Date created: | Last Updated:

Category: Project

Description: When an individual is facing a choice between two options and chooses one alternative over another, what can we infer from this decision? This study aims at investigating the computational mechanisms underlying our inference of other people’s preferences. More specifically, can we learn and infer a person’s social preferences by observing their choice process? Observational social learning can be broadly divided into two approaches: imitation, where the learner copies the observed action, and emulation, where the learner decomposes the observed behaviour into a set of primitives - value of the action, beliefs, and rewards - that dictate the expected value of an action with respect to some goal (Wu et al., 2021). While the former approach is often considered computationally cheaper and simpler compared to other forms of social learning (since no inference is needed about the other person’s goals or intentions), the latter one implies that observed actions first need to be “unpacked” into inferred value representations. This defines a hierarchy of inferences that an observer can draw from a demonstrator’s actions, and a corresponding set of mental representations that the observer can potentially adopt via social learning. In the social- and reinforcement- learning literatures, whether and how one can learn another’s preferences from observing their choices has been well documented (Joiner et al., 2017; Vostroknutov et al., 2018; Charpentier et al., 2020). Yet, the investigation of choices only cannot allow one to infer the strength of the preference, i.e., the confidence with which the person has made their choice or how likely they are to make the same choice again. But choice itself is not the only output of the decision process: the decision process also produces response times, which have been shown to decrease with the strength of preference (Busemeyer, 1985; Alós-Ferrer et al., 2016). In other words, choices between equally-liked options tend to take more time. In comparison to the study of choices, the investigation of how response times impact our learning of other’s social preferences has received little attention. In this study, we aim at filling this gap by proposing an orthogonal design to investigate the involvement of both choices and response times in learning and inferring other’s social preferences. In this task, participants will observe other people’s decision processes in a Dictator Game (Forsythe et al., 2014), where the demonstrators (here referred to as dictators) were asked to choose between different points allocations. In the Dictator Game, participants can be ranked on a selfishness scale, from selfish (choosing the allocation with the higher number of points for themselves) to altruistic (choosing the allocation with the lower number of points for the themselves). Consequently, we assume that dictators’ position on this scale can be determined by their preferred allocation: the allocation of points for themselves and the other that they would like the most. Therefore, a decision problem with two options equally distant from the preferred allocation is equivalent to a choice between two equally-liked allocations, and the response time should increase. When participants will observe the dictators’ decision processes, we will manipulate the amount of information provided to them. They will either have access to both the choices and response times, only choices, only response times, or neither, leading to an orthogonal within-subject 2x2 design. Based on previous literature (Konovalov and Krajbich, 2019), we expect participants to be able to learn, trial by trial, each dictator’s hidden social preference (i.e., their preferred allocation) within a few trials. We expect an ordered profile of prediction accuracy, which is expected to be the highest when complete information is provided to the participants, then choice only, response time only, and finally no information (where the accuracy is expected to be self-biased, i.e., to tend towards the participant’s own preference). Participants will also perform a time-perception task (Broadway et al., 2011), where they will be asked which of two stimuli was presented for the longest time. We expect the accuracy in this task to be correlated with their ability to correctly estimate the dictator's preferred allocation, especially in conditions where the response times are displayed. Finally, in order to estimate participants' own preferred allocation, they will perform the dictator game themselves, and they will be asked to fill the Social Value Orientation scale (SVO, Murphy et al., 2011).