Of curse that the (any) code is dependent of the coder, and AI is probably the hardest thing to do well, yet picking in the teoric scenario given as example in the KZ ppt, if in every dot there is a (even if very simple) 2+ levels structure, soldiers could or not be in there, then the performance needed to run the same AI would be much bigger althought it shouldnt stress that much the rest of the code/system.
What I mean is that AI needs to get "information" about the world (some of the articles I saw about KZ, H and a few more (althought in that site with the KZ AI ppt there is a lot of things that I want to see), they always put info as a primary need, also Napoleon (is this well writen?) used to say something like"it is better a spy in a good pocision than 100000 men on the field") and need to translate their decisions to the world (eg if they know and decide to roll they will need animations to roll), but it is needed more power to get that to work, just like my first eg in the thread the AI will need to get much more info and will need to get much more "replys" if we can indeed control the ball with physical precision because if it can only do the same than eg Top Spin can do it will very bad, so even if it will not get "smarter" it will need to at least more power to scale with the rest.
And other things like the learning AI from Forza that still is being considered "computationlly expensive" (from the Bishop-Lovelace pdf), meybe is XB that cant do it (it does not reach the 1Ghz that you mention).
BTW I am using this
thread as referece too and I think it back me up.
Because I understand that in "pure AI" (research) the problems are others (they are mentioned in the thread) but for a game more things are needed (monkey with all the the info on the net is not smart but a genius in a vegetative state will not outsmart any monkey too).
Soo if we can do much more complex things the AI just to keep as responsible (or smart) in complex worlds should need much more power to run scaling with the complexity, and for me a (at very least in gaming terms) AI that runs in a complex world "as good" as one that runs in a simple one the first one is better than the second, yet it seems the case that (if you already do have all the info/actions/pathfinding that is needed) the extra CPU power may not improve the AI in the simple world game.
Edit: rereading
this
The AIS-1 chip will focus on four key areas: path finding, or computing the most tactically sound way to move from point to point; terrain analysis; moving AI bots in formation and squads; and sensory simulation, which may mean the ability to react to stimuli in the artificial environment.
this also mention things that scale with the complexity of the world, this while it should not give a better way of deciding in RPGs if the girl marry with you or not, in anything more tatical it should provide a good deal of better AI for RTS, FPS just like Cell (or at least math strong CPUs) should give in Chess.
.
