From: Philip Sutton (Philip.Sutton@green-innovations.asn.au)
Date: Tue Aug 19 2003 - 09:30:34 MDT
James/Ben,
> I'll explicitly state the caveat that my hypothesis is strictly for
> AGI, not for specialized sub-fields of AI -- it seems obvious to me
> that it wouldn't apply in the case of specialized AI. It is less that
> I have proof of my hypothesis than I've never seen evidence to the
> contrary. Most of the "contrary" examples I can think of were either
> shallow analysis or do not really address the underlying issue.
Why does your one-best-ideal-system model only apply to AGI but not
to specific applications of AI?
It seems to me that the more complex the system under consideration
the more likely the fitness landscape for optimisation solutions will be
highly uneven (as Ben has described it) leading to more than one
sensible approach. I would have thought that as a consequence AGI
(as opposed to situation-specific AI) would therefore have been a
stronger candidate for multiple solutions - given that the GI task is more
complex than any/most specific applied intelligence task(s).
One major experiment in multiple optimisation solutions in complex
general intelligence systems was played out at the level of the *use* of
an GI system rather than at the level of the creation of the GI
architecture over the last 50,000 years in New Guinea. The geography
of New Guinea is highly broken up due to recently formed mountain
ranges and the food naturally available from the jungle is remarkably
limited - so in large parts of New Guinea villages are very isolated.
Over the last 50,000 years something like (if my memory serves me
correctly) about a 1/3 of the world languages evolved in this one large
island - and the languages are not just variation on each other - they
are as varied within the island in their fundamental linguistic
characterisitics as human languages are varied across the entire globe.
Also the cultural characteristics of the different communities vary
massively - despite thefact that until a few hundred years ago (or less
in many places) the technologies used by the people were all variations
on stone age agricuture augmented by hunter-gatherer strategies.
One interpreation of this data is that, faced with what looks like much
the same challenge of human survival, vast numbers of solution have
evolved - because the villages were isolated enough to allow many
opportunities for experimentation.
The explanation of the much greater linguisitc and cultural
homogeneity of the rest of the world is most likely the invasion of other
cultures by those with current military (and commercial) power. Jared
Diamond's thesis in his book Guns, Germs and Steel suggests that the
global pattern of military (and commercial) dominance that we have
seen in the last 13,000 years, most particularly since the emergence of
agriculture, has owned a great deal more to geograpgical and
ecological differences between the different parts of the world than it
has done to the specific cultural traits of the inhabitants of specific
areas. So the homegenisation found in most parts of the world is not a
counter argument to the multimodal optimisation found in New Guinea -
ie. the dominance of a few cultures is not proof that there really is only
one ideal - it is more an example of first mover adavantage - those
cultures that dominated first got the chance to stamp their culture and
language on other by force of power rather via some kind of objective
test of long term optimality.
Anyway, this story alone give me reason to believe that Ben's basic
hypothesis is right. Also the history of technology is replete with other
examples.
Cheers, Philip
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT