Does the Whole Exceed its Parts? The Effect of AI Explanations on Complementary Team Performance

Explanations alone don't guarantee better teamwork: The study found that while AI explanations increased human acceptance of AI recommendations, they did not actually improve team performance compared to teams without explanations—challenging a common assumption in explainable AI research.

Explanations can increase inappropriate reliance on AI: Rather than helping humans make better decisions, explanations increased the likelihood that people would accept AI recommendations regardless of whether those recommendations were correct, suggesting explanations may be creating false confidence rather than genuine understanding.

The "complementary team" problem remains unsolved: Even with explanations, researchers struggled to achieve "complementary performance" (where the team outperforms both the human and AI working alone), indicating that current explanation approaches may not be addressing the core challenge of effective human-AI collaboration.

Prior positive results had hidden limitations: Many earlier studies showing that explanations improve team performance only observed benefits when the AI was already significantly more accurate than humans—a condition that doesn't reflect many real-world scenarios where human and AI capabilities are more evenly matched.

We need new approaches to build appropriate trust: The paper highlights a critical gap in responsible AI: developing explanations that help humans understand when to trust AI recommendations and when to rely on their own judgment, rather than simply increasing overall acceptance of AI suggestions.