Look, Hofstadter is an amateur. He is obviously a music lover, but he's not a music expert by any stretch. This was plainly obvious when he came to my university and gave his little talk.

He assumed that because his talk was at the music school's concert hall, his audience must be comprised of music students and faculty, without actually ever checking if that were the case. Those of us from the music school in the audience were all very aware that the majority of the audience were students from mathematics, physics, computer science, etc. Most music students don't know who Hofstadter is, and don't care.

So when he jumps to his conclusion that he "fooled" 1/3 of the audience with his recording, it's way way off base. After the talk, we all (music faculty & grad students) went out for a drink with Hofstadter and pointed out the errors in his assumptions, but he refused to listen. His methodology is totally bent, and his conclusions are what he wants to think, not the actual data.

I --and all my fellow musicians-- were pretty much all totally nonplussed by his demonstrations, and underwhelmed indeed by the supposed "quality" of the computer compositions. Don't get me wrong, it's impressive that a computer can do what EMI can do. But it ain't Bach, and it sure ain't Chopin.

Also, do not underestimate the human intereference in the process. A human is involved in "tweaking" EMI and "filtering" its output. EMI probably produces 1000 pieces per minute, 999 of which are total crap, and one of which is vaguely interesting. Run it for a few minutes and you'll have a few vaguely interesting pieces. Pick the best one, and say "the computer composed this, isn't it great?". What about the other N-thousand pieces that were shite?