– SOCIALNET

Social networks play a crucial role in informal transmission of information across society. Although there have been extensive efforts to characterize the neural mechanisms involved in social decision making and learning, the focus has mainly been on dyadic interactions. In contrast, interactions in real-world social networks often occur between large numbers of individuals, creating additional complexity about how an individual incorporates information from their neighbors, neighbor’s neighbors, etc. Such learning has been successfully described using the DeGroot model that relies on one-shot averaging of opinions across one’s neighbors in the network, however, the underlying brain mechanisms are unknown. Therefore, a critical gap in our knowledge of social learning is the understanding of the neurocomputational mechanisms that underlie the learning and integration of information from others during multiple interactions in social networks. In particular, it remains unclear how the one-shot averaging process assumed in the DeGroot model differs from the sequential, error-driven computations observed in reinforcement learning.
A deeper understanding of the neurocomputational mechanisms of social learning in social networks requires identifying both fundamental computational principles as well as the brain systems that perform those computations. By leveraging computational modeling and model-based neuroimaging experiments, our project (SOCIALNET) will lay the foundation for a comprehensive multidisciplinary understanding of the psychological and neurobiological bases of learning in social networks. Specifically, the goal of SOCIALNET is to characterize the brain computations underlying the spread of information in large social networks, thereby connecting computational mechanisms within brain systems to emergent collective behaviors. We propose to combine state-of-the-art computational approaches that contrast Reinforcement Learning (RL) and DeGroot models and integrate predictions of these models with model-based fMRI to unravel the brain computations of learning in social networks. Specifically, we will determine: (1) how uncertainty and volatility of the environment and the topology of the social network influence the adoption of one-shot averaging in the DeGroot model vs. error-driven learning in RL; and (2) brain areas involved in computations required in these two types of learning, and arbitration between them. These allow us to establish a mechanistic understanding of the brain computations that drive learning within social networks and will offer a multi-level perspective on information transmission, from the brain system-level to the levels of individual and collective behaviors.