Social learning—the ability to acquire relevant knowledge, beliefs, or behaviors in the presence or context of others—is a fundamental human skill. Over the past decade, computational models of learning, with formal quantifications using mathematical equations and latent variables/parameters, have offered remarkable insights into both the behavioral and the neural foundations of this essential process. Today, these models in conjunction with bio-psychological methods help uncover the mechanisms underlying various complex social behaviors, underscoring the richness and versatility of this approach.

This symposium brings together scientists across different career stages from three countries (Austria, Germany, and UK) and three different universities (Universities of Vienna, Würzburg, & Birmingham) who investigate bio-psychological mechanisms of social learning in a variety of social contexts.

Our symposium explores Bayesian inference in a novel social learning task and its connection to autistic traits. (Bianca Schuster). Additionally, we examine the neural mechanisms underlying emotional and reward prediction errors in social transgressions (Lei Zhang). We will discuss how interactive gaze fosters closeness through learning (Julian Gutzeit), and a final presentation will highlight how financial incentives and empathy shape social behavior, integrating fMRI and reinforcement learning models (Anne Saulin).

Together, the presentations in this symposium provide a multidisciplinary perspective on the computational and neural mechanisms of social learning. Collectively, we will discuss and integrate these diverse methodologies and modelling approaches to illustrate how social learning models enhance our understanding of key social-cognitive processes and their role in different social contexts.

The social environment is inherently uncertain, requiring individuals to interpret ambiguous and noisy social cues. Hierarchical Bayesian inference, mathematically accounting for multiple levels of uncertainty, provides an optimal framework for investigating how social agents deal with such noisy information. Yet, few experimental studies have used Bayesian models to explicitly capture how individuals process social uncertainty. Moreover, although Bayesian theories of autism - proposing atypical integration of priors and sensory evidence – have become increasingly popular, findings are highly inconclusive. To examine how human agents take into account various sources of uncertainty while learning about others, we designed a novel social learning task wherein prior- and sensory uncertainty were manipulated. In this pre-registered study, 49 non-autistic participants were first presented with either ambiguous or unambiguous information about four fictional characters’ risk proneness and subsequently predicted each character’s bets in a gambling task, with feedback displayed at varying levels of sensory precision. Influences of autistic trait level on uncertainty processing were assessed using Autism Quotient (AQ) data. Analyses showed that participants did take into account the prior information and prior uncertainty: Modelling responses with the Hierarchical Gaussian Filter revealed higher belief volatility (ω) for high-, relative to low-uncertainty prior belief conditions. Contradicting prominent theories, higher autistic traits were associated with a stronger influence of prior belief uncertainty on initial learning. Our results suggest that individuals’ propensity to learn about others depends on the uncertainty of their prior beliefs, and that those with higher autistic traits are more sensitive to this kind of uncertainty.

Traditional decision-making models conceptualize humans as adaptive learners utilizing the differences between expected and actual rewards (prediction errors, PEs) to maximize outcomes, but rarely consider the influence of violations of emotional expectations (emotional PEs) and how it differs from reward PEs. Here, we conducted a fMRI experiment (n = 43) using a modified Ultimatum Game to examine how reward and emotional PEs affect punishment decisions in terms of rejecting unfair offers. Our results revealed that reward relative to emotional PEs exerted a stronger prediction to punishment decisions. On the neural level, the left dorsomedial prefrontal cortex (dmPFC) was strongly activated during reward receipt whereas the emotions engaged the bilateral anterior insula. Reward and emotional PEs were also encoded differently in brain-wide multivariate patterns, with a more sensitive neural signature observed within fronto-insular circuits for reward PE. We further identified a fronto-insular network encompassing the left anterior cingulate cortex, bilateral insula, left dmPFC and inferior frontal gyrus that encoded punishment decisions. In addition, a stronger fronto-insular pattern expression under reward PE predicted more punishment decisions. These findings underscore that reward and emotional violations interact to shape decisions in complex social interactions, while the underlying neurofunctional PEs computations are distinguishable.

Social gazing is a core form of nonverbal communication, with direct eye contact often linked to heightened closeness and trust. Recent research shows that impressions of others develop through repeated interactions, guided by learning processes. However, it remains unclear how repeated gaze interactions shape feelings of closeness and interaction partner preferences over time, and whether these dynamics are driven by reinforcement learning (RL).

In our preregistered online study, n = 106 participants completed 100 trials in which they repeatedly chose one of two female interaction partners, each represented by a face image. The selected partner then either made eye contact or averted her gaze in response. Unbeknownst to participants, one partner made direct eye contact in 70% of the trials in which she was selected, while the other did so in only 30%. After each trial, participants rated how close they felt to each partner.

As hypothesized, participants’ closeness ratings and preference for the partner that made eye contact more frequently increased over time. RL-models showed that direct gaze served as a social reward, driving preference formation through RL, although this reward experience was recalibrated for certain individuals, making eye contact less and averted gaze more rewarding. Importantly, latent Q-values from the best-fitting RL-model predicted trial-wise closeness ratings, suggesting that rewarding eye contact fosters social closeness over time. We discuss these findings in relation to individual differences such as social anxiety and personal dominance.

Prosocial behavior is the glue of society, yet its persistence depends on the underlying motivation—some drivers sustain it, while others fade quickly.

We present two preregistered projects examining the persistence of prosocial behavior driven by empathy for pain (3 studies, https://osf.io/yz9rq/, N_total=104) or monetary incentives (3 studies, https://osf.io/4uvqj/, N_total=195) using model-based fMRI.

In the empathy project, participants completed a baseline block of a binary dictator game, followed by a second block after frequent pain observations and a third after rare pain observations to assess persistence. The monetary incentives project used a reinforcement-learning (RL) approach: in the first block (acquisition), prosocial choices were rewarded (20 cents, 80% probability) whereas rewards were withheld in the second (extinction).

Results showed prosocial behavior remained high across all blocks for empathy, while monetary incentives initially increased prosocial decisions, followed by a decline when payments stopped. Drift-diffusion modeling (DDM) revealed an increased predecisional prosocial bias after frequent pain observation, which persisted after observed rare pain. This parameter of the DDM was also linked to neural activation in the dorso-medial prefrontal cortex.

When using money as prosocial motivation, preliminary combined RL DDM analyses suggest different learning mechanisms for when prosocial behavior was incentivized (acquisition) versus when rewards were removed (extinction). These differences may relate to neural activity in a network including the anterior cingulate cortex, striatum, and anterior insula.

Together these projects shed light on via which neuro-behavioral mechanisms, different motivations can or cannot drive persisting prosocial behavior.