Modeling Emotions

CSGnet Archive CSG2003
1

[From Bruce Gregory (2003.12.15.0955)]

At the risk of muddying the waters, I will simply note that my Traffic
story provides an example of strong emotions that do not have to be
modeled in order to make sense of behavior. A simple perceptual control
hierarchy (a toy model) explains the broad features of my behavior
without any need to worry about the details of adrenalin surges and
expletives.

Bruce Gregory

"Everything that needs to be said has already been said. But since no
one was listening, everything must be said again."

                                                                                Andre Gide

2

[From Bill Powers (2003.12.15.0934 MST)]

Bruce Gregory (2003.12.15.0955)]

At the risk of muddying the waters, I will simply note that my Traffic
story provides an example of strong emotions that do not have to be
modeled in order to make sense of behavior. A simple perceptual control
hierarchy (a toy model) explains the broad features of my behavior
without any need to worry about the details of adrenalin surges and
expletives.

I'm not sure what the function of the expletives would be, but in a
complete model there would be a place for the adrenaline and its effects.
The preparation for action that occurs when large errors appear calls for
various physiological adjustments to support the action. In a model without
such preparations, the output functions would have to be supplied at all
times with enough energy to take care of the most energetic action that
might ever be needed -- otherwise, when the large error occured, there
wouldn't be enough ATP or muscle tone to sustain any large effort, and the
error would not be corrected. However, this continual state of preparation
would come at a cost which is unnecessary most of the time.

So if I were designing a Mars Rover, I think I would probably equip it with
emotions, turning most of the systems down to standby or low-energy states
when not needed, but being able to increase their level of activation
quickly when errors began to get large. The higher-level systems would
probably be organized to crank up the activation at the same time, or even
before, the reference levels calling for strenuous actions. And of course
they would need to detect the resulting change in activation level so as
not to call for large efforts before the power supply was ready to handle
the increased load. My Mars Rover would psych itself up before trying to
climb a steep hill, but remain calm.and cool while trundling along a level
plain or coasting down a gentle slope. And at sunset, when the light of day
was fading and the air was chilling down, I think I'd have my Rover become
depressed, so it didn't have the energy or ambition to do anything.

Best,

Bill P.

3

from [Marc Abrams (2003.12.15.1644)]

[From Bruce Gregory (2003.12.15.0955)]

At the risk of muddying the waters, I will simply note that my Traffic
story provides an example of strong emotions that do not have to be
modeled in order to make sense of behavior.

First, this depends on how you define emotions. I group emotions and bodily
'feelings' together. In my mind 'error' _is_ emotion. The 'error' signal is
in fact an emotional signal for action. But emotions play a much bigger part
than just that. Emotions are the basis for how we 'feel' at any point in
time, and how we 'feel' has a major influence over both what we percieve and
what our imagination might be at any point in time. Emotions provide the
backdrop or context to our perceptions. We tend to try and do things that
make us feel 'good' and try to stay away from things that don't. So we
'pick' things that we imagine will give us 'good' feelings. This is _all_
about emotions.

A simple perceptual control hierarchy (a toy model) explains the broad

features of my behavior

without any need to worry about the details of adrenalin surges and

expletives.

Sure, and so would a Cog sci model and a behavioristic model. Both of those
theories can and do explain that event to many people very plausibly. If
you're happy with what the HPCT model provides you, _wonderful_. Use it,
enjoy it, and have a ball with it. I have no quarrel with you. I just think
there is more to it then that. I guess, after I finish my studies and do a
bit of research I'll be right there where you are, but until then I'm just
going to have to walk down my own path and find out for myself. I appreciate
the effort in trying to save me from some grief and wasted efforts, but If
you can't find an example of where emotion needs to be modeled than we have
nothing to talk about with regard to that subject, because we simply view
things from a different perspective. No crime in that. :slight_smile:

Marc

4

[From Bruce Gregory (2003.12.15.1743)]

Marc Abrams (2003.12.15.1644)

First, this depends on how you define emotions. I group emotions and
bodily
'feelings' together. In my mind 'error' _is_ emotion. The 'error'
signal is
in fact an emotional signal for action.

i don't think that anything is gained by saying that error in the HPCT
model "is" emotion, unless you have a model that incorporates the
conjectured equality. Error is a well-defined mathematical entity;
emotion is not yet quantifiable.

