Not yet human-reviewed
This source record was written automatically by the research pipeline and has not yet been reviewed by the maintainer or integrated into a topic page. Treat it as a provisional summary and verify claims against the original source.
Toward a science of human–AI teaming for decision making: A complementarity framework
Citation. Gonzalez, C., Donahue, K., Goldstein, D. G., Heidari, H., Jalali, M. S., Schelble, B., Singh, A., & Woolley, A. W. (2026). Toward a science of human–AI teaming for decision making: A complementarity framework. PNAS Nexus, 5(3), pgag030. https://doi.org/10.1093/pnasnexus/pgag030
Summary. This peer-reviewed perspective synthesizes cognitive science, AI, human factors, organizational behavior, and ethics to define human–AI complementarity as team performance that exceeds either humans or AI alone. It organizes a framework around reasoning, memory, and attention and identifies sociotechnical conditions—such as role partitioning, calibrated trust, shared mental models, training, and task structure—that shape whether complementarity occurs. It recommends design and evaluation practices that preserve human accountability, transparency, and equity in high-stakes decision settings.
Key claims
- Human–AI complementarity is context dependent: hybrid teams can outperform either partner alone when collaboration is well calibrated, but poorly designed interaction or overreliance can produce worse performance than either working independently.
- A team design should define goals and constraints, partition roles around complementary strengths, orchestrate attention and interrogation, build shared knowledge infrastructure, and continuously train and evaluate the human–AI system.
- Evaluation should assess collaboration quality, workload balance, resilience, trust calibration, shared understanding, and ethical alignment alongside task accuracy.
- The authors argue that human oversight remains necessary where ethical discretion, contextual sensitivity, and accountability are central, while AI can contribute speed, consistency, scalability, and high-dimensional pattern recognition.
Evidence & limitations
- This is an interdisciplinary perspective/framework paper that synthesizes prior evidence rather than reporting a new experiment or dataset.
- The authors characterize existing evidence on complementarity as mixed and context dependent, with much evidence drawn from laboratory studies, small deployments, or early prototypes; long-term, high-stakes, and heterogeneous field settings remain underexamined.
- Published open access under CC BY 4.0 in PNAS Nexus.