Interview
summary
Argues that useful AI is an instrument that helps people do science better, not a closed oracle, and shows concrete concerns: companies hoard training sets so data as moat concentrates advantage in a few clouds, many current models optimize small business problems but rarely reveal new phenomena, and progress depends on tools that expose internals so humans can probe models — interpretability over black boxes is necessary if AI is to be a partner in discovery.
AI-written summary · about
chapters (5)
related talks
46mAI and ML, separating hype from what actually matters
19mAI in the Field: Humans at the Heart of Value
39mThe Search for Transparency and Accountability in the Age of...
11mSuccessful AI adoption- 39mSponsored Session: The Search for Transparency and Accountability in the Age of AI: RAI & XAI as Essential Tools
24mAI for the Common Good: Acting Wisely in Uncertain Times
41mIf We Want AI to be Interpretable, We Need to Measure Interpretability
46mRub a Little AI on It