[From Bill Powers (991025.0000604 MDT)]
Bob Christensen (991022.1020 PT)--
I really like a lot of what you've been writing lately. And I'm also
enjoying your thesis, which I'm about halfway through.
Bill:
A perception is not necessarily an error signal just because it represents
something that we don't like.
Bob:
How is it that we know a perception is one we don't like or want?
Bill:
The most direct way is that we just get a "bad feeling" about it. That's
vague enough that it could mean almost anything. I think what it means is
that we get avoidance feelings; we find ourselves shrinking away from
something, pushing it off, looking for something to do about it. Since
these feelings are _not_ as clear and direct as ordinary perception, it's
hard to describe them clearly. I think we tend to assume that we know more
about them than we do.
Here's a little exercise. Make a cut in your fingertip (just kidding --
wait until one happens, or until some other painful experience occurs).
Then ask yourself, "How do I know that this feeling is bad?" The
overwhelming impression is that pain IS a bad feeling; anything that feels
that way is automatically bad. But that makes no sense. How could a
perception not only tell you what _is_ happening, but also tell you whether
it is good or bad?
Are there two channels hidden in what seems to be a single pain signal? I
think it makes more sense to assume that the value is in the receiver of
the signal, not in the signal.
If you ask what it is, specifically, that's telling you this sensation is
bad, you'll have a hard time answering. All you really know is that you
feel upset and are tensing up, pulling back, trying to get away from
whatever is causing the pain. But those aren't aspects of the pain itself;
they're what you're trying to do about it, voluntarily or involuntarily. I
think you learn to interpret those actions as being part of the pain. I
think you have a low reference level for pain-signals, and that the error
signals which arise when pain occurs readjust many lower-level reference
signals which in turn results in actions whose effects you can feel (that's
what we mean by "tensing up" -- we mean starting to act and going into
conflict with whatever actions we were previously engaged in). We feel the
consequences of having large error signals, but not the error signals
themselves. In exactly the same way, we don't feel the signals travelling
away from our spinal cords and into our muscles (which are first-order
error signals). We feel only the tensions, pressures, and motions that result.
Bob:
From these statements I gather that one infers that one does not like a
perception upon finding one's self taking actions to avoid them or make
them go away. I assume futher that one can tell if one does not like a
perception when one imagines it going away.
Bill:
Not quite. What I'm proposing is that the "badness" part of the feeling is
made of the sensations arising from one's own actions, not just from the
feelings going away. It's the feeling of _wanting_ it to go away that is
taken to be part of the "bad" value. The "wanting it to go away" is the
imagined thrust against it, not just its fading or moving off. If you
imagine taking action to avoid it or push it away, that might serve to give
the same sense of (negative) value. Of course by reversing everything you
also can explain the positive values we put on perceptions.
Bob:
For instance one cannot immediately make a
dent in one's car go away, but one might start imagining the dent not being
there. This is imagining the reference, isn't it?
Bill:
I don't know how you work, but if that happened to me I think my first
thought would be an urge to beat the culprit to a pulp. Of course that's
imagining a reference, too, as you say. And of course it would produce a
conflict, since a nice guy like me doesn't beat people to a pulp. All those
wishes and conflicts would add up to making the dent into a hell of a bad
thing instead of merely a physical condition to be corrected if I wish to
do so. But most of the badness would be my experience of what I want to do,
and also not do. The dent is just a dent.
Bob:
Now I can toggle back and
forth between Perception (dent), and reference (no dent) leading to a
sequence of the two perceptions stored in memory. Now I can perceive the
sequence simultaineously like when perceiving both the A and the B at the
same time in AB. Now I can perceive that the two perceptions (regular
perception, and replayed reference) are different, just like I can tell that
A is different from B. How does this perception of difference work? I
assume that perceptions of difference use some kind of subtraction (like a
comparitor). When such difference perceptions are applied to sequences of
mismatched perception/reference, this can lead to a perception that is
representative of an error signal. Such a representation might be useful
at a higher level, especially if the lower systems are not automatically
working.
Bill:
You think the way I think. It doesn't always lead to a sense of total
conviction, but it's reasonably useful.
Yes, we can perceive differences between perceptions whether present-time
or remembered, in all combinations. And subtraction is a likely mechanism.
In this way we can _simulate_ perceptions of error signals. However, this
process is not as direct and fast as the way a control system actually
works. And the giveaway is that you can decide to keep this simulated error
signal in any state you want, whereas a real control system necessarily
keeps it at zero. You can aim to make the arrows hit the outer ring of the
target instead of the bullseye; you can maintain the knot one inch, or a
foot, to the left of the dot. That tells you you're really dealing with a
perception, not an error signal. The actual error signal remains hidden
from view (actual error signal = synthetic error signal _not_ as big as you
want it).
Bill:
G. Gordon Liddy explained how a person can hold the palm of
his hand over a candle flame until the flesh bubbles ...
Bob:
Can't this situation also be handled by a conflict in which some other goal
(like wanting to become famous or be a martyr) overides, through much higher
gain, the reference for low pain.
Bill:
Sure. But G. Gordon Liddy claimed that he actually learned not to mind the
pain -- that is, to treat it as just being a signal. Of course that would
make his macho demonstration less impressive -- if you really didn't mind
the pain, what were you proving? That you can injure your hand so you can't
use it for a week?
Smart, very smart.
If a certain
amount of food is perceived as a too-little error, then how could
increasing the amount of food also be perceived as an error?
I'm not sure why this is a problem. If errors where perceived, they would
still have to be compared to a reference.
Yes, that's my point. The value is not in the amount of food perceived to
be on the table, but in our reference level for it. If errors were
perceived, they would be perceptions, not error signals. If you were really
perceiving the error signal, then there would already have been a
perception and a reference signal to produce the error signal. And now you
are going to have a _second_ comparator which compares the error signal
with the desired amount of error signal? No, those are the first two terms
in an infinite series. See above.
Or it could be judged an error when visibly compared with a reference in
imagination as explained above.
Yes, but how big should that perceived error be? If you don't _have_ to
make it zero, it's not a real error signal. Speaking of an error signal, a
component of a neural model, is not the same thing as speaking of a
perceived difference between two other perceptions, whether real-time or
imagined.
Bill:
This is what I'm trying NOT to say. You perceive fullness of your stomach.
But do you also perceive the reference fullness, so you can perceive that
one is greater than the other? I say not.
Bob:
With the imagination connection it's not hard to imagine how this can be done
as explained above.
Bill:
Yes, but then you're talking about a perception, not an actual error
signal, and the reference level for it is optional. How much too full do
you want to be?
Best,
Bill P.