There is a myth perpetrated by the non-discipline of cognitive psychology, that knowledge divides into (at least) two kinds of things: procedural knowledge and declarative knowledge.  The former, it is grudgingly acknowledged, is poorly handled by algorithms, and is evidenced only in the doing, as when I tie my bootlaces.  The latter, illustrated by the sentence “Paris is the capital of France”, is assumed to be free of any particular situation, and to be encodable.  It is the kind of knowledge folks in AI believe they are equipping their programs with. But this is nonsense.  There is no sentence in the computer, and the sentence “Paris is the capital of France” could index knowledge only in its use by a linguistic being who can skillfully cope with language.  Words are not fundamentally different from bootlaces, and we can not teach machines anything, for they can not know.

Advertisements