Reinforcement Test

CSGnet Archive CSG2000
1

[From Rick Marken (2000.11.25.1140)]

Bruce Abbott (2000.11.25.1125 EST) --

I see that you are very busy with Fred's Home Heating System.
But could you take just a second to explain what's wrong (if
anything) with my proposed test of reinforcement theory. I
really want to develop a net demo along these lines and it
would be nice to have your help.

So let's go back to the beginning.

You [Bruce Abbott (2000.11.22.1525 EST)] said:

If actions must vary to reproduce specific consequences, then
there must have been some way the organism learned what actions
to vary and in what direction. The behaviorist view would be
that such appropriate behaviors occur now because in the past
such behaviors have been reinforced under these conditions.

That seemed like a pretty clear statement of the reinforcement
theory view: "appropriate behaviors" (actions that reproduce a
specific consequence) occur now because in the past such behaviors
have been reinforced. So an obvious test of reinforcement theory
seemed to be to design a situation where a subject learns to
produce a specific consequence (one that is followed by
reinforcement) using certain behaviors (actions). Then we test
to see if the subject continues to produce this consequence using
actions that are _not_ the ones that produce the consequence
when it was being reinforced. If the subject _does not_ produce
these behaviors when producing them is made necessary in order to
produce the consequence, it is evidence _for_ reinforcement theory.
If, however, the subject _does_ produce these behaviors it is
evidence against reinforcement theory.

This reasoning was the basis for my proposed experiment:

The organism learns to reproduce some specific consequence under
conditions where a specific disturbance requires specific action
variations in order to reproduce it. Once the organism has learned
the action variations that reproduce the consequence, a new
disturbance is introduced so that _different_ action variations
are required to reproduce the consequence.

Is there anything wrong with this experimental test? If so, what
is it and what is a better test of reinforcement theory?

Best

Rick

···

--

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: marken@mindreadings.com
mindreadings.com

2

[From Bruce Nevin (2000.11.26.0317 EDT)]

Question for Bruce Abbott: Are behaviors, operants, etc. always discrete events?

  Bruce Nevin

3

[From Bruce Abbott (2000.11.26.0825 EST)]

Bruce Nevin (2000.11.26.0317 EDT)]

Question for Bruce Abbott: Are behaviors, operants, etc. always discrete
events?

Excellent question!

The organism must learn relationships between what it does and what it
experiences. To learn such relationships, its outputs to the environment
must vary, following which there may be fairly systematic changes in its
experience (i.e., there must be a relationship to learn). This sort of
learning is usually studied by arranging for some particular discrete act to
produce some particular discrete change in the environment that the organism
can perceive and relate to the act (e.g., a rat presses a lever and a food
pellet appears). However, one could arrange a situation in which a
continuous variation in some output (force exerted on the lever, for
example) produced a continuous variation in some experience (e.g., intensity
of pleasurable brain stimulation). If the rat consistently produced a
particular intensity, the explanation offered would be that holding the
lever down with the force required to produce this intensity was more
strongly reinforced than holding it down with forces that produced other
intensities. The data would be taken to indicate the intensity of brain
stimuluation that was most reinforcing for this particular animal, and the
experimental arrangement might be used to track changes in that intensity
(if any) over time.

The relationship between behavior and experience is not the only
relationship the organism learns. It can also learn relationships between
various inputs from the environment, and between input conditions and
behavior-input relationships (e.g., pressing the lever yields food when a
cue light is on and not when the cue light is off).

Bruce A.

4

[From Bruce Nevin (2000.11.26.2248 EDT)]

Bruce Abbott (2000.11.26.0825 EST) --

>Bruce Nevin (2000.11.26.0317 EDT)]
>
>Question for Bruce Abbott: Are behaviors, operants, etc. always discrete
>events?

Excellent question!

The organism must learn relationships between what it does and what it
experiences. [...] This sort of learning is usually studied by arranging for some particular discrete act to produce some particular discrete change in the environment that the organism can perceive and relate to the act [...]. However, one could arrange a situation in which a continuous variation in some output [...] produced a continuous variation in some experience [...].

I gather from this (and other references) that the norm in EAB research is to assume that behaviors, operants, etc. are always discrete events, and that it is not the norm to study continuous co-variation of e.g. intensity of lever pressure and intensity of "pleasurable brain stimulation" (somehow measured in a way that is relevant to the animal's experience of it), though you say that this would be possible.

