notes-cog-ai-combiningSkillsAndSemanticKnowledge hypothesizes that semantic knowledge derives from skill learning:

" Knowing-how and Knowing-that

	The traditional assumption has been that knowing-that is the most fundamental kind of knowing, and that knowing-how is constituted by interrelating various sentences containing knowing-that knowledge. 
     It can be argued that anything which can be properly called "knowing how to do something" presupposes a body of knowledge-that; or to put it differently knowledge of truth or facts. If this were so, then the statement that "ducks know how to swim" would be as metaphorical as the statement that they know that water supports them (Sellars 1963 p.1)

Fodor both ridicules and endorses this view of knowing-how knowledge when he says that there is a little man in our head with a set of books. When we tie our shoes, the little man gets down the book titled “how to tie your shoes”, reads each of the instructions and then follows them in sequence. Fodor claims that this, minus the anthropomorphisms, is a description of a computer with a series of branching subroutines (i.e. the books on the inner shelf.), which is basically how our minds actually work. (Fodor 1981) The pragmatists, however, considered this description to be exactly backwards. Dewey rejected the idea that “knowledge is derived from a higher source than practical activity” ( Dewey 1916 p. 262) and insisted that “There is no such thing as genuine knowledge and fruitful understanding except as the offspring of doing.” (ibid. p.275 italics in original.) In other words, abstract knowledge (knowing-that) is dependent on practical activity (knowing-how), not the other way around. This was a very strange idea in 1916, and Dewey probably didn’t know exactly what he meant by it. However, in the artificial connectionist systems that have partially simulated knowing-that knowledge, this ontological relationship is reversed in pretty much the way Dewey described.

	Instead of creating knowing-how abilities by writing knowing-that subroutines, these connectionist systems create knowing-that knowledge by a specialized truncation of knowing-how knowledge."

This suggests an interesting programme for AI: