Putting intelligence on silicon chips instead of a network of neurons doesn’t magically make it faster. Computers seem faster because we simplify the problems we give them: cleaning up the input data, formalizing the problem, constraining the solution space. This makes sense: why would we bother giving a computer a problem to solve unless it can solve it faster than a human? But if you take away all those training wheels and require a computer to deal with the messy complexity of the real world in the same way that we do, the computer is going to seem a whole lot slower.

As an example, witness IBM’s Watson, which was at best on par with the best human Jeopardy players. In particular, note that even though Watson was given the full text of the question instantaneously at the same time as it was revealed to the human contestants (who then required time to read the question), it still needed time to “think”, and that time was roughly equal to the time that the human contestants took. And it’s not necessarily a question of throwing more processing power at the problem to make it faster. You run up against limits like communication latency between memory and CPU. Especially when you consider that even Watson is nowhere near the intelligence of a human, and the problem it solves is still very simplified relative to the real world, and is even simplfied relative to a real game of Jeopardy. (For example, they could have required it to use a camera to “read” the question from the screen just like the human players.)

More generally, any intelligence playing Xanatos Speed Chess in the real world is going to be primarily limited by the speed at which information comes in, and a human with a better information network will easily beat out a computer.

Bottom line, it’s likely that any computer intelligence that is as smart as a human would also be about as fast (or slow) as a human in practice.

And don’t underestimate what can be accomplished by a non-sentient computer program. Consider the programming behind NPCs and computer opponents in video games. If a sentient computer did make make a bid for world domination, most of the work would probably be done by non-sentient subroutines reminiscent of video game AIs, which the sentient intelligence would periodically check in on, much like a computer-assisted human would.

Indeed, human consciousness is, as you say, a thin veneer over a mess of autonomous processes. And of course, the only reason organic computers are currently superior (for non-simplified real-world tasks) is the 4 billion years’ head start that they have over silicon chips. (Neurons haven’t existed for all 4 billion of those years, but they rely on biochemical reactions that are that old.) So I guess it comes down to a question of how close our brains have come to the theoretical limits of computational power (per watt, per unit volume, per unit mass, etc.). The more of a gap there is between our brains and theoretical max efficiency, the more plausible it is that computerized intelligences could eventually out-compete us. As to that, I don’t really have any idea, though I’m sure people have studied it. But my gut feeling, based mostly on speculation and not actually substantiated by anything, is what I already said: we likely won’t see a computer that can match wits with a human in real time, at least not for a long, long time (centuries).

(Also, by the time computer AIs overtake us, our brains will probably also be computer-augmented, so that should even the odds.)

Yes, that’s partly what I meant by “there are currently a lot of things making it equally impractical for both [humans and computers].” But my point was that Daemon is equally plausible, if not more so, compared to most fiction about sentient AI takeovers.