AGI Can Be Safe
Data Skeptic
00:00
The Unpopularity of Obedience in Machine Learning
There are certain forms of obedience which are safer than others. Mathematically define what an AI should do if it wants to be obedient in that way. Many machine learning researchers feel that this obedience problem is completely intractable and not solved yet. But I solve it by not using any machine learning. This is entirely unpopular and unfashionable, but it does work to give you a kind of obedience - the obedience you like.
Transcript
Play full episode
