#66 – Michael Cohen on Input Tampering in Advanced RL Agents
Hear This Idea
00:00
The Importance of Imitative Learning in AI
An imitation learner is just trying to predict what a human would do in that context. So I guess I would likely be doing things. And yeah, there have been some arguments that that imitation learners would in fact not be safe. We could get into them if you like. But the simple argument that an advanced imitation learning is likely to imitate well, likely to be pretty indistinguishable from a human.
Play episode from 02:07:53
Transcript
