Structure and influence in an interconnected world: neurocomputational mechanism of real-time distributed learning on social networks

Many social species are embedded on social networks, including our own. The structure of social networks shapes our decisions by constraining what information we learn and from whom. But how does the brain incorporate social network structures into learning and decision-making processes, and how does learning in networked environments differ from learning from isolated partners? Combining a real-time distributed learning task with computational modeling, fMRI, and social network analysis, we investigated the process by which humans learn from observing others decisions on 7-node networks with varying topological structures. We show that learning on social networks can be realized by means similar to the well-established reinforcement learning algorithm, supported by an action prediction error encoded in the lateral prefrontal cortex. Importantly, learning is flexibly weighted toward well-connected neighbors, according to activity in the dorsal anterior cingulate cortex, but only insofar as neighbors actions vary in their informativeness. These data suggest a neurocomputational mechanism of network-dependent filtering on the sources of information, which may give rise to biased learning and the spread of misinformation in an interconnected society.