Bill's Gate IIIb

Newsgroups: bit.sci.purposive-behavior
Summary:
Keywords:

From Mervyn van Kuyen (970930 8:30 CET)

[From Bill Powers (970928.0504 MDT)]

Mervyn van Kuyen (970926 20:30 CET) --

Bill, please allow me another chance to make my point.

OK, but you're going to have to give me some help.

(mervyn)

     (Fig.1) (Fig.2) (Fig.3)

     NETWORK NETWORK NETWORK
        A A A
        > > >
   +->(EXC) +->(EXC)<-+ (EXC)<-+
   > A | A | A |
   > > > > > > >
   > (INH)<-+ +->(INH)<-+ +->(INH) |
   > 1 | | 2 | | 1 |
   > > > > > >
  REF SENS REF SENS REF SENS

It is interesting to see how these members each have their distinctions:

(bill)

Before you try to explain that, I need to know what the above arrangement
is designed to accomplish, whichever version you prefer. What does the
network, taken as a unit, _do_? And to what aspect of the behavior of a
living system does its behavior correspond? Does the network have any
physical effects on its environment? Does the sensor detect something in
the physical environment? And what is it that generates REF?

What I'm trying to establish, is what they can accomplish at max,
and _then_ I try to establish what the design should look like
(e.g. where does REF come from). I'm not trying to see how far
I can get with each of them for a _fixed_ set of design choices.
(However, one thing _is_ fixed: SENS is their sensory input)

As I wrote in the original post, I don't think that Fig.3 (on its
own) is capable of very little. I think we want to focus on
the PCT (Fig.1) and neural servo variants (Fig.2)...

But, in Fig 1, REF has to be generated by an independent, predefined
mechanism. Otherwise the system would 'die out' if it faced maximal input
(resulting in no error, no triggering of the network). So this network can
be used to produce _actions_ to minimize the detected mismatch.

In Fig.2 the opposite is true: it detects ALL mismatch, like you've
said before as well, albeit WITHOUT sign. This problem can be solved
however by having the network generate the _reference signals_
as well as _actions_ to try to minimize the mismatch. This allows
it to know the sign perfectly well and exploit this knowledge,
as I wrote two days ago (in 'RE: Bill's Gate IIb'):

[Mervyn van Kuyen (970928 12:30 CET)]

[Bill Powers (970926.0419 MDT)]
If you have just one error signal output, indicating "mismatch" withoutany
indication of the sign, you have at best a hill-climbing system, not a
control system. A control system needs an indication of the sign as well as
the amount of an error, so its action can always be aimed toward reducing
the error. We have discussed this on CSGnet, years ago.

[mervyn]

In PCT the reference is hidden from the network that corrects the
perceived errors. But in my model the network produces the
actions *and* the references. Therefore it has perfect implicit
knowledge about the 'sign' of the disturbance. And you can be sure
it *will* exploit this knowledge because the sign is, as you mention,
a very important property. It will simply react differently to error
signals, depending on the nature of the references it is creating.

So I'm not trying to prove PCT wrong or anything, but where you explicitly
tell us that Fig.2 doesn't work (since it produces no sign), there is a
simple way to have it work, although it makes our designs very
incompatible. Different fundamental design choices _demand_ different
designs. Can we agree on that one?

Regards,

Mervyn

···

From: mervyn@xs4all.nl
Subject: Re: Bill's Gate IIIb

[From Bill Powers (970930.0903 MDT)]

Mervyn van Kuyen (970930 8:30 CET)--

(bill)

Before you try to explain that, I need to know what the above arrangement
is designed to accomplish, whichever version you prefer. What does the
network, taken as a unit, _do_? And to what aspect of the behavior of a
living system does its behavior correspond? Does the network have any
physical effects on its environment? Does the sensor detect something in
the physical environment? And what is it that generates REF?

What I'm trying to establish, is what they can accomplish at max,
and _then_ I try to establish what the design should look like
(e.g. where does REF come from). I'm not trying to see how far
I can get with each of them for a _fixed_ set of design choices.
(However, one thing _is_ fixed: SENS is their sensory input)

What is the output of the neural net, and what does it do to its
surroundings? I'm just asking a general question, not asking about details
of design. As part of a behaving system, what role would this neural net play?

Best,

Bill P.