Searching for Synergy in Shared Workspace Human-AI Collaboration

The study investigates the dynamics of shared-workspace human-AI collaboration, utilizing the Collaborative Gym environment with DiscoveryBench tasks. It finds that while adding simulated human collaborators can enhance performance, it can also lead to process loss if team coordination is insufficient. The introduction of scaffolding that incorporates shared group memory and human-in-the-loop (HITL) gates significantly improves performance, particularly in three-person teams, by clarifying responsibilities and optimizing the integration of expertise. This research underscores the importance of structured collaboration in maximizing the effectiveness of human-AI teams.