied by all the scientists interested in understanding human behaviour not only on markets but on 'decision making' in general. Actually, the book has three missions, and is well conceived to accomplish each of them. Firstly, although collecting different articles (some of them already well-known, others expressly written for this book), it stands as the real 'handbook' of this recent but booming discipline that is attracting many economists, psychologists, or neuroscientists, providing at the same time a "fairly intimate understanding of the accomplishments and challenges in Neuroeconomics". Secondly, it is a very good textbook that provides the needed interdisciplinary and historical background for students of many disciplines that are interested in explaining human behaviour on markets. Thirdly, this book is a sort of 'time capsule' sent to the future, to see in ten or twenty years what of this scientific bet was really prophetic and played the expected role. As a matter of fact, though neuroeconomics has definitely embarked many supporters in recent times, there is still a huge room for colleagues who are sceptical or disagree with the foundations of this discipline (see the conference "Neuroeconomics: Hype or Hope?", Nov 2002, 2008, Erasmus University Rotterdam).Although this discipline has no more than ten years of history, it is revolutionary and aggressive. No doubt that its impact on the cognitive sciences and on economics will be impressive, as well as its impact of the whole cognitive neuroscience. This will bring some fundamental advancements and clarifications, but also the risk of very simplistic views and reductionist approaches. This book is important also in this respect, since it is very serious and well documented, but presents non-converging methodological positions and even divergent 'philosophies', and some ambiguity and contradiction as well. In this review, I will remark on some of these points below.
Given the very rich and quite deep content of the book, which even includes some really nice comparative chapters on primates (e.g., Chapters 6 and 7), it is out of my purpose here to report precisely on all the topics and issues covered by the book, or to discuss some specific chapters in detail. Rather, I will try to reflect on the general importance of this book and of this new 'discipline', touching upon some general challenges and questions that are put forward in this book.
Perhaps, compared with the richness of the book (32 chapters! from neoclassical economics, to Prospect theory, from emotions, to Behavioural Game Theory, from empathy and 'theory of mind', to primates, from learning, to uncertainty, subjective value, etc. ) and with its top-level contributions (Vernon Smith, Kahneman, Camerer, Fehr, Houser and McCabe, Rustichini, Bernheim, Gallistel, Damasio, among others!), my points will appear simplistic and superficial arguments. The only excuse I have is that I'm providing a simple review for a very different and interdisciplinary community, like JASSS community, and that I'm trying to enlighten some general epistemological or theoretical issues, which go beyond the particular contents of the chapters of this book, trying also to look at the self-presentation, the 'social representation', and the cultural inpact of this discipline.
A good question is: Why should this book be of interest for the JAASS community (except that neurosciences have attracted so many psychologists, economists, sociologists)?
Firstly, this book offers a very representative portrait of a discipline that will be one of the protagonists of future scientific trends and revolutions; to understand where science is going, we should be aware of what is happening particularly in these areas. Secondly, this book is worth reading because this important revolution will be joined with reductionist and eliminativistic temptations and simplistic shortcuts, that might require some critical discussions by social scientists. For instance: Is 'trust' just a chemical activation of a generic disposition? Or is a much more complex and analytical mental state (evaluations, expectations, decisions) and social relation? Is 'prejudice' just a specific activation of some hostility reactions in a brain area, or a complex mechanism of ungrounded beliefs and generalized expectations?
A third reason to read this book is because some results are specifically relevant for enlightening exactly those phenomena that we, as computational scientists, are modelling (e.g., cooperation, trust, or subjective values and learning, imitation, and empathy).
A fourth reason is that some epistemological discussion about scientific models or about the real meaning of economic theories or of social laws is of general interest and also pertinent for the social simulation debate.
But, last but not the least, this book is worth reading also because there is an underlying open issue within our own horizon: Do we need realistic models of mental activity, and even simplified artificial model of neural activity in our agents? Do we need bio-inspired and even brain-inspired learning, decision, expectation, affective models, in order to provide credible models of the social phenomena that we simulate? More in general, how is it possible that we successfully simulate complex human social phenomena with very simple ('stupid') rule-based agents? Isn't cognition crucial for human social behaviour?
Let us suppose we are modelling economic/financial phenomena. The question would be: Isn't 'intelligence' necessary for making money? Some possible answers will be that this is possible because: (i) humans frequently appear to be 'stupid' (routine-based, automatic, just based on associations, imitation, etc.); (ii) frequently what really matter are the behavioural consequences of the mental contents. However, in other cases, explicitly modelling mental representation should be necessary. If this is true, does the same hold for realistic brain-based models?
But, let's come back to neuroeconomics. As I said before, the book presents different, even conflicting perspectives. As the authors remark, "scholars within neuroeconomics are still debating whether neuroscientific data will provide theory for economists or whether economic theory will provide structure for neuroscience". I would answer: "Probably both". However, I would also react by saying: "Where is psychology or cognitive science and their theories and models, not just experiments? Doesn't psychology (or better its modelled 'mechanisms') mediate this relationship? Will those notions and models be just jumped?". But, as I said before, it is precisely remarkable that - contrary to some simplistic trends (especially in the media) - in this book an eliminative attitude towards psychological notions, theories, and models is more the exception than the rule. There is some ambiguity on this. The temptation to shortcut and establishing a direct relationship between the behavioural (in this case, economic) notions and the neural mechanisms is always there, but it does not prevail. Sometimes, this temptation, not only in the media but in important published articles, emerges even in a ridiculous way, rediscovering what is obvious and very well-known not only in scientific psychology but even in common sense psychology.
For instance: Do we need a brain 'evidence' (that actually must be 'interpreted'!) to know/prove that "rejection hurts"? Is this a discovery? Wasn't absolutely well-known even before knowing the specific neural or biochemical substratum? Fortunately, the triviality emphasized in the title of Eisenberger et al's paper (2003) is not the only result of the article, which rather raises some questions of utmost interest for us: Is merely psychological sufferance real pain, or better "similar to physical pain"? Are the areas involved in physical pain also activated in non-physical sufferance? Put in more psychological terms: Why we feel 'pain' when nothing is happening to our body?
However, it becomes more and more frequent that even top-level scientific journals (like Nature or Science) make use of a dangerous mass-media style about scientific discoveries, such as in the famous male "gene of infidelity", which was a ridiculous lemon. The added value of this book is that it avoids this fashionable attitude; although the neuroeconomics results are frequently presented in such a way on the mass-media.
Another typical risk of this kind of research is its reduction to 'geography', to mere descriptive and non-explanatory data; to the mere localization of certain mental activities in some brain area. So what? What does this add to our knowledge? We need an interpretation in terms of specific 'functions' (mechanisms or processes) played by those neurons. Without this 'psychological' interpretation, localization or biochemistry say nothing. On the contrary, with this interpretation, it may be enlightening to know that such a specific kind of process is - not accidentally - performed in that area or by that kind of molecule. In general, there is a risk of coarse ideas about the psychological phenomena, while on the contrary only an analytical account, a compositional, structural and process model can enlighten and be enlightened by neural data. It would really poor to say that 'fear' is in the Amigdala, since the Amigdala is activated in fear reaction, and without this there is no felt fear. However, 'fear', in its complex and advanced form, is not simply due to a given stimulus, fired by something, but it is 'about' something (emotion 'intention'), and we do not simply 'move' but we - subjectively and not only objectively - 'escape from ...'. And this content and reason of our fear can just be a mental representation or an idea (a prediction, even a counterfactual thinking, etc.). Similarly, it is very simplistic and naive to claim that altruism receives an internal hedonic reward and thus is motivated by such a hedonic aim. This problem has already been clarified by Seneca: "Sed tu quoque' inquit 'uirtutem non ob aliud colis quam quia aliquam ex illa speras uoluptatem. Primum non, si uoluptatem praestatura uirtus est, ideo propter hanc petitur; non enim hanc praestat, sed et hanc, nec huic laborat, sed labor eius, quamuis aliud petat, hoc quoque adsequetur" (De vita beata, IX). In a few words: Do we cultivate our virtues just because we wish to obtain some pleasure? The fact that virtue gives some pleasure does not mean that we follow it because of this. The pleasure is just an additional result, not our aim or motive. We will get it while pursuing another aim, that is, virtue.
