I'd really like to have an actual transcript with distinctions between pre-programmed responses and the "latent semantic analysis". Obviously the zoo one (and every other question in the article) was pre-programmed, and from what I can tell the way it 'learns' from the internet is basically just looking up synonyms to words or phrases, inserting them in place of the pre-programmed responses. This is another attempt at making a facsimile of AI: it doesn't really learn, it just puts on a good show of pretending to do so.