UEC/NEC summary

CSGnet Archive CSG2002
1

As someone who spends time in the woods following trails with choice points,
on the way to a goal, I think I speed up as I gain confidence that I'm
making the right choices. That's obviously not a smooth gradient, but I
doubt if the measurement of the rats was that fine.
David Wolsk

This reminds me of this weekends Indy 500.

Race teams are given a finite amount of fuel. If they race, flat out, for
500 miles they won't make it. This years winner probably didn't have enough
fuel to finish the race and was being passed by #2 when a yellow flag caused
all drivers to slow down (conserving fuel) and hold their positions. Thus
#1 was able to finish the race (and win)

#2 was increasing his speed (much like the rat) because he had the resources
to accomplish his goal and his goal was in sight. Unfortunately the rules
of the game forced him back into second place position.

It seems to me the further a rat (race driver or any other organism) is from
the goal the more they are conserving and strategizing. The closer the goal
the more resources can be diverted to achieving it. The rat/gradient
example may be moot now, but there are still some points worth
acknowledging.

Do we assume the rat, at the start of the maze, knows there is food at the
end? It could be the rat is always in a search mode for food. This
expends a lot of energy literally going in circles. As a given direction
becomes more promising (closer to the goal) the rat can expend more energy
or at least direct more energy in a given direction. Maybe the gradient
observation is really a change in strategies. The rat shifting from search
mode to acquire mode.

The strategies used near the start of the Indy 500 are not the same ones
used at the end of the race. Rules of the game, track conditions, yellow
flags, and quality/quantity of pit stops all change the dynamics of the
race.

Can we set up a simple experiment with a dog on a leash? A spring scale (or
if anyone has access to a strain gage and computer to log the pull) attached
to the leash. I would move a food dish closer to the dog rather than have
the dog move to the dish. In any case, with the dish several feet from the
dog I think she is going to be in a different mode. She is going to be
looking at the dish and the experimenters and wondering what tricks she is
going to have to perform in order to be rewarded. With a food dish inches
from her nose she is going to be in an acquire or eating mode. The human
handlers are out of her field of view. Surely she has learned that food
next to her nose is for eating and nothing else. Thus she will expend a
large amount of energy in pulling towards what appears to be a sure thing.
Is the gradient smooth or are there transitions?

Steve O

···

-----Original Message-----
From: David Wolsk [mailto:davidwolsk@SHAW.CA]
Sent: Friday, May 24, 2002 1:38 PM
To: CSGNET@listserv.uiuc.edu
Subject: Re: UEC/NEC summary

2

[From Bruce Nevin (2002.05.22 15:11 EDT]
The UEC or NEC (I thought that was a Japanese computer company) is a
proposed mechanism to account for the phenomenon of ‘giving up’, reducing
gain when maximum gain does not reduce error, and conversely a mechanism
that is proposed to increase gain as error diminishes.
First we need clear instances of the phenomenon.
Buridan’s ass and the pill bug are not instances for reasons
discussed.
The ill-mannered horse bolting for the barn to be fed is I believe not an
instance for reasons that I described: I believe the horse is controlling
a sequence perception, the last step of which is getting fed. There is
lots of at least anecdotal evidence for controlling the pace of the steps
of a sequence, performing them rapidly if the last step is desired or
slowly if it is being avoided. I am not aware of any modelling of this.
In fact, I’m not aware of any modelling of varieties of sequence control.
I believe this would be a fruitful place to look.
I think maybe arm wrestling was mentioned as an example in an earlier
spate of UEC discussion. The idea was that a person resists and suddenly
gives up. My experience is that the person resists until their arm is
pushed past the vertical some distance. It appears to me that the strong
muscles of the shoulder, back, and chest that are effective at the peak
of the arc and increasingly so as they push down past the peak are
disadvantaged ‘behind’ the peak of the arc and increasingly so as they
push up toward the descending arm of the other. As you push forward and
down, curving the back forward, to some degree the entire upper body is
being used as a lever whose fulcrum is where you are sitting, and as you
push past the arc you have your body weight to back you up, as it were.
As the other pushes your arm back behind you, to lift your arm you are
curving your shoulder and to some extent your back backward and lifting
your upper body weight as well as the weight of the other transmitted by
their muscular effort. This is why getting past the top of the arc is the
whole game in almost every arm wrestling contest. I believe this has much
to do with the ‘giving up’ in this situation. Also, unless the person is
a neophyte at arm wrestling they remember that it is nearly impossible to
recover once they’ve lost that crucial ground at the balance point.
Nevertheless you do see people resisting until their arm is
actually pressed against the table and indeed it is possible to do this
yourself. So if there is an automatic mechanism for ‘giving up’ and arm
wrestling provides an instance of it, this deliberate prolongation of
fruitless effort shows that it is possible for a higher control system to
override that mechanism. So you have the intervention of higher control
systems even if you add the UEC/NEC to the model.

