Cortical substrates of dynamic social interactions

Daeyeol Lee
Department of Neurobiology, Yale University School of Medicine

During repeated social interactions, decision makers adjust their strategies incrementally to 
approximate optimal strategies, often using simple, model-free reinforcement learning 
algorithms. In addition, such simple learning algorithms can be complemented by high-order 
strategies to evade the exploitation of their opponents. Here, we analyzed the choice behaviors 
of rhesus monkeys performing a biased matching pennies game against a computer. We found 
that the predictions of simple reinforcement learning were systematically violated following 
specific sequences of choices and outcomes that prompted the computer opponent's exploitative 
algorithm. Thus, monkeys used high-order strategies to counter the opponent’s strategies. The 
information about specific choice-outcome conjunctions constituting such high-order strategies 
were found most frequently in the dorsomedial prefrontal cortex.  Furthermore, switching-
related in this cortical area was correlated with the tendency to deviate from the choices 
predicted by a simple reinforcement learning model. Therefore, the medial frontal cortex might 
play an important role in implementing high-order strategies during dynamic social interactions.