A new study published on arXiv investigates the relationship between the functional correctness of explainable AI (XAI) methods and human understanding. Researchers conducted a user study with 200 participants, manipulating explanation correctness at four levels. The findings indicate that while explanation correctness does impact human understanding, not all differences in correctness translate to discernible changes in comprehension. Specifically, understanding dropped significantly only when correctness fell below 70%, with no further decrease at 55%. Furthermore, even fully correct explanations did not guarantee understanding, as some participants still performed poorly, suggesting a need to validate computational XAI metrics against actual human outcomes. AI
IMPACT Highlights the need for XAI evaluation metrics to align with actual human comprehension, potentially guiding future development of more effective AI explanation tools.
RANK_REASON Research paper published on arXiv detailing a user study on XAI. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
