Something new, something different
Hal Feinstein
hfeinstein at cox.net
Sun Jan 16 10:52:11 CST 2005
http://www.csl.sony.fr/Research/Topics/Language/
The Talking Heads experiment studied the evolution of a shared lexicon
in a population of embodied software agents. The agents developed their
vocabulary by observing a scene through digital cameras and
communicating about what they had seen together. To add an extra level
of complexity to their task, agents were able to move freely between
different computer installations located in different parts of the
world. Members of the public were able to influence the course of the
experiment by logging on to the Talking Heads website to create and
teach their own agents.
-----------------------
and
The Project Idea:
Assume a group of small computers each connected to a high frequency
radio. The environment each senses are characteristics of the HF
spectrum: channel occupancy, signal strength, noise and perhaps signal
type and classification. Each can change frequency, listen and transmit
some information which might start as simple
off-on keying and evolve to tones busts or a simple digital information
exchange. The goal of these simple computer/radio is to lean how to
maintain a reliable high frequency link in order to pass messages.
Using ideas of evolution and refinement the radio test strategies that
improve connection reliability by retaining refinements and discarding
lesser performing strategies. Installed in each's strategy, a "prime
directive," that at all costs the units must never become totally
disconnected, but might be allowed to break into more effective smaller
groups to promote innovation. A second "directive" is that cooperating
as a group vastly aids success. Being able to sense and influence
their environment and the intuitive knowledge that groups are
beneficial, the computers/radios would find each other in the RF
spectrum and evolve a simple language to communicate ideas that
improving link reliability and determining better ways to organize.
Devising a better strategy would be a kind of "group grope." Receiving
these ideas via HF a computer/radio would incorporate them into their
strategies while perhaps also discarding substrategies that were not as
successful. The language between them would also evolve, moving away
from easy starter notions to expressing more complex ideas.
Whats in it for us? I think it would be a blast watch these things
talk between themselves and figuring out what they are saying. If we do
the programming right we will see the computers/radio create ideas and
use logic very different from the way you and I reason about common
tasks. This has been the experience with some other machine reasoning
experiments I've read about. The key for us will be learning how to
program the computer/radios not to be good operators but to *learn* to
be progressively excellent operators. We've got to create the tools that
empower the radio/computers to continually get better at what they do
without us saying how to do it.
I know some people will say this is "way out there" or too "sci fi."
But this is fundamentally an engineering problem and requires the same
kind of organized thinking, conceptualization and building skills as
other R&D engineering projects.
Is it practical? It would not be very expensive; computer controlled
radios are pretty common these days. Programming would be done in C
and/or VB on a PC and a sound card might be used for the language
exchanges. The most challenging thing for the "creators" (us) would be
devising a framework to allow the radio/computers to build, test and
communicate their own best strategies and in giving the radio/computers
the "gift of speech" without specifying how to talk or what to say. I
don't believe anything like this has been done before, certainly not by
amateur radio operators and it would take us into a whole new technology
that I think is going to be as common place in 20 years as the Internet
is today.
--hal (wb3kdu)
PS Thanks to Bob, w8mau for the firbe demo at tacos.
More information about the Tacos
mailing list