From: Eugene Leitl (eugene@liposome.genebee.msu.su)
Date: Wed Jul 08 1998 - 06:52:30 MDT
Dan Clemmensen writes:
> It certainly indicates that the trends that we are extrapolating must somehow
[...]
> Yes, this is just playing with numbers. It's still interesting.
But not nearly as interesting as participating in a full-blown
Singularity ;)
> [...]
> Yes, computer hardware and software have been contributing to
> the development of the next generation of computre hardware and
> software, but this is a relatively recent trend. Software and
> hardware development productivity is horrible, and IMO we
> are still taking baby steps. A breakthrough is not unreasonable.
I am very heartened with the promise of evolvable hardware (EHW), GA
acting upon (analog) FPGA. This
- removes the human programmer bottleneck. All you have is to define
fitness (which can be as informal as saying "I want a filet mignon")
- allows very compact/efficient, albeit opaque solutions
- removes the catastrophic failure/brittleness due to unforeseen
input/hardware failure
- allows extremely large/complex, yet robust solutions
- eliminating the von Neumann bottleneck, allows optimal utilization
of the transistor resources
- a critical WSI enabler, allowing better than 0.9 utilization of
silicon real estate
- it scales very well to 3d arrays of quantum dots
> [...]
> In my personal model, the SI is not purely an emergent
> phenomenon of the net. The net serves as the raw material
> for a directed augmentation of an initial proto-SI that first
> emerges as the result of a catenation of a set of development tools
> and a human programmer.
I think that the human is just the primer. If the Internet++ is a
suitable substrate, you need just a tiny, nonsentient SI nucleus. If
network security then is as buggy as today, it will then take over the
Net in minutes, possibly sneaky, so there is no interruption of service.
The bootstrapping human is then essentially disposable.
With some effort, today a more or less clever hacker could build a Wintelian
(that will take care of 90% of everything) worm, packing constructive
buffer overruns together with an IP scanner. As IP addresses of PPP
dialup accounts are typically world-visible, you could participate in
the fun. The worm will not be able to do anything very constructive,
but only because current computer performance and networking
bandwidth is so very pathetic and the state of the art in GP is
in statu nascendi. If there is >Tbit FPGA humming in every cellar
(to better (de)compress your video, to recognize your face and
your voice, to feed your pets, and to render all the nice VR
games), interconnected with several GBit links (why, multimedia),
the situation would seem to be a bit different.
"On February 4, 1998, MSI announced a definitive agreement to merge
with Pharmacopeia, Inc., a leader in application of combinatorial
chemistry and high throughput screening to drug discovery". If you
extrapolate such strong current trends as DNA sequencers, total
organism genome databases, very large organical reaction
databases, hybrid QM/MM forcefield models, automatical scanning
of combinatorial synthetic libraries, GA in synthesis planning,
use of MEMS for combinatorial chemistry, STM/SNOM, you will see that
not many years from now a device not much larger than a desktop
will be sufficient to initiate the bootstrap of some kind of a
molecular autoreplicator very quickly, if used correctly. So
essentially this would be a problem of knowing how, a threshold
of which is probably equivalent to a team of pretty good human
experts. Even a fledgling SI could do it in a few days, probably
much less. After that it has a) basically optimal hardware
b) can grow in exponential progression, with very short
replication times. From the receiving end (you and me), even an
indifferent SI looks very much like a Blight.
> >
> [ 20 yeares more Moore ]
> Instead of asking a radical singulatarian such as myself to comment,
> please allow me to ask you to predict the effect of this level of
> computing capacity on society.
Unemployment, unemployment, unemployment.
'gene
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:49:19 MST