We would expect from modern science at least the same degree of discrimination of philosophical ethics. An expected reward is not the same of a motive, and being motivated (having an explicit aim) is not the same of learning by a positive reinforcement. Some level of naiveté, some brutal simplification, is unavoidable and acceptable, especially at the very beginning of such a hard discipline. But, it is a problem when simplification becomes some sort of manifesto or an ideology.
A very interesting debate in this discipline - very typical also in many others, especially within the cognitive sciences - is about the relationships between theoretical mechanisms or models and the 'real' underlying proximate mechanisms (in this case neural mechanisms) that are responsible for the behaviour that is under analysis. As it is argued in the book: "Prior to the 1990s it had been a completely ubiquitous view in economics circles that models of behaviour, like expected utility theory, were 'as if' models. The model was to be interpreted 'as if' utility were represented internally in the chooser. As Samuelson had argued half a century earlier [but, this detachment from psychology started with Pareto], it was irrelevant whether this was actually the case because the models sought to link options to [behavioural] choices not to make assertions about the mechanisms by which that process was accomplished" (p. 9). On the contrary, at least in part of Neuroeconomics (e.g., Camerer), researchers go exactly in the opposite direction: they claim that we need a precise understanding of how 'utility' is represented in the brain to explain human choices and behaviour.
This debate is not so different from a debate very relevant - under a different perspective - in agent-based social simulation. Are the specific micro-mechanisms that we implement in our agents, the theory and the model of the hidden mechanisms working in human brains that are responsible for that behaviour? Is our 'program' - even in its implementation details - our theory? (a famous claim of Herbert Simon, and Roger Schank). Are the axioms of the theory (capturing behavioural regularities), and even the implemented models of processes, indifferent to their actual mental implementation, without pretending to be the model of the underlying mental mechanisms?
A typical and frequent mistake (also in our JASSS community) is first to take some problematic economic simplifications about rationality for granted, well defined, and intangible aspects, and then saying/showing that humans are irrational. For example, to take for granted the arbitrary reduction of 'rationality' to selfishness, to 'economic' (monetary) incentives, to come later to discover that human beings do not take into account only these 'rewards' but also intrinsic and moral ones, and pursue equity, or the welfare of others ('social preferences'). 'Rationality' says nothing and cannot say anything at all about the 'right' rewards, values, a man should pursue and take into account. It can just say something about 'how' we should calculate a satisfactory or even an optimal choice, once given our personal (internal) values/desires. Therefore, equity or generosity or, let's say, altruism might perfectly be subjectively rational; and one should not confuse an autonomous, self-motivated (self-directed) system with a 'selfish' one.
But, let me come back to the relevance of this book for JASSS and for opening possible relationships between neuroeconomics and computer simulation. As said before, the neurorevolution will deeply transform not only economics but the cognitive and behavioural sciences (consider that the focus of the book is in particular 'decision making'). Which is the relationship with another revolution that is radically transforming the cognitive and social sciences, that is, the computational modeling and simulation of mental and social processes?
These revolutions have in a certain sense something in common: they both want to go beyond mere 'as if' assumptions, or beyond mere 'intervening non-observable variables' and merely postulated 'proximate mechanisms', to propose very specific 'operational' and directly experimental models of those hidden processes. In this sense, they both represent a radical overcoming or development of psychology, with its quite vague verbal or box-based 'models'. As I said, neuroscientists and neuroeconomists are still ambiguous on this: Do they directly want to propose brain models of the proximate causes of behavioural processes? Do they want not only to jump the psychological theory layer but also any functional-formal models, including computational/algorithmic models of functional, informational, control-theoretic processes? Or are they interested in such a level of organization and of theory as a guide to brain research and as an interpretation and understanding of brain data?
