New trends  in decision-making: computational cognitive social  neuroscience

New trends  in decision-making: computational cognitive social  neuroscience

6-7 December 2018  
ENS, salle Dussane, 45 rue d'Ulm, 75005 Paris

Inscription sur eventbrite 

Day 1

Session 1
14h30-15h15: The teaching  brain: general hypothesis and first experiments (Stefano Palminteri)
15h15-16h00: Investigating variations in human susceptibility to social influence: the ecological hypothesis (Pierre Jacquet)

16h00-16h30: Coffee break

Session 2
16h30-17h15: Social and Neural Underpinnings of Individual Decision-making in Inter-group Conflict (Carsten De Dreu)
16h15-18h00: Using Affect to Predict Choice (Tali Sharot)

Day 2

09h00-09h30: Welcome breakfast

Session 3
09h30-10h15: Fairness as an « incentive landscape  » for cooperation (Jean-Baptiste André)
10h15-11h00: Quantifying the rise of trust in history using face-recognition machine learning
(Lou Safra)

11h00-11h30: Coffee break

Session 4
11h30-12h15: Opportunities, costs and apathy: A computational neurology approach  to behavioural and social motivation (Matthew Apps)
12h15-13h00: The brain in the social world:  Integrating approaches from social neuroscience, psychology and social network  analysis (Carolyn Parkison)

13h00-14h30: Lunch break

Session 5
14h30-15h15: The Art of The Deal: a Reinforcement Approach to the Ultimatum Game (Maël Lebreton)
15h15-16h00: Neurocomputational basis of prosociality and selfishness: a reinforcement learning approach (Patricia Lockwood)

16h00-16h30: Coffee break

Session 6
16h30-17h15: Playing with Theory of Mind: a (narrow) computational view on social cognition (Jean Deaunizau)
16h15-18h00: Visual salience in matching and hide-and-seek games (Colin Camerer)

Abstracts booklet

Stefano Palminteri (Ecole Normale Supérieure)
The teaching  brain: general hypothesis and first experiments
This talk is divided into two parts: one theoretic and one empiric. In the first part we propose that teaching (as opposite to imitation) is the form of social learning that allowed humans to develop a complex cumulative culture. Since teaching (as opposite to imitation) requires cooperation, our reflexion suggests that cooperation is a prerequisite of culture and not the opposite. In the second part of the talk we present preliminary findings concerning comparing both forms of social learning (teaching vs. imitation) in the lab in the context of complex reinforcement learning tasks.

Pierre Jacquet (Ecole Normale Supérieure)
Investigating variations in human susceptibility to social influence: the ecological hypothesis There is considerable variability in the degree to which individuals rely on their peers to make decisions. Why in different times and in different places, people display a preference for independence and personal exploration or a preference for following the behaviours,  beliefs  and  attitudes  of  their  peers? Theoretical models predict that environmental risks shift the cost-benefit trade-off  associated  with  social  information  use.  Yet,  this  idea  has  received  little  empirical  support.  I  will  present  two experimental works testing the hypothesis that individuals’ ecology contributes to shape their dependency on social information. The first work aims to reveal associations between perceived vulnerability to extrinsic morbidity risks, the neurophysiological correlates of social information processing, and the individuals’ susceptibility to social influence. The second work aims to track the  impact  that  environmental  harshness and unpredictability individuals experienced during their childhood have on their propensity to follow other people’s opinions. Implications of the reported findings for current views of social information use and cultural transmission will be discussed.

Carsten De Dreu (Leiden University & University of Amsterdam)
Social and Neural Underpinnings of Individual Decision-making in Inter-group Conflict
Inter-group conflict requires that individuals make costly contributions to their in-group’s capacity to attack out-groups, and/or to defend the in-group against out-group attacks. We model such intergroup attacker-defender conflicts as a two-level multi-round contest, and showed that individual’s costly contributions to out-group attacks are lower than to in-group defense, that out-group attacks are less well-coordinated, and that therefore out-group attacks rarely result in victory. I will discuss on-going work in our laboratory focusing on the neural (e.g., prefrontal control and reward processing), hormonal (e.g., oxytocin), and social (e.g., leading-by-example;   group-bonding   rituals)   mechanisms   predicting   individual-level   contributions   to,   and   group-level coordination of out-group attack and in-group defense.

