Index Errans

person in Black box

Tim Miller

Explainability

in Black box

Tim Miller, professor of computer science, in the School of Computing and Information Systems at the University of Melbourne when Explanation in Artificial Intelligence: Insights from the Social Sciences (2019) appeared. His background spans agent-based AI and human–computer interaction, and the paper is his most-cited work.

Stake§

Academic and agenda-setting — the paper effectively defined a research programme, and the large citation count that followed is itself the reputational return. No commercial dimension.

Miller's contribution to this topic is to import the cognitive science of human explanation into explainable AI. He assembles findings that explanations between people are contrastive (why this outcome rather than that), selected from many causes, and social — a transfer of understanding between explainer and audience — and argues that machine explanation should be built to those findings rather than to engineers' intuitions. The move connects the machine debate to the human literature that Nisbett and Wilson belong to, and bears on what it is to explain an agent at all, which is Dennett's question.

Works in this corpus§

their concepts on the territory
ExplainabilityExplainability

1 concept in this scholar's webopen the full territory →

excerpts

However, it is fair to say that most work in explainable artificial intelligence uses only the researchers' intuition of what constitutes a 'good' explanation.
— Explanation in Artificial Intelligence: Insights from the Social Sciences (2019)

Miller's charge against the [[concept:explainability|XAI]] field: it builds explanation interfaces from engineers' guesses about what a good explanation is, ignoring the empirical study of how people actually explain. The corrective he proposes is to read the existing literature on human explanation.

on Explainability

There exists vast and valuable bodies of research in philosophy, psychology, and cognitive science of how people define, generate, select, evaluate, and present explanations…
— Explanation in Artificial Intelligence: Insights from the Social Sciences (2019)

The thesis: explanation between humans is contrastive (why this and not that), selected from many causes, and social — a transfer of understanding in conversation. Miller argues machine explanation should be built to those findings rather than to intuition.

on Explainability