Therefore, if an alternative explanation is possible with the
intervention of higher control systems, that explanation does not add any
complexity, because the intervention by higher control systems has to be
there anyway even if you have the UEC/NEC mechanism.

    /Bruce
3

[From Bill Williams 22 May 02 CST 19:00]

[From Bruce Nevin (2002.05.22 15:11 EDT]

First we need clear instances of the phenomenon.

I don't think you've explained why you don't see the superior good in the
Giffen effect as an instance of the phenomena. From a position in which there
is more than enough money for the consumer to supply calories by consuming the
superior good, an increase in price will result in first more money being spent
on the superior good, and then when the budget and caloric lines crossing less
money being spent on the superior good. This seems to me could be considered an
instance of the phenomena.

Bill Williams

···

______________________________________________________________________
Do you want a free e-mail for life ? Get it at http://www.email.ro/

4

[From Rick Marken (2002.05.22.1800)]

Bruce Nevin (2002.05.22 15:11 EDT)

The UEC or NEC (I thought that was a Japanese computer company) is a
proposed mechanism to account for the phenomenon of 'giving up'

I think the UEC is an explanation of "giving up" control actions without
"giving up" the reference for the controlled variable. So "giving up"
is kind of an ambiguous way of describing the phenomenon explained by
the NEC. It's giving up my efforts to consume the Jack Daniels that I
still want, for example.

This is not an easy phenomenon to demonstrate. It is not demonstrated,
for example, by simply seeing that I no longer act to consume Jack
Daniels (that I no longer produce a particular controlled result). I may
not be drinking Jack Daniels because I no long want (have a reference)
for it. If that's the case, then we already have an explanation for
_that_ phenomenon: hierarchical control. What we have to show is both
that I no longer consume Jack Daniels _and_ that I _want_ to consume it
and _will_ consume it if error (the difference between my goal and the
intended result) is made _small_ enough!

I think we need to develop clever, artificial situations to demonstrate
this phenomenon (_if it occurs_). I am not sure it _does_ occur. It
_seems_ to me that it does occur. But that's just my subjective
experience. I'm still working on an experimental demo of the NEC
phenomenon. So far, no luck.

The NEC is, thus, an explanation for a phenomenon that may not actually
exist. I think it does exist but demonstrating its existence will not be
easy. Again, this is because any such demonstration must show clearly
that the controller has given up the actions but _not_ the goal of
control.

Buridan's ass and the pill bug are not instances for reasons
discussed.

Probably not. I agree.

The ill-mannered horse bolting for the barn to be fed is I believe not
an instance for reasons that I described:

I think this one is still pending. Your explanation (hierarchical
control) might be right; but the NEC could be right, too. We don't know,
from observation alone, whether the horse gave up the goal of getting to
the barn (as per your explanation) or not (as per the NEC explanation).

