How do we represent what others do and receive?
Social Agent Identity Cells in the Prefrontal Cortex of Interacting Groups of Primates. Science 2021
I’ve been fascinated with social interactions all my life. During my Ph.D. research, I discovered a neuronal signal related to who is responsible for self reward (Baez-Mendoza PNAS 2013). However, the problem with this interpretation is that the experimental design could not generate a neuronal signal that can differentiate between an individual and another agent.
To test a social agent identity signal, I needed to test three agents simultaneously. The task is deviously simple; three monkeys sit around a rotary table and take turns to offer a reward (such as an apple slice) to one of the other two monkeys. At the same time, we recorded single neuronal activity from a brain area known to play a role in social cognition — the dorsomedial prefrontal cortex or dmPFC. In the end, we found a much richer brain representation of social interactions in the dmPFC than what was initially anticipated. First, we found a signal related to the social agent identity for reward or action. This type of signal is a building block for social exchange. It is a representation of who performs an action or who receives a reward. Then, using machine learning algorithms, we could read what happened in the previous interaction from the neuronal population. Furthermore, we could decode from the neuronal population whom the animal would give an apple slice to before they even performed the action. This finding suggested that this brain area plays a role in strategic decisions, and helps an individual answer the question “with whom it is more advantageous to reciprocate”.
Together, these findings suggest that the dmPFC plays a vital role in mapping out our actions and outcomes, as well as mapping out what others are doing.
How do we compute other’s beliefs?
Strategic behaviors rely on predicting other’s actions. This ability is highly developed in humans, and is supported by theory of mind (ToM), which is also closely interlinked to our cognitive ability to distinguish self from specific others.
The litmus test of whether someone understands beliefs is whether they can predict (and possibly explain) an agent’s behavior when the agent has a false belief; that is, when the agent is acting rationally from his subjective perspective but irrationally from an objective perspective. We find that we could decode the difference between physical objects and other’s beliefs from up to 20% of recorded dmPFC neurons. The key of the false-belief test is the differentiation between other’s true and false beliefs, and we could decode this difference from 23% of single neurons. Finally, we evaluated if other belief neurons reflected imagined false beliefs in general but found little overlap in these two sets of neurons.
How do we decide to enter a competition?
This is a study led by William Li, who I co-mentored during his PhD at the Ziv Williams lab. My role in this paper was of a consultant and sounding board. First, Will and Ziv developed a clever new paradigm to test group competition in which a group of mice is released into an arena with the goal to be the first one to enter through a narrow passage and eat a food pellet… just like some people scramble to get a new flat-screen TV on Black Friday.
We varied who the animals competed against (higher or lower dominance), for how much (varying amounts of food), and the effort they needed to exert (distance traveled). We found that the relative rank of the mice was the major determinant of who won! We then recorded wirelessly from single neurons in the prefrontal cortex of these mice as they competed. These neurons integrated information about the competitors, the prize and the effort needed to ‘win’ to inform upcoming decisions. Together, these neurons could even predict the animal’s own future success well before competition onset, meaning that they likely drove the animals’ competitive behavior based on whom they interacted with. Manipulating the activity of these cells, on the other hand, could artificially increase or decrease the animal’s competitive effort and therefore control their ability to successfully compete against others. Collectively, these findings reveal neurons in the brain that drive competitive interactions between individuals. Competitive success is not simply a product of an animals’ physical fitness or strength, rather it is strongly influenced by neural signals in the brain that affect one’s own competitive drive.
How do we represent what others do and receive?
When I started my Ph.D. at Cambridge University with Prof. Wolfram Schultz, there were no studies or protocols to investigate the single-neuronal basis of social interactions. My curiosity motivated me to develop an innovative dual-monkey apparatus for behavioral testing and neurophysiological recording. This work resulted in four original research publications as a first and corresponding author. This paper is the most significant one from my PhD. When agents interact with their environment and obtain a reward, they are faced with discovering what sequence of actions led to such an outcome if they want to obtain it again. The same problem occurs when interacting with others; there is a need to associate others’ actions with positive or negative outcomes, which we call the social credit assignment problem. We set out to find neuronal signals that would differentiate between actions that lead to a reward by the self and by another agent. I recorded single neuronal activity from the anterior striatum of one monkey while it took turns with a conspecific to make an action to obtain a reward for both of them. I controlled for several variables, including reward contingencies, action cost, eye position, and animacy of the other (i.e. whether the other was a conspecific or an empty chair). In this paper we describe three main types of single neuronal responses. First, we replicated previous findings of neurons reflecting expected own reward (reward that the recorded animal will receive). Second, we found for the first time single neurons that differentiate between own and other’s actions in the striatum. Third, we discovered single neurons that represented the interaction between others’ actions and self-reward. These types of responses may be used to solve the social credit assignment problem.