Explanation in Human-AI Systems: A Literature Meta-Review, Synopsis of Key Ideas and Publications, and Bibliography for Explainable AI

Good explanations require psychological grounding: The paper emphasizes that AI researchers should draw from established psychological theories of explanation rather than developing explanations in isolation. Understanding how humans naturally explain and comprehend information is foundational to creating AI systems that people can actually understand.

Explainability standards need methodological rigor: The authors recommend that XAI research reports adopt experimental psychology standards, including detailed documentation of participants, procedures, tasks, dependent variables, and control conditions. This suggests current XAI research often lacks the empirical rigor needed to validate whether explanations actually work for users.

Historical lessons from expert systems and tutoring systems matter: The paper traces explainability efforts back through intelligent tutoring systems and expert systems, indicating that the field has decades of prior work on explanation that modern AI researchers should learn from rather than reinvent.

The field lacks consensus on what "good explanation" means: By framing the core question as "What makes for a good explanation?" and noting the vastness of pertinent literature, the paper suggests that explainable AI lacks a unified framework for evaluating explanations—a critical gap for responsible AI development where stakeholders need consistent standards.