The alignment problem is one of the classic problems of artificial intelligence, and it makes for a great many science fiction plot lines. The problem is how we can get an AI to follow the unspoken instructions or the intentions of a user. How can we stop an AI from being too narrowly focused and literal? Here are two examples: Let's imagine we give some powerful AI of the future the prompt ‘Reduce cancer cases in humans.’ Well, ideally, we'd want it to develop a wonder drug or to identify thing