I am guessing that countable "behaviors" are seen to accomplish EAB aims quite well and that that EAB researchers therefore have little or no interest in or reason to get involved in the messiness of continuous co-variation of a behavioral output and an operant. It may even be the case that no EAB study of continuous co-variation of some behavioral output and some desired sensory input has in fact ever been reported.

Why do I suppose the data of continuous co-variation would be too messy for EAB research? Aside from the obvious complexity, there's variation of the reference signal for the perception of the operant. If the intensity of "pleasurable brain stimulation" can be measured in a way that is relevant to the animal's experience of it, the experimental arrangement you sketched could provide a measure of the subject's reference level for that sensory input: by varying pressure on the lever the rat would adjust the "pleasurable brain stimulation" to its reference level. Of course, that reference level would be subject to change, depending on the error signals from which it originates and the reference input function that combines them, adding to the messiness of co-variation data. Lacking a model that includes reference signals and continuous negative feedback control, how is a researcher to make any sense of such data? No, to get any sensible EAB results you have to treat lever presses and "pleasurable brain stimulations" as discrete events. Right? Isn't this born out by the absence of such studies from the EAB literature?

I would be delighted to learn that my guess about this wildly off the mark.

         Bruce Nevin

···

At 08:26 AM 11/26/2000 -0500, Abbott, Bruce wrote:

5

[From Bruce Abbott (2000.11.27.1720 EST)]

Bruce Nevin (2000.11.26.2248 EDT) --

Bruce Abbott (2000.11.26.0825 EST)

>Bruce Nevin (2000.11.26.0317 EDT)]
>
>Question for Bruce Abbott: Are behaviors, operants, etc. always discrete
>events?

Excellent question!

The organism must learn relationships between what it does and what it
experiences. [...] This sort of learning is usually studied by arranging
for some particular discrete act to produce some particular discrete
change in the environment that the organism can perceive and relate to the
act [...]. However, one could arrange a situation in which a continuous
variation in some output [...] produced a continuous variation in some
experience [...].

I gather from this (and other references) that the norm in EAB research is
to assume that behaviors, operants, etc. are always discrete events, and
that it is not the norm to study continuous co-variation of e.g. intensity
of lever pressure and intensity of "pleasurable brain stimulation" (somehow
measured in a way that is relevant to the animal's experience of it),
though you say that this would be possible.

Point of clarification: It would not be the norm in EAB research to _assume_
that behaviors are discrete events, but it _is_ the norm to study behaviors
whose occurrence is defined by discrete events -- the lever-press or
key-peck being examples in which the behavior is defined as occurring when
the criterion of switch-closure occurs. In these cases the dependent
variable under study is the frequency with which the behavior, so defined,
occurs.

I am guessing that countable "behaviors" are seen to accomplish EAB aims
quite well and that that EAB researchers therefore have little or no
interest in or reason to get involved in the messiness of continuous
co-variation of a behavioral output and an operant. It may even be the case
that no EAB study of continuous co-variation of some behavioral output and
some desired sensory input has in fact ever been reported.

Certainly most of the work has focused on developing an understanding of the
factors that affect in predictable ways the frequency of such countable
acts. Other properties of those acts, such as duration or force, have also
been studied, but less well.

Why do I suppose the data of continuous co-variation would be too messy for
EAB research? Aside from the obvious complexity, there's variation of the
reference signal for the perception of the operant. If the intensity of
"pleasurable brain stimulation" can be measured in a way that is relevant
to the animal's experience of it, the experimental arrangement you sketched
could provide a measure of the subject's reference level for that sensory
input: by varying pressure on the lever the rat would adjust the
"pleasurable brain stimulation" to its reference level. Of course, that
reference level would be subject to change, depending on the error signals
from which it originates and the reference input function that combines
them, adding to the messiness of co-variation data. Lacking a model that
includes reference signals and continuous negative feedback control, how is
a researcher to make any sense of such data? No, to get any sensible EAB
results you have to treat lever presses and "pleasurable brain
stimulations" as discrete events. Right? Isn't this born out by the absence
of such studies from the EAB literature?