But emotions play a much bigger part
than just that. Emotions are the basis for how we 'feel' at any point
in
time, and how we 'feel' has a major influence over both what we
percieve and
what our imagination might be at any point in time. Emotions provide
the
backdrop or context to our perceptions. We tend to try and do things
that
make us feel 'good' and try to stay away from things that don't. So we
'pick' things that we imagine will give us 'good' feelings. This is
_all_
about emotions.

Everything you say may be perfectly true and still not be relevant to
the HPCT model. It seems to me that you are not describing a model. I
can't imagine anyone disagreeing with your generalizations.

A simple perceptual control hierarchy (a toy model) explains the broad

features of my behavior

without any need to worry about the details of adrenalin surges and

expletives.

Sure, and so would a Cog sci model and a behavioristic model.

Can you point me to such models? If they exist I am not aware of them.
it would be very interesting to compare the predictions of these models
with the prediction of a simple HPCT model.

Both of those
theories can and do explain that event to many people very plausibly.

Indeed. But plausibility is not the point, is it? I associate
plausibility with just-so-stories, not with the predictions of models.

If
you're happy with what the HPCT model provides you, _wonderful_. Use
it,
enjoy it, and have a ball with it. I have no quarrel with you. I just
think
there is more to it then that.

If you mean that there is much work to be done on developing HPCT
models, I suspect that everyone will agree with you. Certainly Bill and
Rick do.

I guess, after I finish my studies and do a
bit of research I'll be right there where you are, but until then I'm
just
going to have to walk down my own path and find out for myself. I
appreciate
the effort in trying to save me from some grief and wasted efforts,
but If
you can't find an example of where emotion needs to be modeled than we
have
nothing to talk about with regard to that subject, because we simply
view
things from a different perspective. No crime in that. :slight_smile:

I don't think it is a different perspective, I think it is a different
understanding of models and their limitations.

Bruce Gregory

Everything that needs to be said has already been said. But since no
one was listening, everything must be said again."

                                                                                Andre Gide

5

from [Marc Abrams (2003.12.15.1842)]

[From Bruce Gregory (2003.12.15.1743)]

I don't think it is a different perspective, I think it is a different
understanding of models and their limitations.

Yes, your probably right here. But those different understandings provide
different perspectives. It's interesting that Rick did not know what a 'toy'
model was. I guess we all have different ideas about what a model should or
shouldn't have. Why do you think your definition is the 'correct' one? And
since I'm a little slow in the uptake, please explicitly lay out what a
model does and does not have and how a model differs from a theory? This
would be appreciated, so at least when conversing with you I would know what
your boundaries are. Thanks,

Marc

6

[From Bruce Gregory (2003.12.15.1911)]

Marc Abrams (2003.12.15.1842)

[From Bruce Gregory (2003.12.15.1743)]

I don't think it is a different perspective, I think it is a different
understanding of models and their limitations.

Yes, your probably right here. But those different understandings
provide
different perspectives. It's interesting that Rick did not know what a
'toy'
model was. I guess we all have different ideas about what a model
should or
shouldn't have. Why do you think your definition is the 'correct' one?
And
since I'm a little slow in the uptake, please explicitly lay out what a
model does and does not have and how a model differs from a theory?
This
would be appreciated, so at least when conversing with you I would
know what
your boundaries are. Thanks,

I share Rick's view with one small modification. I would replace
"current model" in the first line with "current theory." I don't think
it is the "correct" view. It simply reflects what scientists mean when
they talk about theories and models.

