Toward a science of human-AI teaming for decision making: A complementarity framework.

Area of Science:

• Cognitive Science
• Artificial Intelligence (AI)
• Human Factors
• Organizational Behavior
• Ethics

Background:

• Artificial intelligence (AI) is increasingly integral to critical decision-making processes in health, safety, finance, and governance.
• The primary challenge has shifted from human-AI collaboration to structuring this interaction for optimal complementarity.
• Human-AI complementarity signifies a synergistic relationship where combined human-AI teams surpass the performance of either humans or AI operating independently.

Purpose of the Study:

• To advance the science of human-AI teaming for decision-making.
• To propose a framework for understanding and engineering effective human-AI teams based on collective intelligence and core cognitive processes.
• To identify sociotechnical factors and design principles crucial for achieving human-AI complementarity.

Main Methods:

• Integrated insights from cognitive science, AI, human factors, organizational behavior, and ethics.
• Proposed a framework grounded in collective intelligence, focusing on reasoning, memory, and attention.
• Examined sociotechnical factors (team composition, trust, mental models, training, task structure) and outlined design principles for complementarity.

Main Results:

• Identified key sociotechnical factors influencing human-AI team effectiveness.
• Outlined actionable design principles for achieving complementarity, including role partitioning and continuous evaluation.
• Emphasized the importance of transparency, trust, and human-centered design in AI collaboration.

Conclusions:

• Human-AI complementarity is achievable through careful structuring of teams and tasks.
• Effective human-AI teams require attention to cognitive processes, sociotechnical factors, and ethical considerations.
• The proposed framework and principles offer a roadmap for developing high-performing, adaptive, transparent, and trustworthy human-AI systems aligned with human values.