I think the only way to approach this question (whether or not there is
an NEC phenomenon) is to do the research that reveals it. Of course,
that places the burden on those of us who think there _might_ be an NEC
phenomenon. Those of you who don't believe in such a phenomenon can rest
easy because there has been no demonstration of it. So the UEC
hypothesis is, at worst, unnecessary

Best regards

Rick

···

---
Richard S. Marken
MindReadings.com
marken@mindreadings.com
310 474-0313

5

[From Bill Williams 22 May 02 CST 20: ]

There have been experimental economic studies in which a consumer buys an
icecream cone for one price. Then the consumer is offered an increasing
sequence of prices for the icecream cone. But, the consumer demands quite a bit
more for the cone than the purchase price. This result creates difficulties for
the orthodox economic theory of choice-- which would lead one to think that the
consumer ought to be willing to sell the cone for only slightly more than was
paid for it.

Bill williams

···

______________________________________________________________________
Do you want a free e-mail for life ? Get it at http://www.email.ro/

6

[From Bruce Nevin (2002.05.22 15:11 EDT]

Bill Williams (22 May 02 CST 19:00) –

···

At 02:58 AM 5/23/2002 +0300, William Williams wrote:

I don’t think you’ve explained why you don’t
see the superior good in the

Giffen effect as an instance of the phenomena.

You’re right, I forgot that. Mea culpa. This looks like an instance of
the phenomenon, but the model that produces this behavior does so
by the interaction of control systems, it does not depend upon a UEC or
NEC mechanism to do so. So it’s a counterexample to the UEC/NEC
explanation of the phenomenon.

    /Bruce
7

[From Bruce Nevin (2002.05.22 15:11 EDT]

Bill Williams (22 May 02 CST 19:00) --
>I don't think you've explained why you don't see the superior good in the
>Giffen effect as an instance of the phenomena.

You're right, I forgot that. Mea culpa. This looks like an instance of the
phenomenon, but the model that produces this behavior does so by the
interaction of control systems, it does not depend upon a UEC or NEC
mechanism to do so. So it's a counterexample to the UEC/NEC explanation of
the phenomenon.

OK. So, maybe it would be helpful if a distinction could be introduced between
UEC/NEC as a caption for a Phenomena and UEC/NEC as a caption for an
explaination?

Bill Williams

···

At 02:58 AM 5/23/2002 +0300, William Williams wrote:

______________________________________________________________________
Do you want a free e-mail for life ? Get it at http://www.email.ro/

8

[From Rick Marken (2002.05.23.0850)]

Bruce Nevin (2002.05.23 11:17 EDT) --

>
>it would be helpful if a distinction could be introduced between
>UEC/NEC as a caption for a Phenomena and UEC/NEC as a caption for an
>explanation?

Indeed.

As a phenomenon or as a family of phenomena, we are looking for examples.
The Giffen effect seems to be an example, maybe the only sure example we
have so far.

The Giffen effect seems to be a clear cut example of hierarchical control. It
doesn't look like a NEC phenomenon to me.

The Giffen effect is paradoxical only if we look at the relationship between
the price of and consumption of one product, say, bread. What we see is that
an increase in price leads to an _increase_ in consumption of bread. This
relationship between price and consumption is paradoxical only in the context
of certain (admittedly common) assumptions about what the relationship between
these variables _should_ be. But when you look at the picture from a
hierarchical control perspective there is nothing paradoxical happening at
all. The consumer is adjusting consumption of a variety of different foods,
each costing a different amount, as the means of controlling both caloric
intake and expenditures simultaneously. A simple, two level hierarchy of
control systems readily accounts for this result (when the higher level
reference for expenditures is set to an appropriately low -- poor -- level).

