Re: Where is behavior AI now?

"Curt Welch" <curt@xxxxxxxx> wrote in message
news:20060827023211.671$2R@xxxxxxxxxxxxxxxxx
In my simple-bot example which has 2 binary inputs sampled 10
times per second, if you were to record only 1 second of sensory >
data, (10 x 2 bit samples) and produce reactions to that small set
of sensory data, you would already have 2^20 or 1,048,576
different sensory conditions to react to. If you recorded only 2
seconds of sensory data, only 40 bits, you would have over
a trillion input states to respond to.

You have a long and reasoned argument. I'm sorry, but I do not buy it. I
think the argument is based on falacious reasoning, via application of
reductio ad absurdum. It starts with bump switches. No magic demon sits
on the bump switch and sends back detailed morse code describing the
tecture and color and arangements of the world sliding by the wisker. No
UART sits on them coding up ascii messages of depth and complexity. No
such information content exists no matter how often the bump switch is
read.

There are no trillions of inputs with 20 readings on 2 sensors taken
over 2 seconds. All that is there is a 1 one part in 20 representation
of a switch closure, not 40 bits of information. There are two parts to
the information. One is purely binary. The switch is open, the switch is
closed. The other part is purely a measure of time. Now matter if the
switches are read 1 time a second, or 1 million times a second, only
(largely useless) resolution is gained on the timing of that bit change.
It is of little "information" significance. Given two bump switches as
the only inputs, the information content is relative between the two
switches, so which switch went first (assuming both closed before escape
action was taken and releaved the first) is about all of significance
that can be made of the situation.

But even my premises that there are only a few bits of information at
most in two bump switches misses the point I tried to make in my post,
by putting the horse before the cart (or some such).

I was focused on outputs. If your robot has limited output combinations,
it doesn't matter how intelligent it might otherwise be. It can only
express itself by having a very few states.

Here the poster child should not be Helen Keller, but Terri Schiavo. If
there are no motor functions available to evidence intelligence, we tend
to assume there is none.

jBot is a wonderful and talented robot, a credit to the state of the
art. But it can't even wag it's tail, when it is finished hunting it's
spot.

So without much over splitting of the analog control of outputs (so as
to have infinite responses, like: this is a one meter circle, this is a
1001 millimeter circle, this is a 1002 millimeter circle, and so on) we
ought to be able to characterize all possible behaviors, as evidenced by
output settings. Then given all the possible inputs in terms of
relationships (again avoiding analog hair splitting) we should be able
to come up with an equivalent machine, by linking all output states by
input data relations, and be able to describe the complexity of the
machine we see.

Complex behaviors like escapes? perhaps they are simple behaviors
applied according to passage of time. Then that takes me back full
circle to the original premise. The intelligence is in the changes of
behaviors, and not the behaviors themselves.

--
Randy M. Dumse
www.newmicros.com
Caution: Objects in mirror are more confused than they appear.


.

Relevant Pages

  • Re: Where is behavior AI now?
    ... of a switch closure, ... something, like the bump switches, increases the information content in ... than the desired reading. ... behaviors (oh, ...
    (comp.robotics.misc)
  • Re: Where is behavior AI now?
    ... representation of a switch closure, ... history of what has been happening, even if that history is only the switch ... the past 5 switch transitions of data. ... behaviors (oh, ...
    (comp.robotics.misc)
  • RE: Caching a sniffer
    ... I can think of at least four behaviors, each of which would give different ... Dump the entire MAC table. ... Switch acts as if power on reset just ... Shutdown port - assume hostile intent and stop forwarding traffic. ...
    (Security-Basics)
  • Re: What can be done in FORTRAN that cannot be done in C/C++?
    ... >>compliant) simply use a switch to get the old behaviour. ... > If something is declared to be standard, ... > than a whim to declare it to be non-standard. ... need to do is memorize those default behaviors. ...
    (comp.lang.fortran)