The Transition to Adversarial Learning in AI

The idea of adversarial learning is at the core of a lot of modern AI techniques. For example, if you look at deep fake photos behind the scenes, there's one AI that creates photos and the other AIs tries to tell if they're real or fake. They both don't know how to do it but after doing this for millions or billions of cycles, they get very, very good at it. The same thing we did back then with Eureka. We had models that fit small,. a small subset of the data and those subsets of the data moving around to try to defeat the model.