Check this one out. A computer which shows awareness of 1. its environment 2. its history 3. ambiguity in human speech.
http://hci.stanford.edu/winograd/shrdlu/ Granted, it only does something very, very specific, but pretty much any language AI which does a competent job at replicating human interaction is limited to a small field of discourse, just think about ELIZA.
Actually, no. Eliza was a terrible AI therapist.
Anyway, I'm thinking about an approach to building a convincing artificial intelligence: taking something very simple as a base, we could build a computer program out of various specific subroutines. I started to think of human cognition as something almost modular. I think the self-consciousness kind of arises from this. Still trying to formulate a theory, but not quite there yet.