From: Eugene Leitl (eugene.leitl@lrz.uni-muenchen.de)
Date: Mon May 15 2000 - 01:56:10 PDT
Tom Sweetnam writes:
> When does machine-made intelligence cross the line from "artificial" to
> something else entirely? I remember reading last year about a new and
Apropos, just in the unlikely case you're not aware of Hugo de Garis:
http://foobar.starlab.net/~degaris/
Despite borderline kooky presentation, the fundamental approach is imo
very sound (heck, I'd be doing it very much the same way. Now only if
he didn't try to integrate GA-bred modules by hand...)
He breeds artificial neurons as circuitry pattern for a 3d cellular
automaton via darwin in machina, currently using FPGA-accelerated
cellular automaton machine to let it execute at a less than glacial
speed. We can do much better in near future, of course, if people come
around to use embedded RAM processes to implement hardware
CAMs. Molecular circuitry can still do much much much better, but we
ain't there yet.
Here's some demo C code, illustrating the concepts:
http://foobar.starlab.net/~degaris/source_code/dynamic2Dpattern.html
As to why at all using cellular automata in hardware, have a look at
http://psoup.math.wisc.edu/papers/margolus.ps.gz
More rigorous argumentation towards the ultimate "computronium"
hardware can be found here:
http://www.ai.mit.edu/~mpf/Nano97/paper.html
To drop a few more names, Fredkin and Toffioli are good names to put
into google, and Oz author Greg Egan has been recently popularizing
the concept (Permutation City, Diaspora).
This archive was generated by hypermail 2b29 : Mon May 15 2000 - 02:01:46 PDT