A lot of “crystallized intelligence” is really just having a large corpus of related problems in your history, and being good at relating or transforming new problems into old ones that you’ve already conquered. This is one reason I think some human triumphalism over AIs memorizing their training distribution is misplaced — if that distribution is sufficiently large and general, and if you’re sufficiently good at pattern matching against it, then this is a wickedly powerful tool. One way to view it is that this is the meta-technique that lets AIs be very effective even without much “intelligence,” but another way is that more of human intelligence than we want to admit is actually just this.
While crystal intelligence is a huge part of expert skill, I would also argue that vaguely mylenation-like brain circuits (i.e turning routine reasoning into hardware-accelerated brain circuits) is a huge part of the speedup. Highly skilled engineers will have enormously fast mental math, analysis, and spatial reasoning skills. These are built like any other skill (playing an instrument, football, Starcraft) with practice.
-- Review, How To Solve It, John Psmith
While crystal intelligence is a huge part of expert skill, I would also argue that vaguely mylenation-like brain circuits (i.e turning routine reasoning into hardware-accelerated brain circuits) is a huge part of the speedup. Highly skilled engineers will have enormously fast mental math, analysis, and spatial reasoning skills. These are built like any other skill (playing an instrument, football, Starcraft) with practice.