Aligning Judgment Using Task Context and Explanations to Improve Human-Recommender System Performance

Shared context may be as effective as explanations for human-AI collaboration: The study found that providing humans with task context and environmental information produced equivalent agreement on final decisions compared to showing algorithmic explanations, suggesting that helping humans understand the situation itself—not just the AI's reasoning—is a viable approach to improving human-recommender system performance.

Context helps prevent over-reliance on AI recommendations: Participants who received task context information showed less tendency to blindly follow the recommender system's suggestions, indicating they maintained more appropriate skepticism and independent judgment compared to those who only saw algorithmic explanations.

Understanding when AI fails matters: Humans given task context were better able to identify the specific conditions under which the recommender system made errors, which is crucial for responsible AI deployment since operators need to recognize failure modes to prevent poor decisions.

Explainability alone has limitations: While explainable AI has been promoted as a solution to black-box algorithms, this research reveals gaps remain—simply explaining how an algorithm works may be less effective than ensuring humans understand what the algorithm is deciding about, highlighting the need for complementary transparency approaches.

Task environment design is underutilized in AI transparency: The findings suggest that responsible AI implementation should prioritize designing systems that provide relevant contextual information alongside or instead of post-hoc explanations, representing a shift in how organizations should approach human-AI collaboration.