Alt-BEAM Archive
Message #02313
To: beam@corp.sgi.com
From: Steven Bolt sbolt@xs4all.nl
Date: Wed, 14 Apr 1999 15:17:25 +0200 (CEST)
Subject: [alt-beam] Re: CPU again? (was Beam genome)
On Tue, 13 Apr 1999, Terry Newton wrote:
> >*De*coupling from reality? It's coupling you want.
>
> Everything you say is true, we do need *much* better sense of and
> influence over the environment. I might have mis-spoke when I said
> "alive", I was thinking "conscious life".
Ah. Consciousness is many steps ahead of me. Robotically speaking,
it's not on my agenda at all. Lack of ambition, perhaps...
> In order to be conscious, something must be able to "think" and
> be able to "decide" on its own, not as a result of networks code
> or anything so closely coupled to reality that no other option
> is possible.
Introducing random factors is easy enough - far easier than
providing a bot with sufficient input from its environment.
Sufficient to make relevant decisions, I mean.
> In other words, it's got to have soul to think,
That's not something we can design, even in principle. Like we
can't design the seed for a tree starting with a clean sheet of
paper. Any little seed will always be immensely more powerful than
any designable machine. It's a combinatorial explosion - which
grows into not just a tree, but a forest, and eventually develops
other species.
> and I'm afraid that won't happen using existing sequential
> computers. Best one can do is simulate it well enough to fake it.
> But it isn't *really* thinking. Chaotic networks of analog
> neurons influencing each other probably has more potential to
> become "conscious" than cold rigid code. Although I'm working on
> a solution that will result in more potential for code life...
> the idea of unknown value registers. Don't look at them directly
> or all the magic goes away.
Lost me there :)
> Still looking for a good low power noise source, I'll never get
> there trying to use a counter for "random" numbers, no matter how
> much shifting I do there is still only one possible answer at any
> given moment.
Diode noise is used. Reverse bias a base-emitter with a large
resistor, amplify, and you get a random signal. Something like
that?
> It's something besides whether it's cpu or analog that will determine
> if something electronic is "alive" (there's that word again!), who's
> to say our little four neuron walkers are not alive? Can we tell?
Take it from me, they are dead as doornails.
> Would we recognise that kind of life?
A semantic debate is always possible. But not productive.
> I'm working on something I can't even verify. I can only go by
> intuition which says chaos needs to be involved if true
> multiple-choice responses are to be achieved. A phase space
> where many things can happen, so sometimes they do.
This again touches on what can and cannot be designed.
> With a cpu, generally it works as well as programmed, current
> programming methods prevent just anything from happening for
> with any set of inputs and starting conditions only one thing
> can happen. Unless I find that noise source... really it comes
> down to what is used for the random number generator. Is it really?
In general terms, adding a random factor merely lifts the
predictability of machines to the Brownian motion level.
Life and consciousness won't rise Phoenix-like from the ashes of
a little chaos. Remember the monkeys, their typewriters and
Shakespeare.
> A cpu is still just a part with advantages and drawbacks.
> One can explore deep intellectual challenges like thought with
> analog or cpu with a roughly equal chance of success - true
> because there is no definition of what thought even is. Whatever
> level it has, if it supports the robot, then it is a success.
> A cpu-controlled robot would probably have more recognisable
> thoughts, certainly capable of more advanced processing, and
> a network-based solution might have a more advanced thought
> process but it occurs in a world we do not easily perceive
> or understand. The "Living Machines" paper helps. De-coupling
> can be seen here... even though the environment is coupled
> to the neurons they do not force them. Rather it is a rhythmic
> heartbeat that mostly determines behavior, the sensors only
> influence it. The environment needs to be tightly coupled to the
> robot, it is the thought process that should be one step removed
> to open up the possibility of conscious decision. Maybe?
You've put your dream into words very well :)
Best,
Steve
----------------------------------------------------------------------
# sbolt@xs4all.nl # Steven Bolt # popular science monthly KIJK #
----------------------------------------------------------------------
------------------------------------------------------------------------
eGroup home: http://www.eGroups.com/list/alt-beam
Free Web-based e-mail groups by eGroups.com
Home