The NEC phenomenon is a decrease in effort with an increase in error and an
increase in effort with a decrease in error. This seems to be happening in the
horse and barn example. As the horse moves away from the barn (resulting in a
presumed increase in error) there is eventually a decrease in effort (the
horse stops trying to get back to the barn) and as the horse returns to the
barn there is a sudden _increase_ in effort as the horse starts getting close
to the barn (resulting in a presumed decrease in error). So the horse and barn
phenomenon seems like it _might_ be based on a NEC. But it might not; the
horse may be revising it's reference for distance to the barn depending on its
distance from the barn; so the phenomenon may actually be based on
hierarchical control; but at least it _looks like_ it could be based on a NEC.
I don't see why you think the Giffen effect suggests a NEC.

Best regards

Rick

···

At 06:49 AM 5/23/2002 +0300, William Williams wrote:

--
Richard S. Marken, Ph.D.
The RAND Corporation
PO Box 2138
1700 Main Street
Santa Monica, CA 90407-2138
Tel: 310-393-0411 x7971
Fax: 310-451-7018
E-mail: rmarken@rand.org

9

[From Rick Marken (2002.05.23.0915)]

Bruce Nevin (2002.05.23 11:17 EDT) –

Actually there are other examples of the phenomenon
in existing control theory demos. We know what the control systems in the
demos are doing that results in the appearance of ‘giving up’. This can
help us to look with a discriminating eye at the appearance of ‘giving
up’ in [our perceptions of] the natural world around us.
Again, what is “giving up”? Is it “giving up” actions while the reference
for input remains or is it giving up the reference for input itself? The
NEC explains the former type of “giving up” (if it happens); hierarchical
control explains the latter. You cannot tell what type of “giving up” is
actually occurring by simply looking – even with a “discriminating eye”
– at the appearance of behavior.
I don’t believe there is any valid way to resolve the question of
whether or not there is really a need for the NEC hypothesis other than
by doing the appropriate PCT-based research. Modeling alone won’t do
it. You’ve got to “dare to disturb” the universe. That is, you’ve
got to carefully push on some controlled variables and see what happens.

Best regards

Rick

···

Richard S. Marken, Ph.D.

The RAND Corporation

PO Box 2138

1700 Main Street

Santa Monica, CA 90407-2138

Tel: 310-393-0411 x7971

Fax: 310-451-7018

E-mail: rmarken@rand.org

10

[From Bruce Nevin (2002.05.23 11:17 EDT]

it would be helpful if a distinction could be introduced between
UEC/NEC as a caption for a Phenomena and UEC/NEC as a caption for an
explanation?

Indeed.

As a phenomenon or as a family of phenomena, we are looking for examples. The Giffen effect seems to be an example, maybe the only sure example we have so far.

As an explanation, the label has referred to the proposal that something in the output function of control systems is the cause of the phenomenon. An advantage is that if this is valid then modelers can write a bit of code once and use it everywhere.

To make the distinction, maybe we should reserve NEC for the phenomenon and use related terms that refer specifically to the class of mechanisms being proposed. For example:

         NECOF -- NEC from an output function mechanism
         NECCI -- NEC from control system interaction
         NECS -- NEC from control of a sequence

as distinct from

         NEC -- Nonlinear error curve contrary to expectation,
                 either reducing output despite increasing error
                 ('giving up') or increasing output despite
                 diminishing error. Unequivocal instances are
                 being sought.

         /Bruce

···

At 06:49 AM 5/23/2002 +0300, William Williams wrote:

11

[From Bruce Nevin (2002.05.23 11:17 EDT]

Bruce Nevin (2002.05.23 11:35 EDT)–

The Giffen effect seems to be an example,
maybe the only sure example we have so far.

Actually there are other examples of the phenomenon in existing
control theory demos. We know what the control systems in the demos are
doing that results in the appearance of ‘giving up’. This can help us to
look with a discriminating eye at the appearance of ‘giving up’ in [our
perceptions of] the natural world around us.

In the e. coli chemotaxis demo, the agent could be said to ‘give up’
because of rising error when it starts tumbling. (The e. coli organism
‘gives up’ consuming nutrients and starts expending energy
tumbling.)

