From: Eugen Leitl (eugen@leitl.org)
Date: Thu Jun 13 2002 - 16:33:17 MDT
On Thu, 13 Jun 2002, Smigrodzki, Rafal wrote:
> ### Well, in that case indeed there is nothing to fear in the foreseeable
> future, and you can safely dismiss Eli, instead of expressing outrage at his
> efforts.
If I thought such efforts had a chance I'd use plastique, not words. But
the memes are dangerous, since possibly motivating people who'd have a
a higher chance of succeeding.
> I don't share this attitude. AI research was not what it was cracked up to
> be, but it will.
I hope you like your popcorn. The movie is going to be *so great*.
> ### I'm not so sure. There are places out there in autocratic states where
> even nuclear weapon projects could go unnoticed, a bunch of geeks with
> workstations won't be even a blip on your radar.
I'm not interested in a bunch of geeks with workstations. This goes
nowhere.
> ### Do you mean the estimate for the computational power of the human brain
> is incorrect or the Moore's Law will not hold long enough? Integrated 3D
There are three things wrong with this sentence. The estimates are
off-cuff by orders of magnitude, Moore is about integration density, not
performance, and the assumption that raw performance immediately
translates in AI kinda tenuous.
> molecular circiutry doesn't seem to be necessary to achieve this kind of
> power.
Let's wait and see. 300 mm Si is about the biggest wafer size we've seen,
and even with WSI and smallest structure size this is still flatland. How
much crunch can you get from a 300 mm diameter surface?
> ### See this quote from Ben:
>
> "I think we're talking about a few hundred thousand lines of C++ code, maybe
>
> 5-10 different specialized data structures, 10-20 different specialized
> learning algorithms, and 100 configurable parameters altogether. Smaller by
>
> far than the Windows OS, but vastly harder to debug and tune... "
>
> But then, since he is an AI researcher, we shouldn't pay attention to him,
> right?
I don't actually disagree with Ben. Biology is subject to physics, and
physics can be pretty stark at the TOE level: just a small vector of
integers. Human genome is a rather small vector of integers, come to think
of it.
But you'll notice that he kinda ignores the state of above algorithm
(where all the details lie), and you'll notice that he's not giving the
url to the tarball which contains the beef of above outline. Plus, what is
his track record? He writes code, but what is the impact of said code, and
the trend in it? Excuse me for being incredulous, but we've seen this so
many times before.
> ### Just wait for Windows 2015 on a multiprocessor Pentium 10 machine, with
> Cyc v7.9, and a Flare 4 AI development suite.
We always knew Armageddon is a side effect of Redmondware. If only we'd
acted while we had a chance! Corruption of precious body fluids...
> Basically, what worries me about your position is the seeming discrepancy:
> on one hand you are quite forcefully remonstrating against the current FAI
> attempts, for all the danger they are putting us in, but then on the other
> side you dismiss them as so incompetent as to be unable to even get properly
> started towards their goal. Which is it then?
You've catched it? Kinda obvious. No, I can't think of any current
attempts that are a threat. But I do think that the memes are dangerous,
and need counterpressure. Time well invested, I think.
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:14:46 MST