Explainable AI (XAI)

Also see index ethical AI .

The term XAI is coined by DARPA, cf. TFCA lecture 1: intro to XAI . See specifically D. Gunning, Explainable Artificial Intelligence (xAI), Technical Report, Defense Advanced Research Projects Agency (DARPA), 2017.

Definitions

  • "explainability is defined as the ability a model has to make its functioning clearer to an audience." source, p.100

General

For an overview of notes on the more broader and philosophical question what "explanation" is: What is explanation? .

That index note also contains references to Bayesianism.

Technical Methods

Feature attribution

Example-based and counterfactual

Case-based reasoning

Case-based reasoning is also one of the hybrid strategies to explainable AI, e.g. identified in the deep learning taxonomy of BarredoArrieta. I'm collecting papers on that topic in a note on case-based reasoning.

Evaluation

XAI Taxonomies and ontologies