In the Crowd demo, we frequently observe an agent ‘giving up’ a path
toward its target and backtracking. Indeed, we might say this of a person
making a way through a crowd, that they ‘gave up’ trying to get through
where there were lots of people and backtracked to try another
route.

What other examples of the appearance of ‘giving up’ can we see in our
existing demos? Are there any examples of the converse, increasing effort
as error diminishes?

    /Bruce
···

At 11:17 AM 5/23/2002 -0400, I said:

12

[From Rick Marken (2002.05.23.1330)]

Me

> I don't see why you [Bruce Nevin] think the Giffen effect suggests a NEC.

Bill Williams (23 May 02 CST 13:30)

Its the behavior of the superior good that looks as if it might be an instance
of a Non-linear Error Curve. If the budget is well above the level required to
purchase all of the calories by consuming only the superior good then-- when the
price of the superior good is increased, _at first_ the amount expended on the
superior good increases. Then when the price of the superior good has risen to
the point that it is neccesary to consume some of the inferior good to ubtain
the required calories _then_ expenditures which had gone to the superior good
will begin to shift into the cheaper source of calories. So, first as a result
of the price increase for the superior good expenditure increase, and then when
the price increases still further expenditures on the superior good begin to
decrease.

Ah. I think I see. The consumption of the superior good is the controlled variable.
The amount expended on the superior good is the action used to get the amount
consumed to its reference level. The price of the superior good is a disturbance to
consumption of that good. So what you are referring to is the relationship between
action and disturbance, which is an inverted U: as the price of the superior good
(disturbance) goes up the amount purchased (action) first increases and then
decreases. This inverted U looks like an NEC and _would be_ an NEC if the
disturbance were actually proportional to error. Is this right, Bill?

This is a good example of an apparent NEC, since we know that, in the model,
anyway, the change in action results from a change in the reference for the amount
of superior good consumed.

Best regards

Rick

···

---
Richard S. Marken, Ph.D.
The RAND Corporation
PO Box 2138
1700 Main Street
Santa Monica, CA 90407-2138
Tel: 310-393-0411 x7971
Fax: 310-451-7018
E-mail: rmarken@rand.org

13

[From Bill Williams 23 May 02 CST 13:30]

[from Rick Marken ]

> [frm Bruce Nevin]

> As a phenomenon or as a family of phenomena, we are looking for examples.
> The Giffen effect seems to be an example, maybe the only sure example we
> have so far.

The Giffen effect seems to be a clear cut example of hierarchical control. It
doesn't look like a NEC phenomenon to me.

I don't see why you [Bruce Nevin] think the Giffen effect suggests a NEC.

Its the behavior of the superior good that looks as if it might be an instance
of a Non-linear Error Curve. If the budget is well above the level required to
purchase all of the calories by consuming only the superior good then-- when the
price of the superior good is increased, _at first_ the amount expended on the
superior good increases. Then when the price of the superior good has risen to
the point that it is neccesary to consume some of the inferior good to ubtain
the required calories _then_ expenditures which had gone to the superior good
will begin to shift into the cheaper source of calories. So, first as a result
of the price increase for the superior good expenditure increase, and then when
the price increases still further expenditures on the superior good begin to
decrease.

Bill Williams

···

______________________________________________________________________
Do you want a free e-mail for life ? Get it at http://www.email.ro/

14

[From Bruce Nevin 2002.05.23 17:51 EDT]

Rick Marken (2002.05.23.1330)--

···

At 01:31 PM 5/23/2002 -0500, Richard Marken wrote:

This is a good example of an apparent NEC, since we know that, in the model,
anyway, the change in action results from a change in the reference for the amount
of superior good consumed.

Keeping the distinction between phenomenon and mechanism to which Bill Williams so astutely called our attention, the Giffen effect is a good example of NEC the phenomenon, but not of NECOF, a mechanism in the output function of control systems whose effect is the NEC phenomenon.

