A POMDP Framework for Social Learning of Identity and Trustworthiness

Abstract

The amygdala has traditionally been studied through fear conditioning paradigms, positioning it primarily as a “fear center” responsible for detecting threats and orchestrating fear responses. However, this classical view has been challenged by inconsistent replicability, and emerging research indicates that the amygdala's role extends beyond fear processing, such as in trustworthiness assessment and social decision-making [1], [2]. The amygdala may be better conceptualized as part of a hierarchical predictive coding system, where the central amygdala and the basolateral amygdala collectively form a dynamic system for integrating interoceptive and exteroceptive information [3]. 

This thesis therefore supports a research project, with the overarching aim of developing a new neuroimaging paradigm to provide a more nuanced account of the role of the amygdala and associated brain circuits in social learning. This paradigm moves beyond passive fear conditioning and emotion discrimination tasks, to include active learning about identity (e.g., name, country of origin) and personality (trustworthiness).

My role in this project will be to develop a computational model using the partially observable Markov decision process (POMDP) framework, guided by active inference principles. The study aims to simulate how individuals learn about others under uncertainty. Participants will learn facts about their opponent’s identities, while simultaneously learning how trustworthy they are in a betting game. The computational model of this paradigm will generate trial-by-trial computational variables such as belief states and prediction errors, which may in the future be correlated with functional MRI data to elucidate the neural mechanisms underlying social learning.

A central hypothesis is that higher uncertainty will predict increased exploratory actions, and that this reflects the amygdala's role in balancing exploration and exploitation during social decision-making. By operationalizing constructs like identity and trustworthiness within this computational framework, the research aims to provide a more nuanced understanding of the amygdala's involvement in social cognition.

References

[1] R. Sladky, F. Riva, L. A. Rosenberger, J. Van Honk, and C. Lamm, “Basolateral and central amygdala orchestrate how we learn whom to trust,” Commun. Biol., vol. 4, no. 1, 2021. doi: 10.1038/s42003-021-02815-6.

[2] R. M. Visser, J. Bathelt, H. S. Scholte, and M. Kindt, “Robust BOLD responses to faces but not to conditioned threat: challenging the amygdala’s reputation in human fear and extinction learning,” J. Neurosci., vol. 41, no. 50, pp. 10278–10292, 2021. doi: 10.1523/jneurosci.0857-21.2021.

[3] R. Sladky, D. Kargl, W. Haubensak, and C. Lamm, “An active inference perspective for the amygdala complex,” Trends Cogn. Sci., vol. 28, no. 3, pp. 223–236, 2023. doi: 10.1016/j.tics.2023.11.004.