>EDIT: If/when we build a human-level AI, perhaps we could use the number of transistors / artificial neurons involved as a proxy for an absolute measure of how difficult it is to answer some problems. This would be imperfect but better than nothing.

Wouldn't that be roughly equivalent to equating brain size/volume to intelligence? I know there's a decent correlation between intelligence and head-size, but it's not that consistent. Some brains just work better for their size.