Hybrid human-AI teams: New framework maps collective intelligence

A new preprint on arXiv (2607.05593) from researchers Babak Hemmatian, Razan Baltaji, and Lav R. Varshney tackles a fundamental gap in our understanding of hybrid intelligence: how collective cognition scales from individuals and dyads to entire groups when humans and AI agents work together. While existing frameworks have clarified attention, memory, and reasoning differences at the individual and pair level, a formal account of group-level dynamics has been missing. Most network science studies focus on either human-only or AI-only systems, leaving the translation to hybrid teams unclear.

The authors bridge this divide by synthesizing network science, collective cognition, and multi-agent systems through the lens of attention, memory, and reasoning. They review how task environments, group topologies, agent-level processes, and incentive structures shape outcomes in homogeneous networks, then extend those findings to heterogeneous human-AI groups. Crucially, they treat hybrid networks as nodes and links with distinct individual and transactive constraints—e.g., humans and AI have fundamentally different capacities for memory and attention.

The comparative analysis reveals which network effects are robust across agent types and which require revision. A standout finding is the structural centrality of configurations that were peripheral in single-type traditions, such as human gatekeepers managing AI subnetworks. The paper also shows that classic exploration-exploitation and efficiency-redundancy trade-offs operate differently in hybrid teams, suggesting new design principles for organizations. This work is a non-authoritative author's version of a forthcoming chapter in the Springer Nature Handbook of Hybrid Intelligence.