What do you think of this method of creating AI?

Question

No one can really argue that recreated intellegience as it is in humans, in computers, would be extremely difficult.
I had an idea - there was this Miller-Urey experiment conducted in 1953 -
http://en.wikipedia.org/wiki/Miller-Urey_experiment
where these scientists successfully created organic molecules from inorganic molecules in a primordial-earth-like chamber.
I was thinking, what if you somehow make this supersuper computer that could would keep track of every atom / subatomic particle in a virtual miniworld. Then run the same Miller-Urey experiment, but virtually. This time however, make the computer simulation speed it up as much as possible, and have it run till (hypothetically) it would create living organisms. Eventually organisms with brains and the such would form. Once you get some form of intellegience in this virtual world, all you have to do is copy their atomic code onto another computer, and there you'll have AI. What do you think?

Answer 1

Computers are already being used to model many organic and inorganic molecules for use in medical and scientific research. The problem with your proposal is one of scale. Recently a team of researchers announced that they had successfully modeled an organism at the atomic level, with something on the order of half a million atoms. This organism was a virus, chosen because it was much smaller than most virii, and the simulation ran for a tiny fraction of a second by my recollection. In order to simulate an entire ecosystem for evolution to occur in, over a timescale of probably millions of simulated years, would require far more computing power than is currently available or will be for the foreseeable future.

Answer 2

That's got to be one of the best ideas I've heard on Yahoo! Answers. I love it. True, the technical obstacles are high right now, but that's not the point. There is no FUNDAMENTAL reason why this can't work. The idea is that if we can't figure out how to make a sentient robotic brain because we're just not smart enough to know how, why not leave it up to simulated evolution to do the job? But we don't have to do a repeat of the Miller-Urey experiment virtually, we can jump ahead and start with something similar to DNA that will construct "brains" and let virtual nature take its course. Of course, we'd have to provide an environment for those brains to interact and maybe destroy each other for dominance and some kind of virtual sex.

Answer 3

Well... it's going to take a LOT of processing power to model all those molecules. And, it's not clear how closely you'd need to track everything. (The closer you track, the more processing power.)

I'd think you might get intelligence faster using an approach like genetic programming. Let the programs evolve, but without doing so much physical modeling.

Answer 4

Go for it!