Skip to content Skip to navigation

Neural computations underlying social interactions - Keren Haroush

Keren Haroush
January 13, 2020 - 5:10pm
Sloan Hall, Math Building 380, Room 380-C

Keren Haroush

Stanford University

Abstract

A cornerstone of human interaction is the ability to build internal models of other individuals in our environment, based on our past interactions, which in turn enable assessing and predicting another individual’s current hidden state of mind, for example, what other individuals are thinking or feeling. Such predictions are key for successful social engagement, mutual reciprocity and cooperative behavior, the glue that holds together our societies. Yet, despite their importance, how social prediction computations are implemented at the single-neuronal and population level, and their causal underpinnings have remained a mystery. This presents a major roadblock to the development of neural circuit-based therapies for an array of neurological and psychiatric disorders in which social interaction deficits are a debilitating factor. Key to our unique approach for rendering the complex psychological problem of predicting another’s hidden state of mind a biologically tractable question is using game theory to provide a mathematically driven, well-controlled encapsulation of real-world interactions. Specifically, we adapted the canonical iterated Prisoner’s Dilemma (iPD) game in which each agent can choose to cooperate or defect on each trial. Critically, as one’s outcome depends on the other’s decision in a series of repeating encounters, anticipating the other’s intention and upcoming choice is key to one’s success. Temporal separation between a monkey’s choice, the time the opponent’s decision was revealed, and delivery of reward, allowed to explicitly dissect the neuronal signals that predict the other’s yet unknown decision from one’s own concurrent choice, while dissociating past responses, social context, reward expectancy and outcome. Using this approach, we discovered that monkeys play similarly to humans, using mixed strategies, and identified a previously unknown class of “other-predictive” neurons in the dorsal Anterior Cingulate Cortex (dACC), that signal the opponent’s choice well before the other’s decision. These neurons distinguish between self and other agency and are modulated by social context, likely constituting a key circuit component for social prediction. Disruption of dACC activity using microsimulation selectively biased mutually beneficial interactions but, surprisingly, had no effect on their decisions when no net-positive outcome was possible. These result indicate that the cingulate plays an important role in guiding animals’ behavior in social context. Following up on this work, we found that similar to humans, inhaled Oxytocin enhances cooperative social behavior in monkeys. Surprisingly, Oxytocin decreased the prevalence of other-predictive neurons in the dACC and weakened the population prediction of the other’s upcoming decision, further corroborating the role of other-predictive neurons. Specifically, neuronal prediction was correlated with the monkeys’ own decision following Oxytocin but not following saline. These results shed light on the neural basis by which Oxytocin influences interactive social behavior, with implications for the targeted treatment of disorders such as ASD and schizophrenia. To complement our non-human primate results, we investigate the representation of the other in humans. We will discuss our work recording single neuronal activity in intraoperative patients and using deep brain stimulation to further dissect how interconnected brain regions such as the dorsolateral prefrontal cortex and the periaqueductal gray take part in building the internal subjective representation of other agents. Together, this body of work begins to delineate the brain-wide neuronal foundation of social prediction.

Curriculum vitae

Related Articles

[1] Neuronal Prediction of Opponent’s Behavior during Cooperative Social Interchange in Primates