Been Kim: Interpretable Machine Learning
The Gradient: Perspectives on AI
00:00
How Can the Fuzzero Surprise Us in a Way That's Productive?
The idea of expanding the human space of concepts is really exciting to me as well. I see sometimes a little bit of overlap with things like feature engineering, right? So I might tease a very simple regression model to do something more interesting,. If I hand a quadratic concept, collections of things that don't really mean anything to me. But it's like, hey, this performs better. And in the same way, a machine might hand back something to me once I inherited its space of concept. It's like, well, obviously, this produced move 37 or something of that flavor. But to me as a human, it makes no sense, at least right now. We have
Transcript
Play full episode