         /Bruce

15

[From Bill Williams 23 May 02 21:14]

[From Rick Marken (2002.05.23.1330)]

Me

> > I don't see why you [Bruce Nevin] think the Giffen effect suggests a NEC.
>
Ah. I think I see. The consumption of the superior good is the controlled

variable.

The amount expended on the superior good is the action used to get the amount
consumed to its reference level. The price of the superior good is a

disturbance to

consumption of that good. So what you are referring to is the relationship

between

action and disturbance, which is an inverted U: as the price of the superior

good

(disturbance) goes up the amount purchased (action) first increases and then
decreases. This inverted U looks like an NEC and _would be_ an NEC if the
disturbance were actually proportional to error. Is this right, Bill?

Almost. Starting from an initial condition in which the consumer has more than
enough money to purchase more than enough meat (the superior good) to supply
all the calories required, then-- at first when the price of meat increases
more money will be spent on purchasing meat. So, at first as the price of meat
increases more will be spent on meat while the consumption of meat stays the
same. THe "meat error" when the budget line is above the caloric requirement on
the superior good axis is zero. But, the "output" the money expended to keep
the meat error zero is increasing with increases in the price of meat. When the
consumer gets to the point where it is neccesary to purchases some of the
cheaper ( cheaper in the sense of dollars/unit of calories )good then a "meat
error" develops. But, this meat error is less urgent than the possiblity of
developing a caloric errror. Therefore, to prevent a caloric error
expenditures shift from meat to expenditures on some cheaper source of calories
like bread or potatoes. This generates a plot for the price of meat vs
expenditures on meat in which at first the expenditures on meat increase in
proportion to the increasing price of meat-- up to the point at which it is
neccesary to shift consumption away from meat to a cheaper source of calories.
At this point the relationship between the price of meat and the expenditures
on meat shifts. Now, further increases in the price of meat generate a
decrease in expenditures on meat.

This, the reaction of expenditures on meat-- first increasing and then
decreasing as the price of meat is increased, isn't the relationship upon which
the Giffen Paradox literature has focused. Orthodox economists are bothered by
the possiblity that there might be a good for which an increase in the price of
the good generates an increase in the volume of the good purchased. They worry
about this possiblity because--- if such behavior actually occurs, then they
will not be able to assume stablity in the market-- where they need to be able
to believe that a small disturbance ( a price ) increase will generate a
"negative feedback." If the intial price increase results instead in more of
the good being demanded, then this it is thought will generate a price increase
in the market and the "positive feedback" will destroy their argument that the
market is always stable and consumer behavior will always generate equilibrium
and other good stuff.

The point I think regarding the behavior of the superior good ( meat ) is that
using ordinary control loops, an output "expenditure" is generated that first
increases as the disturbance ( the price increases ) is applied and then at
some definable point begins to switch and generate a decrease in response to
still further price increases. It doesn't quite match the UEC in the sense
that while the price of meat is a first generating more output ( expenditure )
the error for meat consumption remains at zero. So, what is involved is a
disturbance to output relationship rather than an association between error and
output. The peculiar feature of this case may be a result of the increasing
disturbance, at first, not generating an error.

Bill Williams

···

______________________________________________________________________
Do you want a free e-mail for life ? Get it at http://www.email.ro/

16

[From Bill Powers (2002.05.23.1951 MDT)]

Bruce Nevin (2002.05.23 11:17 EDT)--

>In the e. coli chemotaxis demo, the agent could be said to 'give up'
because of rising >error when it starts tumbling. (The e. coli organism
'gives up' consuming nutrients >and starts expending energy tumbling.)

In the Crowd demo, we frequently observe an agent 'giving up' a path
toward its target and backtracking. Indeed, we might say this of a person
making a way through a crowd, that they 'gave up' trying to get through
where there were lots of people and backtracked to try another route.

What other examples of the appearance of 'giving up' can we see in our
existing demos? Are there any examples of the converse, increasing effort
as error diminishes?

