Four: Brian Christian on artificial intelligence
Effective Altruism: Ten Global Problems – 80000 Hours
00:00
I Don't Want to Drink Alcohol
An inverse reinforcement learning system would kind of learn, potentially, that an alcoholic that there one of their goals is to drink of alcohol. And i guess you suggest that we might want to have rules that allow people to inspect the imputed values that some system has figured out that we hold and then say, no, actually, this is a misunderstanding. We are not in fact, able to always operationalize our values to the maximum degree in real life, until you cold learn the wrong lesson.
Transcript
Play full episode
