Sometimes if I'm having trouble sleeping, I start doing an easier IMO shortlist problem to think about something while drifting off. I'd prefer to think about some nicely packaged AI Safety theoretical questions instead. Do you know of any AI safety questions that feel similar to math problems?

By "nicely packaged", I mean something that is already phrased kind of like a math problem and for which it's sort of obvious in my mind whether something is a solution or not. Something like "what is intelligence" is too philosophical for my taste. ELK is an example of the kind of thing I'm looking for.