A Cognitive Framework for Delegation Between Error-Prone AI and Human Agents

Dynamic delegation improves human-AI teamwork: The paper demonstrates that systems can improve overall performance by strategically shifting control between humans and AI based on who is better suited for each specific task, rather than having one party handle all decisions.

AI systems need to understand human capabilities and limitations: The research emphasizes that for effective collaboration, AI agents must be able to predict and respond to human behavior and performance, showing that "comprehension and response to the actions or capabilities of a human from an agent's perspective" is essential for improved system utility.

Cognitive models help predict both human and AI performance: The framework uses cognitively inspired models to forecast how both humans and AI agents will behave and perform on tasks, enabling smarter decisions about who should handle each decision point.

An intermediary entity can manage the handoff: The paper proposes using an intermediary system to coordinate delegation between humans and AI agents, suggesting that effective human-AI collaboration may require a dedicated mechanism to manage the transition of control rather than leaving it to chance.

Complementary strengths overcome individual shortcomings: By recognizing that both humans and AI have different error patterns and capabilities, the delegation framework allows teams to "overcome potential shortcomings of either humans or agents," creating more robust decision-making systems than either could achieve alone.