US-German Collaboration: Computational Modeling of Cooperative Success using Neural Signals and Networks

Scripps College, Claremont CA

Investigators

Abstract

Societal progress requires humans to excel at cooperation over time. When people cooperate successfully, everyone on the team comes together to achieve a goal while being fair to the entire team. Successful cooperation ideally includes achieving the good of each person on the team, as well. To sustain successful cooperation, people need to coordinate what they see, hear, think, and do, especially about who is in the best position at any given moment to make the best decision or to take the best action for the team as a whole. Expertise changes with changing circumstances, and team members need to respond quickly and flexibly. This project will enhance understanding about how the mind and brain work to initiate and sustain successful cooperation in a rapidly changing world. The research will reveal aspects of the mind and brain that best support winning strategies for cooperative success. It will combine computational models of team coordination with measurements and models of brain networks during successful cooperation. The outcomes of the research could help enhance innovation in teams for technology design, assist centers of disease control to contain sudden outbreaks, and create new methods to help robotic assistants better respond to human direction. The research proposal seeks to answer the following primary questions: 1) What is a team member's mental representation of other agents with whom that person is cooperating, especially during periods of cooperative success?; 2) How do these representations contribute to knowledge of others' decision-making processes and changing expertise?; and 3) What are the neural signals that predict cognitive updating and successful cooperative action? To model the cognitive processes during cooperation requires explicitly modeling social reasoning capabilities, especially capacities to think about other partners's decision process and their sophistication in understanding both the task and communication among team members. These features cannot be observed directly and will be estimated using computational approaches for multi-agent settings (i.e., interactive partially observable Markov decision processes). The underlying neural networks will emerge via dense array EEG and analysis of functional connectivity patterns within and across brains. Network connectivity measures include the circular correlation coefficient and time-varying, adaptive multivariate autogressive modeling, using independent components that covary strongly with choice outcomes and/or cognitively modeled parameters. These measures of between-brain connectivity are less susceptible to spurious correlations induced by underlying task demands. A companion project is being funded by Federal Ministry of Education and Research, Germany (BMBF).