The only reason these examples can "be said" to show the "giving up"
phenomenon is that the meaning of the term changes, and we use a very
liberal interpretation of what we see. I would not like to see the
phenomenon trivialized in that way -- if it really is that trivial, we
should just forget about it.

Perhaps someone with access to a large University library can find the
paper describing the effect as I first heard of it. The authors will (I
believe) be Dollard and Miller, or perhaps one of them alone, although the
first author could have been a graduate student. The subject is "approach
gradients," and the observation was that as a rat approached a food dish,
its efforts to get closer increased, as measured through a tether
restraining the rat. It is this effect we have to explain, although if it
is seen only in rats and only in that experiment, finding an explanation
may not be very important. Look up _Dollard Miller approach gradient_ in
Google and you'll see what you get into.

I'm concerned about that experiment, because it seems to contradict PCT: as
any error gets smaller, the effort used to correct it ought to get smaller,
or its rate of increase ought to be smaller. Instead, the proximity of the
food dish seems to serve as a stimulant to the approach efforts, which is
of course the S-R interpretation Dollard and Miller put on that sort of
thing. Whoever did that experiment should have sensed something wrong,
because if the effort really kept increasing all the way to the food dish,
the rat should have gone shooting by it instead of stopping to eat. But we
still have to worry about the behavior far from the food dish. And at least
as I do things, I can't explain it just by cooking up some imaginary
higher-level systems defined so they will do just what we observe. If
that's all that's happening, there are too many scenarios that would give
the same appearance, and none of them is compelling for that reason.

A key feature of this phenomenon was that there was a _gradient_ of effort,
so the effort was a smooth function of distance (in the wrong direction for
PCT). That is hard to explain in terms of a logical system that is
reasoning symbolically about what's happening. Whatever is creating the
appearance of a change in gain is doing it in an analog fashion.

We can model a system in which the gain of the output function is high for
errors below some threshold and declines for errors above that threshold in
proportion to the excess. We don't need to say what is causing the gain to
decline with increasing error in this way -- whether it's a learned
algorithm, competition between contradictory goals, something
physiological, or merely an illusion created as something else entirely is
controlled. The explanation will not alter the effect of having an output
function behave in that way as the error changes.

The simplest simulation experiment will be to vary the sensitivity of the
gain-changing relationship to excess error. With zero sensitivity, the
effort will simply go on increasing as long as the error increases. As the
sensitivity is slowly raised, we will start to see the increase of effort
falling below a linear relationship. At some sensitivity, the effort will
just level off, as if the output function were saturating, and at still
higher sensitivities we will start to see the effort decreasing as error
increases further (and of course, increasing if the error then decreases).

These changes in the output function will not be directly reflected in the
relationship between disturbance and effort. A lot will depend on the
feedback function and the disturbance function. If the disturbance is
coupled to the controlled variable as through a spring, it is likely that
the whole system will become unstable at the critical point, or begin to
behave as a flip-flop as the error fluctuates around the critical point.
This should give us a better idea of the kinds of relationships we should
look for in real behavior. It will certainly help us distinguish the sort
of "giving up" behavior that results from this ertror-effort relationship
from other behavior that might also fall into a category indicated by those
words.

I'm fooling around with Bill Williams' source code for the Ass and will
communicate about the results if there are any.

Best,

Bill P.

17

[From Bruce Nevin (2002.05.23 23:28 EDT)]

Bill Powers (2002.05.23.1951 MDT)--
The only reason these examples can "be said" to show the "giving up"

phenomenon is that the meaning of the term changes, and we use a very
liberal interpretation of what we see.

The point of bringing this up is that all we have initially is what we see. In the models we know what is going on. When we observe something happening in nature or in the laboratory we don't. The models, as always, are instructive as to what might be going on behind the appearances in nature or in the laboratory.

the observation was that as a rat approached a food dish,
its efforts to get closer increased, as measured through a tether
restraining the rat.