I don't think you need to treat lever presses and brain stimulation as
discrete events, but you do need to assume that the subject can perceive
changes both in its own behavior and in its inputs and detect a relationship
between them. EABers see no problem with talking about a pigeon perceiving,
for example, a relationship between its _rate_ of responding and _rate_ of
reinforcement, both of which are continuous variables, and talking about
higher rates of responding being reinforced by resulting increased rates of
reinforcement (on ratio schedules). The continuous case I described would
be viewed as another example of the same, therefore yielding nothing that
can't be seen in the frequency variations of discrete actions and input
changes. I suspect that the reason for sticking to discrete lever-presses
etc. is that levers and response keys are cheap and reliable, the apparatus
is already equipped that way, and there is no perceived need to change among
those in the EAB community.

Bruce A.

6

[From Bruce Gregory (2000.1127.1747)]

Bruce Abbott (2000.11.27.1720 EST)

I suspect that the reason for sticking to discrete lever-presses
etc. is that levers and response keys are cheap and reliable, the apparatus
is already equipped that way, and there is no perceived need to change among
those in the EAB community.

Not as long as they continue to receive reinforcement while using that kind
of apparatus.

BG

7

[From Rick Marken (2000.11.27.1515)]

Bruce Abbott (2000.11.27.1200 EST)--
Bruce Abbott (2000.11.27.1720 EST)--

Now that you're back, perhaps you could take a moment to
reply to [Rick Marken (2000.11.25.1140)]. I would sure
appreciate your comments on my experimental proposal. It
would also be nice to see the simulation that Bill Powers
(2000.11.26.01255 MST) asked for just before you left.

Thanks

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
MindReadings.com mailto: marken@mindreadings.com
www.mindreadings.com

8

[From Bruce Abbott (2000.11.27.1940 EST)]

Bruce Gregory (2000.1127.1747) --

Bruce Abbott (2000.11.27.1720 EST)

I suspect that the reason for sticking to discrete lever-presses
etc. is that levers and response keys are cheap and reliable, the apparatus
is already equipped that way, and there is no perceived need to change among
those in the EAB community.

Not as long as they continue to receive reinforcement while using that kind
of apparatus.

I used to have a cartoon on my door in which two lab rats are talking and
one says to the other "I've got him trained to give me a pellet whenever I
press this bar." You're right -- the experimenter's behavior is just as
subject to shaping via reinforcement as the rat's. But if you are implying
that the experimenter arranges the situation so that it always produces the
data she wants, that's just plain wrong. There are plenty of experiments in
which what happened was not at all what the researcher who designed the
study was expecting or wanted. A clear example is autoshaping, in which
pigeons began to peck at an illuminated key even though there was no
apparent reward for doing so. This of course proved to be a challenge to
reinforcement theory and led to a reformulation of thinking about what
processes are at work in such situations. More recently, John Staddon has
demonstrated situations in which subjects appear to be driven to respond in
ways that are not in their best interests -- that actually increase the time
to food delivery. Again this has stimulated new thinking about what
underlying processes are at work to produce this apparently nonadaptive
behavior. To me, this doesn't sound anything like researchers arranging
their experiments so that they produce data that always confirm their prior
view. Maybe it does to you; if so, then I don't understand your reasoning.

As for "just-so stories," that label applies to situations in which
explanations are invented post hoc to fit the data, for which no evidence
exists other than the data they are invented to explain. That's a problem
for evolutionary theory because nobody was there to watch, say, a small
dinosaur's physical characteristics gradually change as a result of
variation and natural selection until it could fly. We can only say "well,
birds exist, and here's a reasonable scenario as to how they might evolve.
Until you begin to find fossil or other evidence to support this reasoning,
it's only a possibility, a "just-so story." That is not the case with
operant conditioning research, in which the changes in behavior can be
observed as they happen, and related to variables whose effects can be
assessed via experimental manipulation. Clearly this excludes the
explanations developed to account for the data from the category of "just-so
stories" -- unless you have some other definition in mind.

Bruce A.

9

[From Bruce Nevin (2000.11.27.1804 EDT)]

Bruce Abbott (2000.11.27.1720 EST)--

···

At 05:22 PM 11/27/2000 -0500, Abbott, Bruce wrote:

ABers see no problem with talking about a pigeon perceiving,
for example, a relationship between its _rate_ of responding and _rate_ of
reinforcement, both of which are continuous variables, and talking about
higher rates of responding being reinforced by resulting increased rates of
reinforcement (on ratio schedules).

These rates are continuous variables controlled by an observer with the aid of metering equipment. Is there evidence that they are perceptions controlled by the rat, pigeon, or human subject?

