creators

Been Kim - Interpretability at Google DeepMind

Been Kim is a director and principal scientist at Google DeepMind. She created TCAV and now works on agentic interpretability and concept transfer.

Been Kim is a research scientist known for her work on machine learning interpretability. She holds MS and PhD degrees in computer science from MIT, where her 2015 thesis was on interactive and interpretable machine learning models for human-machine collaboration, and she worked at Google Brain before moving to Google DeepMind, where her own page now gives her title as director and principal scientist. Her best-known contribution is TCAV (Testing with Concept Activation Vectors), a 2018 ICML paper with Martin Wattenberg, Fernanda Viegas and others, which tests whether a human-understandable concept, such as "stripes" for a zebra classifier, influenced a network's prediction. TCAV won one of the ten Netexplo Awards, chosen from more than 2,000 innovations, at the UNESCO Netexplo Forum in Paris in April 2019. Kim now works on what she calls agentic interpretability: getting knowledge out of models and back into human hands. She was general chair of ICLR 2024 and senior program chair of ICLR 2023, and sits on the ICLR board and on the FAccT and SaTML steering committees.

Specialization

interpretability, concept-based explanations, human-AI interaction, AI safety

Country

United States