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

Shane T. Mueller, Robert R. Hoffman+3 more|arXiv (Cornell University)|Feb 5, 2019

Explainable Artificial Intelligence (XAI)Computer Science44 references218 citations

TL;DR

An integrative literature review that defines what constitutes a good explanation in Human-AI systems, surveys historical and theoretical foundations, and highlights exemplary works while calling for fuller empirical reporting.

ABSTRACT

This is an integrative review that address the question, "What makes for a good explanation?" with reference to AI systems. Pertinent literatures are vast. Thus, this review is necessarily selective. That said, most of the key concepts and issues are expressed in this Report. The Report encapsulates the history of computer science efforts to create systems that explain and instruct (intelligent tutoring systems and expert systems). The Report expresses the explainability issues and challenges in modern AI, and presents capsule views of the leading psychological theories of explanation. Certain articles stand out by virtue of their particular relevance to XAI, and their methods, results, and key points are highlighted. It is recommended that AI/XAI researchers be encouraged to include in their research reports fuller details on their empirical or experimental methods, in the fashion of experimental psychology research reports: details on Participants, Instructions, Procedures, Tasks, Dependent Variables (operational definitions of the measures and metrics), Independent Variables (conditions), and Control Conditions.

Motivation & Objective

• Address the question: What makes for a good explanation in Human-AI systems?
• Summarize the history of computer science efforts to create explainable and instructive AI (e.g., tutoring and expert systems).
• Present capsule views of leading psychological theories of explanation relevant to XAI.
• Highlight articles notable for their relevance, methods, results, and key points.
• Recommend reporting practices for empirical studies in AI explainability, analogous to experimental psychology.

Proposed method

• Conduct an integrative literature meta-review.
• Synthesize key ideas and publications into capsule views of explanations.
• Highlight select articles with strong relevance to XAI and summarize methods and findings.
• Provide bibliographic focus and a bibliography for Explainable AI.
• Advocate for thorough reporting of empirical/experimental methods (Participants, Procedures, Variables, Controls).

Experimental results

Research questions

• RQ1 What constitutes a good explanation in human-AI systems?
• RQ2 What historical and theoretical foundations (e.g., tutoring systems, expert systems, psychological theories) inform XAI?
• RQ3 Which articles provide the most relevant methods and results for explainability in AI?
• RQ4 What are the recommended reporting standards for XAI empirical studies?

Key findings

• The report integrates the history of CS efforts to create explain and instruct systems, including intelligent tutoring and expert systems.
• It articulates explainability issues and challenges in modern AI.
• It presents capsule views of leading psychological theories of explanation relevant to XAI.
• It highlights certain articles for their methodological relevance and key results.
• It recommends researchers include detailed empirical methods and operational definitions in AI explainability studies.