         Bruce Nevin

10

[From Bruce Abbott (2000.11.27.2155 EST)]

Bruce Nevin (2000.11.27.1804 EDT) --

Bruce Abbott (2000.11.27.1720 EST)

EABers see no problem with talking about a pigeon perceiving,
for example, a relationship between its _rate_ of responding and _rate_ of
reinforcement, both of which are continuous variables, and talking about
higher rates of responding being reinforced by resulting increased rates of
reinforcement (on ratio schedules).

These rates are continuous variables controlled by an observer with the aid
of metering equipment.

No, they're not. These rates are continuous variables controlled by the pigeon.

Is there evidence that they are perceptions
controlled by the rat, pigeon, or human subject?

Without looking at a timepiece, tap the table with your finger at roughly
two times per second. No problem? Good, how did you do it? Can you
distinguish tapping quickly from tapping slowly? You can? How do you
manage to do it without measuring equipment?

Yes, there is plenty of evidence that such rates can be perceived and
controlled by rat, pigeon, or person. Of course that does not mean that
they are always controlled.

Bruce A.

11

[From Bruce Abbott (2000.11.28.1420 EST)]

Bruce Nevin (2000.11.28.1336 EDT) --

Bruce Abbott (2000.11.27.2155 EST)

>Is there evidence that they [rates of tapping, etc.] are perceptions
>controlled by the rat, pigeon, or human subject?

Without looking at a timepiece, tap the table with your finger at roughly
two times per second. No problem? Good, how did you do it? Can you
distinguish tapping quickly from tapping slowly? You can? How do you
manage to do it without measuring equipment?

Sure makes my question sound absurd, naive, and downright silly, doesn't it.

I thought it was an excellent question; I didn't intend my reply to make it
sound silly, and I'm sorry if it did.

If the rat is controlling a rate, then the interval between presses is
regular and the appearance of regular periodicity is not a product of
averaging. Is that the case?

How about the case where higher rates of responding produce higher average
rates of pellet delivery? That possibility assumes that the rat can
perceive changes in average rate that correlate with changes in its response
rate.

As for rate of responding as a controlled variable, rats can successfully
adjust their rates of responding to meet some rate criterion, as in
differential reinforcement of high rate (DRH) or differential reinforcement
of low rate (DRL) schedules, or in schedules that require that the rate fall
into some window.

I can control a relationship of synchrony in which a reference of "less
delay" or "more delay" is involved but no "rate" reference is involved. One
term of the relationship can be a memory of just-prior performance. In this
model, rate is a notion introduced by the observer.

Is the rat controlling the rate or the interval between events? Hard to
distinguish the two cases: one variable is just the inverse of the other.

This issue has been at the core of a continuing debate in EAB referred to as
the "molar" versus "molecular" views. The molar view says that the subjects
are able to integrate events over time, the other says that the are
sensitive only to relatively immediate experiences, such as the interval
between two events.

I once carried out a study in which rats had to choose between two
conditions that were identical except for a difference in the average length
of a series of variable-length foot-shocks. The rats had little trouble
distinguishing which condition offered the shocks of shorter average
duration, down to differences as small as 0.1 seconds. In another study,
rats had to choose between two variable-time shock schedules differing only
in the average frequency of shock (or if you prefer, the average time
betweens shocks). They had no trouble choosing the condition having the
lower shock density.

I believe that such performances require the ability to integrate over time.

Bruce A.

12

[From Bruce Gregory (2000.1128.1630)]

Bruce Abbott (2000.11.27.1940 EST)

To me, this doesn't sound anything like researchers arranging
their experiments so that they produce data that always confirm their prior
view.

Nature is rarely so accommodating. The main requirement is that the
experiment generate data that can be subject to canonical statistical
analysis. I assume this is what the experimenters find reinforcing.

Clearly this excludes the
explanations developed to account for the data from the category of "just-so
stories" -- unless you have some other definition in mind.

Exactly why is it incorrect to say that temperature changes reinforce the
thermostat's behavior?

BG

13

[From Bruce Nevin (2000.11.28.1336 EDT)]

Bruce Abbott (2000.11.27.2155 EST)--

>Is there evidence that they [rates of tapping, etc.] are perceptions
>controlled by the rat, pigeon, or human subject?

Without looking at a timepiece, tap the table with your finger at roughly
two times per second. No problem? Good, how did you do it? Can you
distinguish tapping quickly from tapping slowly? You can? How do you
manage to do it without measuring equipment?

