Learning and evolution, continued

[From Bruce Abbott (951106.1500 EST)]

Shannon Williams (951105) --

Bruce Abbott (951104.0945 EST)

Well, THAT was interesting! I had intended to "park" the post I was
composing while I went off to teach class, and snagged the "send" button
instead. So, out it went, unsigned and incomplete. Here's the rest.

When the finches' beaks are selected to be longer, what is the 'coincidence
detector'? Is this 'coincidence detector' only active for a short time?

Me:

Do control systems reproduce? Does "like beget like"? What does it mean to
be "successful"? How does it happen that successful variants are "preserved"?

Before I got butter-fingered, I had meant to complete this paragraph by
stating that, as these questions suggest, the evolution of species and the
evolution of behavior are not strictly parallel; your questions assume that
they are. Thus the implied absurdity of my "coincidence detector" notion
emerges from an improper comparison.

I would say that when a control loop exists, it always exists. But it is
only emitting output when it is activated. When its reference level is
restored, it stops emitting output. There does not need to be a
'coincidence detector'.

You are describing how an ECU (elementary control unit) works, not how one
is "selected." Yes, there is no need for a "coincidence detector" in an ECU.

your "mechanism X" is hypothesized to be activated when the door opens.
Why? What's significant about the door opening?

When the door opens is when the new control loop is restored to its
reference level.

How does the new control loop come to have "door open" as its reference level?

And how does the cat identify which of its actions is the effective one?

That is mechanism X. Somehow the control loop must *remember* what outputs
worked. It must select for these and inhibit others.

How does it "know" which outputs "worked"? What is it "remembering"?

(Are you always in such a hurry?) Once we start testing loops, then we
will develop a picture of them. We will start describing them. We will be
in 'functional specification' phase again.

I like to know where I'm going; it helps me to assess whether the path I'm
thinking of taking will get me there. In this regard, it seems to me that
the final structure of the control system is unlikely to provide much
information about how it got that way. Would it really make much
difference, theoretically, if it turned out that the cat had learned to
control for "pressure of pole against flank" as opposed to "approach pole
from right and turn left"? Once I know what perceptions the cat is
controlling, how does that tell me how those systems came into being? It
seems to me that I am not very much further along in finding an answer to my
question.

Regards,

Bruce

[From Shannon Williams (951108)]

Bruce Abbott (951106.1500 EST)--

Do control systems reproduce?

Don't know yet. (I think) that they cause new systems to come into
existence. Is that reproduction?

Does "like beget like"?

Don't know yet. Doubt it.

What does it mean to be "successful"?

Control loop A is disturbed. Control loop A attempts all of its normal
outputs to restore its reference level. The normal outputs fail. Loop A
spawns a potential control loop B. Loop B activates pre-existing loops
C,D,E... in random order. If C,D,E... have been activated when the
reference level to loop A is restored, then C,D,E... is "successful", and
is remembered by loop B.

Or...

Control loop A is disturbed. Control loop A attempts all of its normal
outputs to restore its reference level. The normal outputs fail. Loop A
spawns loop B which activates loops C,D,E... Also spawned is loop F which
activates E,D,C... Also spawned is loop G which activates D,C,E... The cat
synchronously runs through B,F,G,etc. When loop A is restored, the loop
B,F,G that is being attempted is considered "successful".

Or...

There are many different scenarios. We need to know more about the loops
first.

How does it happen that successful variants are "preserved"?

Don't know yet.

the evolution of species and the evolution of behavior are not strictly
parallel

How can they be other than parallel? Evolution occurs because random
processes/happenings get trapped/mapped into control loops.

When the door opens is when the new control loop is restored to its
reference level.

How does the new control loop come to have "door open" as its reference level?

It doesn't necessarily. Its reference level could be set by the control
loops which motivate it. The new control loop(s) are probably unaware that
a door exists.

Would it really make much difference, theoretically, if it turned out that
the cat had learned to control for "pressure of pole against flank" as
opposed to "approach pole from right and turn left"?

If you want to verify that the movements of the cat have purpose and
eventually, through a process of selection, the cat organizes its purposes
to cause its escape, then yes, it makes a difference.

Once I know what perceptions the cat is controlling, how does that tell me
how those systems came into being?

You may not know how those systems came into being, but you can predict
what those systems might cause the cat to do in another environment. Or,
if you know the controlled variables of a person, then you can predict how
the person will behave in different environments.

-Shannon