Alignment problems
The alignment problem
In this World Economic Forum interview, AI expert Stuart Russell explains why asking an AI to do something for you is so much more complex than asking a human to do the same thing.
Ensuring that the AI models we create capture our norms and values, understand what we really want from them and, above all, do what we want them to do, is one of the most urgent questions in computer science.
As AI systems grow increasingly pervasive and powerful, we find ourselves increasingly in the position of the sorcerer’s apprentice.
“We conjure a force, autonomous but completely compliant, give it a set of instructions, then scramble like mad to stop it once we realise our instructions are imprecise or incomplete – lest we get, in some clever, horrible way, precisely what we asked for.”