[From Shannon Williams (960302.11:00)]
Bill Powers (960301.1540 MST)--
There may be completely internal feedback paths, but we can't
investigate them experimentally.Neither can you.
I am not interested in investigating internal feedback paths. I am
interested in designing an intelligent, autonomous unit. If I have to use
internal feedback paths to make my design work, then I will use them.
Besides, you do use internal feedback paths because some of the elements in
an HPCT hierarchy depend upon loops that never see the external world.
To repeat a behavior simply by repeating the signals that produced it
previously is impossible.I agree. What could I say which would convince you that I
understand this?Easy. Stop talking as if it is possible simply to "emit" an organized
behavior.
Stop reading me as if I were saying 'it is possible simply to "emit" an
organized behavior'.
This kind of functional division I can agree with. And if this is
what Hans meant when he mentioned "functional clusters", then I
agree that "functional clusters" are needed.But this is not the division that you are talking about when you
say "going up a level".Yes it is.
When you talk about two people solving their conflicts by "going up a
level", you are not talking about changing one's focus from emitting
behavior to determining what behavior to emit. The two levels are
functional independent, even though the second cannot exist without the
first.
In other words, one cluster determines which way you should jump. The other
cluster determines how you go about jumping.
If you see "going up a level" as the same kind of division as the
physically functional division above, then I think you are assuming
divisions where no divisions are needed.Well, are "functional clusters" needed, or are they not? Make up your
mind.
Sometime they are needed, and sometimes they are not. You do not need two
functional clusters that do the same job.
I think that when we "go up a level", we simply associate a
perception to a different perception.I'm glad to know what you think. Do you have any evidence to support
this idea?
Every bit of your evidence which supports HPCT, supports a control loop
with neural nets. A control loop with neural nets has the added advantage
of explaining learning.
[diagram omitted]. Hans' model has been implemented as a simulation. Can
you demonstrate that your proposed model would actually do anything?
No. I am in the design review stage, and you are not helping me any.
I guess I will return to this list when I have a working product.
-Shannon