Tali Sharot (University College London)
Using Affect to Predict Choice
A common assumption in  psychology and behavioural economics is that feelings guide choice. Very little is known, however, about  the  computational  rules  by  which  feelings  are  transformed  into  decisions.  Here,  I  draw  on  basic  principles  from economics (e.g., the law of diminishing returns) and neuroscience (e.g., neural adaptation) to understand how affect (i.e., a physiological reaction that is experienced as a feeling) is translated to choice. I will present two recent studies in which we measure and quantify affect in order to predict decisions to (i) lie in self-interest and (ii) punish a selfish agent. In the first study we show that the negative affective response to ones’ own selfish behavior is reduced with repetition, predicting an escalation in such behavior over time. In the second study we show that the affective response of a person observing a selfish agent can be modeled using principles adapted from prospect theory. This model is then used to predict whether and by how much the observer will punish the agent. The work illustrates that understanding the dynamics of affect provides novel insight into decision-making.

Jean-Baptiste André (Ecole Normale Supérieure)
Fairness as an « incentive landscape  » for cooperation
Evolutionary theoreticians understand since the 70s that natural selection can lead to the evolution of cooperation. Yet, so far, they have been less successful in explaining the fine grained properties of this cooperation: Why do we cooperate the way we do and, in particular, why is our cooperation governed by the logic of fairness? In this talk, I will present modeling work, done with Nicolas Baumard, focusing on this second question. I will argue that, in a cooperative species, individuals are under a specific selective pressure: they must be good at attracting cooperation. To do so, human beings have evolved a mental organ, the sense of fairness, whose function is to make it adaptive for others to cooperate, that is to incentivize their cooperation.I will show that this view of fairness (i) explains the most important properties of our cooperation such as our intuition that people should be rewarded in function of their efforts and talent, (ii) resolves apparent moral paradoxes such as the fact that fairness often consists in refusing the consequences of power-struggle, and (iii) can account for the variability of human cooperation across societies.

Lou Safra (Sciences Po)
Quantifying the rise of trust in history using face-recognition machine learning
Compared to ancient societies, industrialized countries are characterized by a high level of social trust and a relatively low level of inequality. In the social sciences, this culture of trust and equality is usually regarded as the result of the institutional and religious revolutions. However, recent advances in in behavioural ecology have demonstrated that social preferences, which have long been seen as fixed, actually respond to environmental pressures such as resources scarcity. More specifically, when resources are scarce, individuals are more risk averse in their social relationships, they trust other less and they decrease their their investment in social cooperation. The aim of this project is to put to test the ecological theory of cultural difference by using the tools of affective sciences and test whether the evolution of prosocial preferences parallels changes in standards of living. To do so, we build on recent work showing that prosocial preferences are consistently associated with facial and emotional displays. We first developed an algorithm that models provides humans’ ratings of photos regarding trustworthiness. We then applied this model to two series of portraits of historical: the National Portrait Gallery (N = 1947 portraits from the United Kingdom) and the Web Gallery of Art (N = 5538 portraits from 28 countries). Both series confirmed the ecological model of trust:   an increase in trustworthiness, correlating with the improvement of living standards. Together these results provide further evidence of the role of ecological variables in explaining individual differences in cooperative attitudes.

Matthew Apps (University of Oxford)
Opportunities, costs  and  apathy:  A  computational neurology approach  to  behavioural and social motivation
Apathy - a quantifiable reduction in goal-directed behaviour – is one of the most common symptoms in Neurology, particularly in Parkinson’s Disease )PD), coming at a major cost to society patients and carers. Apathy is also present in a milder form in the healthy population and is comprised of three independent dimensions: behavioural, social and emotional. Previous work has suggested that behavioural apathy is associated with both a reduced sensitivity to rewards (e.g. money) and a heightened sensitivity to effort, with more apathetic people less willing to put in effort to obtain themselves rewards – both functions linked to the brain’s dopaminergic systems. However, the mechanisms underlying the different dimensions of apathy are poorly understood.Is social apathy linked to the effort required for social acts? Is behavioural apathy linked to dopaminergic dysfunction? Here, using the tools of computational psychiatry I show that both behavioural and social apathy are associated with a heightened sensitivity to effort, with social apathy linked specifically to the willingness to put in effort into social behaviours. In addition, I show that behavioural apathy is linked to the role that tonic dopamine has in the signalling of rewards both in healthy people and PD. These findings provide some of the first insight into the different dimensions of apathy in health and PD.

