Re: [Swarm-Modelling] Crichton's "Prey"

[Rick Riolo]

i read this a couple of years ago.  i thought it
was a pretty poor read, frankly, even after suspending
disbelief.   it was years since i read his andromeda strain,
but at least in my memory of it, that was a much better book.
(also, i think in prey he mentioned a prof john holland
as big name in complex systems / evolutionary algorithsm research
but said he was at the university of chicago!!)

at any rate, to your point:

it doesn't seem like this is impossible, at least "in theory".
that is, what if each agent was some kind of finite state machine,
like there is in each cell of a cellular automata.
that means each will have some sort of function specifying
how its inputs and current state map to a new current state
(and perhaps to outputs, including actions).
thus the agents can "learn" by changing their functions.

now suppose the agents were actually connected like a virtual
cellular automata...i.e., despite buzzing around a bit, they
maintained a more or less fixed set of neighbors.
then, as we know from wolfram's work (not to mention all
the precursor work that wolfram doesn't mention!!), it doesn't
take much to come up with a system of such agents that
is a universal computer (in the turing sense), and so
can be as "intelligent" as any computation device
(of that type).

but maybe your main question is about *how they could learn*.
well, that is a good question.   all the above argument
establishs is that such an intelligent system could be
implemented in that set of agents, not how to
find the implementation (ie how to find the rules).

how about this view:
learning could consist of looking at neighbors, seeing how
well they are doing (at whatever...), and then copying
the behavior rules of some neighbor that is doing better than you
(with some error).   that has change dynamics that are similar
to evolutionary learning, i.e., spread of fitter, with some error.

that also pushes some/much of the problem off onto
the key issue of deciding who to copy (who is "fitter").
there is a long history of "war stories" from lots of work on evolutionary
computation, because evolutionary algorithms will take advantage of
all kinds of unforeseen consequences of the "fitness function"
we give them to work with, and end up doing things
far from what we, the designers, had hoped they would do.

maybe there are other learning algorithms that
might work, depending on what the group has to learn.
but i suspect there is a lot we have to learn about
what kinds of capabilities agents have to have, including
what kind of interaction rules they have that allow
them to structure their interaction patterns,
not to mention the learning algorithms they would need,
to support the kind of swarms he has in the book.

here is another view:
we people are one example of such swarms!
(ie a big group of simple agents that collectively does some smart
things).  so a queustion is: if there could be "systems" as smart
as us that have the kind of loose organization of his swarms,
then why haven't they evolved?
just bad luck? (ie historical accidents)
or are there some deeper reasons that such swarms
would not support such intelligent behavior?
(ie, if we ran the tape over many times, we'd never see such swarms).
perhaps there are organizational reasons such swarms
are limited in what they can do, e.g., having to do with the
competition/cooperation that has to be managed between
selective units at one level (the agents) and higher levels
(metazoans, etc), along the lines discussed by
maynard smith and szathmary, or franks, or buss, or ...

in any case, i agree that it does take a hefty amount
of belief suspension to get to the rapid learning
that is portrayed in the book.

- r

Rick Riolo                           address@hidden
Center for the Study of Complex Systems (CSCS)
4477 Randall Lab
University of Michigan         Ann Arbor MI 48109-1120
Phone: 734 763 3323                  Fax: 734 763 9267
http://cscs.umich.edu/~rlr

On Fri, 3 Jun 2005, Steve Railsback wrote:

> Date: Fri, 03 Jun 2005 08:32:27 -0700
> From: Steve Railsback <address@hidden>
> Reply-To: Swarm Modelling <address@hidden>
> To: Swarm Modelling <address@hidden>
> Subject: [Swarm-Modelling] Crichton's "Prey"
>
>
> Someone just gave me Michael Crichton's 2002 book "Prey". (I would bet a
> lot of money that he wanted to name it "Swarm" but his lawyers told him
> not to...too bad, could have been a nice fund-raiser for us.)
>
> Am I nuts, or is a central assumption of the book nonsense? He has a
> swarm of independent, persistent, artificial organisms; and the swarm
> could learn very rapidly. Our brains can learn rapidly, but the
> connections are all hardwired; a population of simple organisms can
> "learn" but only  via evolution, which requires birth and death and
> selection... Is there any way a collection of independent agents,
> interacting locally with a changing set of neighbors, can "learn"
> without evolution?
>
> Steve
>
> --
> Lang Railsback & Assoc.
> 250 California Ave.
> Arcata, California 95521
> 707 822 0453
> _______________________________________________
> Modelling mailing list
> address@hidden
> http://www.swarm.org/mailman/listinfo/modelling
>