Think Like a Team: Graph-based Representation of Shared Mental Models in Human-Agent Collaboration

Authors: Moinam Chatterjee, Behnam Ensan, Andreas Wendemuth, Ayoub Al-Hamadi

Abstract:

We introduce a conceptual temporal Shared Mental Model (SMM) that maps changing human-human and human-agent team dynamics, allowing Human-Agent Teaming Systems (HATs) to achieve shared teaming goals. We use temporal graph neural networks (TGNNs) to capture the evolving roles and tasks within the team, learnt from time-stamped team interactions. We conduct a proof-of-concept exploratory small-scale user study to observe, in real time, the evolution of team dynamics and interrelationships among team members. This study simulates collaborative tasks involving human and AI agents, enabling direct observation and measurement of teaming behaviours. The proposed model bridges the current research gap in HATs and SMMs by offering a graph-based representation of agentic teaming dynamics.