but we need to characterize e.g. the Metallode F by a relationship such as "involves grabbing pairs from the AtomTable." Until we can characterize !Actallodes using abstract descriptors like this, one might argue we're just doing "statistical learning" rather than "general intelligence style" procedure lea