Speakers

Daniel Houser

Interdisciplinary Center for Economic Science (ICES), George Mason University

Working for Others: Theory of Mind in the Absence of an Intentional Agent

(Joint work with Steven Saletta, Frank Kruger, Erte Xiao and Kevin McCabe)

Intentionality detection is frequently identified during interpersonal exchange. Previous studies in this area have focused on strategic interactions between a decision-maker and a counterpart. How a potentially altruistic decision-maker evaluates outcomes accruing to another person who not present during decision making (and who cannot reciprocate) remains an open question. Using fMRI, we implemented a modified dictator game without a punishment option, augmenting or replacing money cost with effort. In a delayed match to sample task, subjects (n=18) earned money for themselves or a known counterpart who was not present during decisions and who had no opportunity to reciprocate or influence payoffs. Comparing outcomes accruing to counterpart versus those accruing to self yielded activation in regions previously identified in mentalizing tasks and two-person interactive games: medial prefrontal, superior temporal, and temporal pole regions. We argue these regions respectively bind shared mental intentions, social identity, and social feature knowledge to predict how outcomes are evaluated by the counterpart. These results shed important new light on the biological foundations of human altruism.

Rosemarie Nagel

Universitat Pompeu Fabra

Assessing strategic risk with fMRI

(Joint work with Giorgio Coricelli (CNRS Lyon), Andrea Brovelli (CNRS Marseille) and Frank Heinemann (TU-Berlin))

We used fMRI to measure the neural correlates of strategic uncertainty in games and risk in lotteries. Participants played a series of stag hunt games, entry games, and lotteries, all framed in a similar way. The two games differ in their equilibrium properties: stag hunt games are games of strategic complementarity (e.g., an investment pays off if and only if a sufficient number of agents invest in the same industry, so all invest and nobody invest are two Nash equilibria) while entry games are of strategic substitutability (e.g., if too many agents invest in a new market all get nothing; here we should not all do the same, but instead choose mixing strategies in equilibrium). A mentalizing network (mPFC, TPJ, STS, precuneus) is activated in games playing vs. Lotteries, thus distinguishing the social and the private nature of the choice context. Furthermore, we found a behavioral correlation and a similar pattern of activity in the striatum between choosing lotteries and choosing the stag hunt game; while insula and lateral OFC activity was mainly related to entry games choices. Interestingly, we found a clear separation of insula activity in lotteries and stag hunt games when distinguishing between risk averse and risk loving players. However, in entry games this distinction is not at all found. We conclude that the entry game creates more strategic uncertainty as predicted by the nature of the theoretical equilibrium which also involves levels of reasoning. While the strategic uncertainty of the stag hunt game can be “reduced” to standard risk, the uncertainty underlying entry games is higher and analogous to ambiguous choices.

Aldo Rustichini

University of Minnesota

Strategic Theory of mind in young children

(Joint work with Melissa Kenig and Itai Sher)

Having a Strategic theory of mind (SToM) is defined as the ability to think about the behavior and the thinking of the others, based on the information you have on the other's incentives and the assumption that the others pursue their interest as you do. SToM is different from ToM because it has to integrate the information on incentives with other information that players have.

We test the development of SToM in young children, age 3 to 8, on the basis of two simple games, male and female. The games are simple enough that can be understood at all ages. We also gather information on thinking ability.

The development of effective strategic behavior is similar in the two games, suggesting a general ability for recursive thinking. The difference in performance across ages is clear in the first move, which suggests that this ability is different from ability to learn. We also find that the age in which SToM skill develops is substantially higher than the age in which the non strategic ToM develops.

Older children (7-8 years) develop ability to think about cooperation in repeated games. In the talk we will compare these results with behavior of adults in a repeated game of differential information.

Peter Shizgal

Center for Studies in Behavioral Neurobiology, Concordia University

Valuation of opportunity costs by laboratory rats

My research team and I study the neural, behavioral, and computational mechanisms underlying the evaluation, selection, and pursuit of goal objects by laboratory rats. Direct electrical stimulation of particular brain sites produces a powerfully rewarding effect that leads the rats to work hard in order to procure additional stimulation. The rewarding effect arises from an observable volley of nerve impulse propagating along the axons of identifiable neurons. Thus, this preparation provides an entry point for tracing brain reward circuitry and for studying its interaction with modulators of reward, such as abused drugs and hormonal signals that regulate energy balance. Unlike natural rewards, “consumption” of the electrical reward does not lead to satiety or habituation, making it possible to study reward-seeking behavior over long periods of time under stable, highly controlled conditions.
Although game-theoretic ideas and paradigms have been applied successfully in research on brain reward circuitry in non-human primates, research on brain stimulation reward in rodents has been carried out typically outside of a game-theoretic context. The rodent interacts with a computer-controlled “agent” that does not take into account the strategy employed by the experimental subject and implements instead a fixed “schedule of reinforcement,” a set of rules for delivering rewards according to the amount of work performed and/or the distribution of responses over time. That said, an agent that does take account of the strategies employed by the other agents with which it interacts should benefit from forming an accurate predictive model of the algorithms used by these agents to evaluate benefits, costs and risks. If so, insights into the core processes involved in the evaluation, selection, and pursuit of goals may be germane to the concerns of game-theory researchers.
I will describe experiments on the evaluation of opportunity costs by rats working for rewarding brain stimulation. By allocating time spent working for the electrical reward, the rat must forgo benefits that would accrue from competing behaviors, such as grooming, resting and exploring. We have estimated the psychophysical function that translates the objective opportunity cost into its subjective equivalent. This function has interesting properties. When opportunity costs are very low (e.g., working for only a small fraction of a second triggers delivery of a reward), the subjective cost is invariant. Although the rat may be capable of detecting the difference between 0.1 s and 0.2 s, it does not appear to increment its subjective cost function over such intervals. This makes sense in that there may be no beneficial activities that the rat could substitute for work over such short intervals. As the objective opportunity cost rises, the subjective cost begins to rise as well, and beyond values of ~4 s, we estimate the subjective cost to rise at the same rate as the objective cost. This make sense as well given that a steeper rise in subjective costs would cause the rat to forgo very large expensive rewards that are profitable to pursue, and a more gradual rise would cause the rat to chase smaller, expensive rewards that are not worth the trouble. The form of this “subjective opportunity-cost function” is quite different from the form predicted by hyperbolic temporal discounting. I will propose that time spent working for reward is evaluated differently than time spent waiting for reward after a work requirement has been satisfied. I will also discuss the implications of these results for theories that hold psychological time to be either a scalar or logarithmic function of objective time.