As artificial intelligence (AI) plays a larger role in choices about health, safety, finance, and government, the main question is how to make human and AI work together well. A new paper titled “Toward a Science of Human–AI Teaming for Decision Making: A Complementarity Framework” offers a clear way to understand and build these partnerships. The work was published in PNAS Nexus.
The paper draws on studies of collective intelligence. It centers on three main mental activities: reasoning, which means thinking logically to solve problems; memory, which is storing and recalling information; and attention, which is focusing on what matters. These activities can be shared between people and AI systems so each handles what it does best.
Complementarity happens when the combined human-AI effort produces better outcomes than humans working alone or AI working alone. The authors describe the conditions needed for this to occur. These include how the partnership is put together, how trust is balanced, how everyone shares the same understanding, how training is done, and how tasks are organized.
Key conditions and design approaches for success
Several design principles help reach this complementarity. They involve setting clear goals and limits, dividing responsibilities, guiding what receives focus, creating systems to share knowledge, and keeping up ongoing training and checks. The framework gives people a shared language to spot where partnerships will likely work, where they might struggle, and how to fix problems.
The paper discusses effects for theory, real-world use, and government rules. It stresses keeping AI aligned with human values, maintaining clear responsibility, and supporting fairness. AI is now part of many group decisions in areas such as health care, emergency response, finance, transportation, and governance. Success depends on careful design, careful testing, and good oversight. The ideas in the paper provide steps toward partnerships that perform well, adapt easily, stay open, earn trust, and stay centered on people.