  Rick Marken (2003.12.14.1415)

I think the current model is a set of principles that can be used as
the basis for building detailed models of specific behaviors. That's
what Bill did with the Crowd agents and the Little man. That's what I
did with the baseball outfielder and Rx writer. The model of the
outfielder was not described in B:CP. I had to figure out how to build
it based on principles that were implicitly and explicitly described in
B:CP. For example, the principle of one perception controlled per
control system is implicit in the HPCT model described in B:CP. This
simple little principle (along with the much more important, but
equally simple, principle of control of perception) is the basis of the
outfielder model and the reason for its success.

Bruce Gregory

"Everything that needs to be said has already been said. But since no
one was listening, everything must be said again."

                                                                                Andre Gide

7

[From Bruce Gregory (2003.12.15.1937)]

Marc Abrams (2003.12.15.1842)

Since you are studying physics, let me suggest the following example.
Newton's theory is that successful models of the physical world can be
built using three 'laws" of motion and the "law" of gravity. It is left
as an exercise for the reader to actually construct such models and to
compare their predictions with observations. Newton provided some very
helpful examples, of course.

Bruce Gregory

"Everything that needs to be said has already been said. But since no
one was listening, everything must be said again."

                                                                                Andre Gide

8

from [Marc Abrams (2003.12.15.2008)]

[From Bruce Gregory (2003.12.15.1911)]

I share Rick's view with one small modification. I would replace
"current model" in the first line with "current theory." I don't think
it is the "correct" view. It simply reflects what scientists mean when
they talk about theories and models.

  Rick Marken (2003.12.14.1415)

I think the current model is a set of principles that can be used as
the basis for building detailed models of specific behaviors. That's
what Bill did with the Crowd agents and the Little man. That's what I
did with the baseball outfielder and Rx writer. The model of the
outfielder was not described in B:CP. I had to figure out how to build
it based on principles that were implicitly and explicitly described in
B:CP. For example, the principle of one perception controlled per
control system is implicit in the HPCT model described in B:CP. This
simple little principle (along with the much more important, but
equally simple, principle of control of perception) is the basis of the
outfielder model and the reason for its success.

Wonderful, :slight_smile: You still haven't told me _what_ a model consists of. What
makes something a model and something a 'generalization'? What is a
'detailed' model versus one without detail? When does a generalization
become a model? Doesn't a model have to do with the ability to plug numbers
into it and predict system behavior based on the data? Please help me
clarify this.

Thanks,

Marc

9

[From Bruce Gregory 92003.12.15.2052)]

Marc Abrams (2003.12.15.2008)

Wonderful, :slight_smile: You still haven't told me _what_ a model consists of.
What
makes something a model and something a 'generalization'?

F=ma is a generalization. "If a ball is thrown directly upward with an
initial velocity of 30 m/sec, how long wit take to reach the ground?"
is a problem that can only be solved with a model. In this case the
model incorporates Newton's law of gravity and takes the form of

V = 30 m/sec - 10 m/sec2 x t sec.

This model tells us that the velocity of the ball will be zero (the top
of the trajectory) when t= 3 sec. So the total travel time will be 6
sec and when the ball reaches the ground it will be traveling at 30
m/sec.

What is a
'detailed' model versus one without detail?

In this case the model is realistic to the extent that we can ignore
air resistance, wind and the spin imparted to the ball. A more detailed
model would include these two effects.

When does a generalization
become a model?

Never. Models are built using generalizations, generalizations do not
become models.

Doesn't a model have to do with the ability to plug numbers
into it and predict system behavior based on the data?

Yes. See above example.

Bruce Gregory

"Everything that needs to be said has already been said. But since no
one was listening, everything must be said again."

                                                                                Andre Gide

10

[From Bruce Gregory (2003.12.16.1207)]

Bill Powers (2003.12.15.0934 MST)

So if I were designing a Mars Rover, I think I would probably equip it
with
emotions, turning most of the systems down to standby or low-energy
states
when not needed, but being able to increase their level of activation
quickly when errors began to get large.

You old romantic, you. I've never seen love described in quite that way
before.

Bruce Gregory