OpenHEXAI: An Open-Source Framework for Human-Centered Evaluation of Explainable Machine Learning

Human evaluation of AI explanations is critical but underutilized: The paper highlights that while explainable AI (XAI) methods have proliferated for high-stakes scenarios, properly testing whether these explanations actually help people understand AI decisions requires human studies—a step many researchers skip due to complexity and cost.

Current XAI evaluation lacks standardization: The framework addresses a reproducibility crisis in XAI research, where "numerous design choices in the design space of user study lead to problems of reproducibility," meaning different studies testing similar explanations often reach different conclusions due to methodological differences.

OpenHEXAI removes barriers to responsible AI testing: By providing pre-built datasets, models, explanation methods, and a ready-made web application for user studies, the framework makes it feasible for ML researchers and practitioners to conduct human-centered evaluations without specialized expertise in study design.

Practical tools support better research planning: The framework includes "convenient tools for power analysis and cost estimation," helping researchers design appropriately-sized studies and understand resource requirements before committing to expensive human subject research.

Standardized metrics enable meaningful comparison: OpenHEXAI provides "comprehensive evaluation metrics for the effectiveness of post hoc explanation methods in the context of human-AI decision making tasks," allowing researchers to compare results across studies and build cumulative knowledge about which explanations actually work.