I think there is a huge gap between AI being better than humans at theorem proving and AI being able to “do all the things that humans can reliably and measurably do.” Theorem proving is a lot like go or chess; it’s intuition-guided search over a fairly simple search space. It’s the kind of thing that we would expect computers to surpass humans at soon, even in a world where human-level AI is a long ways away.
I do agree with that. It just seems to me like the solution at some point becomes “Run a pilot study on a comparable real-world task that needs doing anyway” rather than “Develop a standard benchmark.”
I think there is a huge gap between AI being better than humans at theorem proving and AI being able to “do all the things that humans can reliably and measurably do.” Theorem proving is a lot like go or chess; it’s intuition-guided search over a fairly simple search space. It’s the kind of thing that we would expect computers to surpass humans at soon, even in a world where human-level AI is a long ways away.
I do agree with that. It just seems to me like the solution at some point becomes “Run a pilot study on a comparable real-world task that needs doing anyway” rather than “Develop a standard benchmark.”