Carolyn Parkinson (University of California Los Angeles)
The brain  in the social  world:  Integrating approaches from  social  neuroscience, psychology and social network  analysis
The cognitive demands of navigating large groups comprised of many varied, intense, and enduring social bonds are thought to have significantly shaped human brain evolution. Yet, much remains to be understood about how the human brain tracks, encodes, and is influenced by the structure of the social networks in which it is embedded. In this talk, I will review a variety of collaborative recent work that integrates theory and methods from experimental psychology, cognitive neuroscience, and social network analysis, as well as the theoretical motivation for combining these lines of inquiry. One set of research seeks to elucidate if, when, and how the human brain retrieves knowledge of familiar others’ structural positions in one’s social network when  encountering  them.  Related  research  tests  how  this  knowledge,  once  retrieved  by  the  brain,  goes  on  to  shape downstream processes and behavior, including attention to others’ mental states, interpersonal trust, and altruism. An additional set of studies tests if human social networks exhibit assortativity in how their members perceive, interpret, and respond to the world around them. Consistent with this possibility, inter-subject similarities of neural responses to naturalistic stimuli can be used to accurately predict the geodesic distance between individuals in their shared social network, such that friends have exceptionally similar neural responses to the world around them. All human cognition is embedded within social networks, but research on neural information processing within individuals has progressed largely separately from research on the social networks that those individuals inhabit. The set of findings to be reviewed in this talk suggests that integrating approaches from social network analysis and cognitive neuroscience can provide new insights into how individuals perceive, shape, and are shaped by the structure of their social world.

Maël Lebreton (University of Amsterdam & University of Geneva)
The Art of The Deal: a Reinforcement Approach to the Ultimatum Game
The ultimatum game is a simple social dilemma paradigm in which one player (the proposer) decides how much of an endowment to offer a second player (the responder). The responder then decides whether to accept the offer, in which case the endowment is divided as proposed, or reject the offer, in which case both parties receive nothing for that trial. From the proposer’s viewpoint, it is therefore necessary to find the lowest possible offer that would still be considered acceptable (or fair) by the receiver. In this talk, I will present results from a new behavioral task, where proposers have to find, by trial-and-error, how to optimize their offers to different types of responders.Our results show that, although proposers learn to adapt their offer to  responders  characterized  by  different  acceptance  functions,  this  learning  process  is  paradoxically  less  efficient when proposer interact with human rather than computerized responders. We then show that a modified version of simple reinforcement-learning algorithms capture the critical features of this learning. Finally, we used functional neuroimaging to investigate the neurobiological correlates of the critical variables identified by our computational model.Overall, our results shed new  lights  on  the  behavioral,  computational and neurobiological processes involved in learning the fairness norms which prevails in different environments.

Patricia Lockwood (University of Oxford)
Neurocomputational basis of prosociality and selfishness: a reinforcement learning approach
The question of whether humans are fundamentally selfish or prosocial has intrigued many disciplines from philosophy to economics for centuries. From small acts of kindness to major sacrifices, just how willing are humans to help others? Here I will use models derived from reinforcement learning to understand ‘prosocial learning’ namely, how people are able to learn which of their actions help others. I will then discuss how basic associative learning processes might also underlie our tendency to be biased towards self rather than other-related information. Neurally I will show how distinct portions of medial prefrontal cortex reflect social computations related to selfishness and prosociality, whereas classical areas implicated in reinforcement learning such as ventral striatum track domain general learning signals regardless of the social context. These findings could have important implications for understanding everyday social learning and decision-making and its disruption in disorders of social behaviour.

Jean Deaunizau (Brain and Spine Institute)
Playing with Theory of Mind: a (narrow) computational view on social cognition
A defining feature of human social cognition is our insight that others' behaviour is driven by their beliefs and preferences, rather than by what is objectively true or good for them. In fact, a great deal of our social interactions is concerned with guessing others' mental states. But what are the specific computational processes underlying such "mentalizing", if any? Can mentalizing sophistication be defined in computational terms? Does the adaptive fitness of mentalizing sophistication depend upon  the  type  of  social  interactions  (e.g.,  competitive  versus  cooperative)  agents  engage  in? What type of evolutionary pressure  eventually  drove  differences  in  mentalizing  sophistication  across  (here:  seven)  primate  species?  How  does mentalizing sophistication relate to the specific "cognitive style" of people suffering from autism spectrum disorders? These are the questions we address in this work, by combining computational modelling with behavioural investigations in dyadic games against online  artificial mentalizing agents.

Colin Camerer (Caltech)
Visual salience in matching and hide-and-seek games
We measure and study visual salience in two-player games, in which players both prefer to match choices of locations in images, or hide-and-seek (HS) games one prefers matching and the other prefers mismatching. Visual salience is predicted a priori from a computational algorithm based on principles from theoretical neuroscience and previously calibrated by human free gaze data. Salience is a strong predictor of choices, which results in a matching rate of 64% in two samples. Both seekers and hiders choose salient locations more often, though seekers also choose low-salience locations. The result is a “seeker’s advantage” in which seekers win about 9% of the games, compared to a mixed-Nash benchmark of 7%. A salience-perturbed cognitive hierarchy (SCH) model is estimated from the hide-and-seek data. Those estimated parameters accurately predict the choice-salience relation in the matching games. If levels of thinking in the SCH take time, then time pressure should bias choices toward salience and increase the seeker’s advantage. This prediction is confirmed.