From: Eugene Leitl (Eugene.Leitl@lrz.uni-muenchen.de)
Date: Fri Mar 01 2002 - 04:59:01 MST
On Fri, 1 Mar 2002, Remco Gerlich wrote:
> > I mean, what would you do with a ~1Pflops computer???
>
> Well, not create an intelligence, since I haven't found out how to program
> one yet.
So don't, model a critter. Learn how to image it first. Or fiddle with
molecules, preferably ones useful to build faster computers.
> I believe the bottleneck on a Singularity is not computer performance;
> they're already somewhat faster than we need at the moment and will become
> lots and lots faster.
Um, if we're talking about human level intelligence people have been
claiming this for >20 years. If you look at the actual system performance
of current architectures, you'll see that they're highly skewed (benchmark
XY, not AI), and hence are not a good estimate of human processing
equivalent (even if we had a good idea what a human processing equivalent
is, which we don't yet -- Kurzweil has been using the glass pipe, as
usually).
> There is so much work left to do to create a general intelligence, and all
> of it hard human work that won't scale well in teams.
Yep.
> I don't believe genetic engineering is going to produce significant advances
> in intelligence in the next 50 years.
True. Given that you have generation iterations of about three decades,
plus acceptance time before it gets deployed widely, plus time before
enhanced people will start having an impact.
> Also, nanotech may happen in some very limited form but not enough to lead
> to a singularity without superhuman intelligence guiding it.
We should get molecular circuitry in 3d bulk (computronium) in less than
two decades. We should get molecular memories (2d) in less than a decade,
2d computronium soon after.
> I expect at least 20 years, probably more like 30 years for programming a
> good enough general intelligence. I hope to be able to be a part of this,
I do not see anybody who goes on two legs programming a general
intelligence at all. Creating boundary conditions for emergence of a seed,
yes. A seed without control, however.
> I'm just now finishing my CS study, I need to make some money somehow, but
> otherwise I want to work on this.
The longer you wait, the less time you will have. Of course, you will need
to get rid of the college debt first.
> But that's a very wild guess - just my intuition.
>
> Another part of me says that we will continue as before, gradual improvement
> in technology, no singularity. People will have enough of technology at some
> point. Research money will go into entertainment only.
Lots of hardware innovation comes from entertainment. Resources to run
simulated game worlds and rendering them would seem to drive towards
parallelism.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT