I was following your argument happily until you used the word “deterministic”. What does determinism have to do with intelligence?
Current parallel computers are not deterministic [this is still true in 2012], and parallel programs that you run on them have to take special measures if you want them to give the same results every time. So is Photoshop intelligent if I run it on a dual-processor machine? Not half! Would it be easier to write an intelligent program on a multiprocessor? Yes, because the machine is faster, but nondeterminism will only cause debugging headaches.
Suppose I have a black box which acts exactly as if there were a human inside (maybe because there is). How do you tell if it’s deterministic? It has memory, so you’ll never be able to get it into the same state twice, so there won’t be any way to figure out whether it’s deterministic without taking it apart.
Original version, January 1998.
Updated and added here January 2012.