What I mean is that the IBM Watson deals effectively deals with semantic information.

You can make some analogies between sparse matrix data representations and what happens in the mammalian neocortex. It is of course not self conscious, but it effectively deals with semantic information and relationships among it.