Sometimes, it seems to me that such a relationship between brain processes and models and economic phenomena is be a direct one, without any intermediary layer. Sometimes, it seems that researchers in this discipline are interested in more abstract and functional, computational, models of brain activity. For example, in Ch. 30: "We [...] saw how mechanistic models can offer a solution to this problem by providing reasonable proxy variables with which to correlate physiological signals" (p. 478). In such a case, the relationship between neuroeconomics and agent-based simulation can be multiple, with many aspects involved. On the one side, multi-agent simulation may be a possible approach for modelling brain organization and processes, and its complexity-based and emergence phenomena. This is a very promising research direction. It is in part anticipated in the introduction when the authors claim that one might - for example - (economically) account for the psychological 'dual-process' theory in terms of "two (or more) independent interacting agents being locked in a bad equilibrium by their own self-interests". On the other side, computational agents (being simple or cognitively sophisticated) can be a good platform for implementing some basic principles postulated by neuroeconomics (e.g., in term of decision making, or of social preferences and dispositions, or of trust and reciprocity, or of reinforcement learning of values), as well as a better means to model their architectural and procedural properties, or to experimentally demonstrate which interactive or collective phenomenon does really follow from those 'mental rules'. Question like: "Is cooperation motivated by altruism (social preferences), by the personal reward that emanates from relationship-building (goodwill) in exchange, or by failure to backward induct" (p.17, and Ch. 15) are typically crucial questions in both behavioural economics, experimental studies, now in neuroeconomic studies, but also in social simulation studies: are the postulated internal (behaviour controlling) rules and mechanisms really able to produce the predicted social conduct and phenomenon?
Is the maximization of the total obtained utility an 'aim' of our mind (or brain) in the sense that the brain works 'as if' the system was aimed at this, governed by this 'goal', but actually is not a real 'goal' (THE only dominating 'goal' of mind, to whom all the other goals - desires, needs, ambitions - just are occasional means)? Is subjective utility maximization just a functional way of working of our mental machinery, its function but not a real 'goal': an explicit, anticipatory representation of the results, evaluating, selecting and controlling our behaviour?
To answer this question - which sounds rather crucial to me - we need explicit models of the cognitive and motivational architecture, of 'goals' (ambiguously called 'preferences' by economists, given their dominating 'as if' behavioural paradigm), of decision processes, and so on.
To explain my evident 'ambivalence' towards this revolutionary discipline, let me take, as example, the very important chapter 15 on Social Preferences, and its Janus' faces.
How to deny that there are very interesting, even exciting (but confused), results, for example, on the issue of 'social preferences' (altruism) and of trust? This chapter presents the large "experimental evidence ... indicating that a substantial percentage of people are motivated by other-regarding preferences and that neither concerns for the well-being of others nor for fairness and reciprocity can be ignored in social interaction" (p.217), and the identification of the brain areas or processes involved in those motivations and concerns. For example, there are data showing that trust (but in which sense?) is influenced by oxytocin neuropeptide, which does not reduce generically a risk-taking attitude, but more specifically reduces the fear of social betrayal, and would inhibit diffidence and defensive behaviour. For sure, this is a basic component of trust (especially of trust as feeling): feeling her/himself as non-exposed and safe, not perceiving threats and dangers from others. Of course, a nice result. However, why it should be also true that: "trust decisions are likely to involve perspective taking and theory-of-mind" (p. 228)? Why and how this should be connected with the inhibition of 'defensive behaviour'? Why and how should this be also connected to the so-called 'reputation' (sic!), that is, the estimation of the probability of the other's reciprocation based on learning and previous interactions?
This very high level and inspiring researches nevertheless give the sensation to grope one's way in the dark, and to disregard their social or psychological 'objects' of research; the common-sense concepts, or the constructs of psychological sciences, and their models. They seem not at all theory driven. Fehr - convincingly - "argue[s] strongly in favour of using a mix of methods in future neuroeconomic experiments" (p. 230), but he doesn't seem to worry at all about the dominating empiricism and conceptual nonchalance, as follows: the no-problem identification between 'subjective value' and 'hedonic value' or utilities, or the very bad and reductive notion of 'reputation', or of 'punishment', or of 'envy' ('disutility to be worse off'), or of 'compassion' ('disutility of being better off'). But, this was already a problem typical of behavioural economics, not of this excellent chapter.
Is it correct and productive to grope one's way in the dark, without the guide of a model, for example, of an unitary theory of trust, presenting trust as a casual and occasional set of features enlightened time by time by one discipline or the other, or by one experiment or another? Wouldn't be a task (mission) also of neuroscience providing us a 'theory' of trust, with its organic aspects and functions, or a theory - not merely inductive, but based on the failure of a unitary attempt - of the existence of different phenomena mixed-up under a bad common-sense notion of 'trust'?
Let me conclude this review, coming back to the limits of brain 'explanation' in terms of activation and localization. For example, chapter 5 examines the relation between behaviours and neural activation of subject engaged in a strategic game. The subjects play a trust game either against an anonymous human opponent or against a computer. The neurological data show - for example - that in some subjects the medial prefrontal cortex becomes more active when they play a cooperative move and deviate from the normative economic prediction. What does this kind of geographic results/data really add? Wasn't really important the experimental behavioural results that the subject - in certain context - deviates from the 'rational' precepts and seems moved by 'social preferences' or other social motives (e.g., dealing with other humans, reciprocation, fairness, trust building). The authors' explanatory hypothesis is that this deviating cooperative pattern would originate in the circuits of the prefrontal cortex; but this seems just a restatement: it doesn't explain anything of the behaviour. In case, it requires to add something (some hypothesis) about the nature and function of that brain area.
This limited contribution (simple localization) is, in my view, in part responsible of the situation described by Daniel Kahneman in his "Remarks" (the final chapter). He wrote: "My impression is that [...] it is still early days for neuroeconomics. When I was asked by a well-known critic of the field what I had learn that had changed my mind about decision making, I did not have much to say [...] The findings of neuroeconomics research have generally confirmed the expectations of behavioural decision theorists and behavioural economics. However, we are beginning to learn more, and I am confident that the pace will accelerate in coming years". Let me just add to this that - in my perspective - neuroscience should soon overcome a merely 'descriptive' attitude (what I call 'geography') to assume a much stronger (and hard) explanatory perspective, which presupposes explicit 'models' of functional architectures, of specific orchestrated 'mechanisms', of specific (explicit or implicit?) representations, and so on; thus, possibly, psycho-computational models. Let's say - a bit brutally and cynically - that to show that two different responses (e.g., behavioural, affective ones) activate different brain areas it is simply obvious and does not add or explain anything. Fortunately, several important data in this book are really challenging economic or psychological models. Under this perspective, the Kahneman's conclusion - balanced and polite as it is - is not very generous, and too soft as well. On one side, he does not recognize some important contributions and debates motivated by neuro-research on choice. In particular, I have in mind the important studies about how values are represented and 'calculated' in brain (Ch 22; Ch. 29; Ch. 32, and others). In particular, it is worth mentioning the studies aimed to understand whether the 'decision' about where to automatically address our attention, and which is the most relevant stimulus, is the same process guiding our motor-behaviour 'choices' (left hand or right hand?), and is based on the same mechanism governing our choice between two goods. Neural data importantly show that there are various representations of 'values' in different areas. The hypothesis of some authors (see Ch. 29) is that those different value signals in various regions contribute to different mental processes, and very different kinds of 'decisions': frontal areas value signals contribute to a choice between goods, while in sensory areas the value signals determine attention, and so on. But, this gives rise also to an important theoretical debate between two models: (i) the 'actions-based model' where the choice, even the choice between goods, is embedded in premotor processes of action selection, and is due to general associative laws of reinforcement learning; (ii) the 'goods-based model' that "assumes a level of mental abstraction - the space for goods - computationally removed from sensory and motor representations" (p. 457), in a sort of independent cognitive 'module'. These are important evidence for a crucial and general discussion on these important issues. On the other side, Kahneman seems too soft and not-assertive. Although oriented towards an interdisciplinary collaboration and integration, what we need is a more 'aggressive' fight. An example of this is Caccioppo et al. 2003, where the authors show a deep criticism against neuro-trivialities and re-discoveries, and useless localizations in terms of "centres for", suggesting at the same time some principles for a mutual understanding and collaboration between social psychologists and social neuroscientists.
ReferencesEISENBERGER NI, LIEBERMAN MD, WILLIAMS KD. (2003) "Does Rejection Hurt? An fMRI Study of Social Exclusion." Science, 302, 10 October, pp. 290-292.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2009