Re[8]: The dumb SAI and the Semiautomatic Singularity

From: Cliff Stabbert (
Date: Tue Jul 09 2002 - 01:04:15 MDT

Tuesday, July 9, 2002, 12:27:01 AM, Tomaz Kristan wrote:

TK> On Mon, 08 July 2002, Cliff Stabbert wrote:

TK>>> The list of all medical situationa is huge. But finite.

CS>> So you claim. I beg to differ. If your simulation, or whatever, is
CS>> based on physics, and if even a single atom can be in an infinity of
CS>> states (positions, electron spin, whatever degrees of freedom quarks

TK> For the first aid - it is definitly not so. And for the medicine as
TK> well, also not. The latest should be bigger, yes. Still very finite. you say. Do you have any argument for this? Isn't there
an infinity of states even one atom or molecule can be in, let alone
the number that make up a human body?

CS>> could still be NP-hard.

TK> It's not.

How do you know?

CS>> I think you're overconfident of the ability of standard
CS>> computational techniques to solve problems.

TK> That is my point. You can do GA with them - can't you? Just see,
TK> what the evolution is capable of. Needs a lot of computation -
TK> but so what, if you have it?

OK. But how do you get massive amounts of computation in our lifetime?
By building a seed AI, which can learn, and grow, until it can design
new versions of itself, and new hardware to run them on. At that
point, building a non-conscious AI (which I don't think is possible,
for reasons I've outlined in other posts) becomes irrelevant.

In any case: talk is cheap, code is hard. You say x, y and z is
possible? Fine, I don't think it is. Now write the software and
prove me wrong.

If we were sitting around wondering how to get rich, and I said
"I know! Let's become famous rockstars!" -- that wouldn't be very
helpful, you know?

...or wait, I've got it! You're somebody's early prototype, aren't
you? You can tell me, it's OK...Cyc, maybe?

I'm definitely getting that expert system, "symbolic manipulation
without meaningful underlying structures" vibe...


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:40 MDT