Explore the crucial need for interpretable AI metrics in this 41-minute talk by Jordan Boyd-Graber, PhD. Delve into two innovative metrics for unsupervised and supervised AI methods, including the "intruder" interpretability metric for topic models and a multi-armed bandit approach for optimizing explanations in question-answering systems. Gain insights into the broader applications of these methods in fact-checking, translation, and web search. Learn how to transform AI from a mysterious black box into a transparent tool, and understand the importance of measuring interpretability in artificial intelligence systems.
Measuring Interpretability in AI - Transforming Black Box Systems into Transparent Tools
Overview
Syllabus
- Intro
- AI Should be Interpretable
- We Should Measure Interpretability
- Proposal for Unsupervised Methods Topic Models
- Proposal for Supervised Methods Question Answering/Translation
Taught by
Open Data Science