We're assuming that the rat is controlling the quantity 'distance from food', that error is inversely proportional to distance (or directly proportional to proximity to the food), and that this is all that is going on.

If I'm lifting a heavy rock, and I've almost got it to the top, I exert tremendous effort, as much as I can, to put it over the top. Why is that? My rationalization of this is that I make extra effort because I'm almost there and also because if I don't get there on this try the chances of getting there on a second try aren't likely to be better. And I want to get on with whatever it was that I was lifting that rock up there for, or whatever I was going to do next.

Suppose this is a sequence that the rat is controlling:

1. Move mouth close to food.
2. Eat.

Then error in the sequence control system is due to Step 1 not being completed. It seems to me, subjectively, that if a step in a sequence is almost completed (and the sequel is desired rather than averted) the gain increases for completing that step and proceeding to the next.

Some experimentation and modelling of sequence control might determine whether this is true, and could provide more insight into the rat experiment.

         /Bruce

···

At 08:30 PM 5/23/2002 -0600, Bill Powers wrote:

18

[From Rick Marken (2002.05.23.2230)]

Bill Powers (2002.05.23.1951 MDT)

Perhaps someone with access to a large University library can find the
paper describing the effect as I first heard of it. The authors will (I
believe) be Dollard and Miller, or perhaps one of them alone, although the
first author could have been a graduate student. The subject is "approach
gradients," and the observation was that as a rat approached a food dish,
its efforts to get closer increased, as measured through a tether
restraining the rat. It is this effect we have to explain, although if it
is seen only in rats and only in that experiment, finding an explanation
may not be very important. Look up _Dollard Miller approach gradient_ in
Google and you'll see what you get into.

I do remember that study, vaguely (I think I saw it referred to in an
undergraduate course on "Motivation"). I can cull up an image of it from
memory. The image is a picture of a rat wearing a harness that is connected to
a force measuring device. I don't have a memory of the force distance plot,
if such a plot was presented. I looked on Google and found articles that refer
to Miller and Dollard's approach-avoidance "theories" and their notion of an
approach gradient. But no reference to that particular "rat harness" study.
Finding this study may require some effort, which may increase as I get closer
to it. If anyone can find it I bet Bruce Abbott can!

Best

Rick

···

--
Richard S. Marken
MindReadings.com
marken@mindreadings.com
310 474-0313

19

snipped

I'm concerned about that experiment, because it seems to contradict PCT:

as

any error gets smaller, the effort used to correct it ought to get

smaller,

or its rate of increase ought to be smaller. Instead, the proximity of the
food dish seems to serve as a stimulant to the approach efforts, which is
of course the S-R interpretation Dollard and Miller put on that sort of
thing. Whoever did that experiment should have sensed something wrong,
because if the effort really kept increasing all the way to the food dish,
the rat should have gone shooting by it instead of stopping to eat. But we
still have to worry about the behavior far from the food dish. And at

least

as I do things, I can't explain it just by cooking up some imaginary
higher-level systems defined so they will do just what we observe. If
that's all that's happening, there are too many scenarios that would give
the same appearance, and none of them is compelling for that reason.

A key feature of this phenomenon was that there was a _gradient_ of

effort,

so the effort was a smooth function of distance (in the wrong direction

for

PCT). That is hard to explain in terms of a logical system that is
reasoning symbolically about what's happening. Whatever is creating the
appearance of a change in gain is doing it in an analog fashion.

As someone who spends time in the woods following trails with choice points,
on the way to a goal, I think I speed up as I gain confidence that I'm
making the right choices. That's obviously not a smooth gradient, but I
doubt if the measurement of the rats was that fine.
David Wolsk

···

----- Original Message -----
From: "Bill Powers" <powers_w@FRONTIER.NET>
To: <CSGNET@listserv.uiuc.edu>
Sent: Thursday, May 23, 2002 7:30 PM
Subject: Re: UEC/NEC summary