The Alignment Problem: Reward, Values, and Policy
The question Turing left behind In 1950, Alan Turing asked his readers to imagine a game. A judge would exchange written messages with two unseen participants, one human and one machine. If the judge could not reliably tell them apart, the machine would have succeeded at the imitation game. I like Turing’s move because it is so practical. Instead of waiting for a definition of thinking that everyone could accept, he replaced the philosophical argument with something a person could observe. Can the machine hold up its side of a conversation? ...