Sure makes my question sound absurd, naive, and downright silly, doesn't it.

If the rat is controlling a rate, then the interval between presses is regular and the appearance of regular periodicity is not a product of averaging. Is that the case?

I can control a relationship of synchrony in which a reference of "less delay" or "more delay" is involved but no "rate" reference is involved. One term of the relationship can be a memory of just-prior performance. In this model, rate is a notion introduced by the observer.

         Bruce Nevin

···

At 09:55 PM 11/27/2000 -0500, Abbott, Bruce wrote:

14

[From Bruce Abbott (2000.11.28.1715 EST)]

Bruce Gregory (2000.1128.1630) --

Bruce Abbott (2000.11.27.1940 EST)

To me, this doesn't sound anything like researchers arranging
their experiments so that they produce data that always confirm their prior
view.

Nature is rarely so accommodating. The main requirement is that the
experiment generate data that can be subject to canonical statistical
analysis. I assume this is what the experimenters find reinforcing.

What do you mean by "cononical statistical analysis"?

Clearly this excludes the
explanations developed to account for the data from the category of "just-so
stories" -- unless you have some other definition in mind.

Exactly why is it incorrect to say that temperature changes reinforce the
thermostat's behavior?

Whether the thermostat will continue to actuate the furnace when the
temperature falls below reference does not depend on the past success or
failure of this action in raising the temperature toward the reference level.

Bruce A.

15

[From Bruce Nevin (2000.11.28.2102 EDT)]

Bruce Abbott (2000.11.28.1420 EST)--

I thought it was an excellent question; I didn't intend my reply to make it
sound silly, and I'm sorry if it did.

My apology for being too touchy. It causes as much trouble to take offense as it does to give it.

What perhaps you mean by "rate" is events per minute (or some other time unit), where the momentary speed (interval between two events) can vary considerably. The "rate" in this sense is inherently an average over many inter-event intervals. If a rat is controlling a rate in this sense, it must be controlling a count of events in relation to a perception of units of time and calculating an average (integrating over time). Do you have in mind some simpler way that could be accomplished?

What I mean by "rate" is a constant momentary speed (interval between pairs of events). No averaging is required. This is what I intend by the term in the following responses.

>If the rat is controlling a rate, then the interval between presses is
>regular and the appearance of regular periodicity is not a product of
>averaging. Is that the case?

How about the case where higher rates of responding produce higher average
rates of pellet delivery? That possibility assumes that the rat can
perceive changes in average rate that correlate with changes in its response
rate.

To do it faster does not necessarily require the rat to perceive a rate. If the rate of pellet delivery increases the quantity of food available increases, so there is no need for the rat to perceive the rate of pellet delivery. Just because these are the things measured does not mean that they are relevant -- like Nasrudin looking for his house key under the streetlamp, where there's light to see by, rather than in the alleyway where he lost it.

As for rate of responding as a controlled variable, rats can successfully
adjust their rates of responding to meet some rate criterion, as in
differential reinforcement of high rate (DRH) or differential reinforcement
of low rate (DRL) schedules, or in schedules that require that the rate fall
into some window.

Again, doing it faster (in a range of rates called H) or slower (L) or in some other range ("window") does not demonstrate that the rat is perceiving a rate and controlling the rate perception, as you acknowledge:

Is the rat controlling the rate or the interval between events? Hard to
distinguish the two cases: one variable is just the inverse of the other.

It looks to me as though Ockam's razor shaves the rate and leaves the interval, since the former is built upon the latter. A steady rate of discrete events is necessarily a succession of like intervals between events; an interval between events bears no necessary relation to a rate.

This issue has been at the core of a continuing debate in EAB referred to as
the "molar" versus "molecular" views. The molar view says that the subjects
are able to integrate events over time, the other says that the are
sensitive only to relatively immediate experiences, such as the interval
between two events.

I once carried out a study in which rats had to choose between two
conditions that were identical except for a difference in the average length
of a series of variable-length foot-shocks. The rats had little trouble
distinguishing which condition offered the shocks of shorter average
duration, down to differences as small as 0.1 seconds. In another study,
rats had to choose between two variable-time shock schedules differing only
in the average frequency of shock (or if you prefer, the average time
betweens shocks). They had no trouble choosing the condition having the
lower shock density.

The rats are getting shocked less. The rate is how you are measuring it, but you have no idea how they perceive that.

I believe that such performances require the ability to integrate over time.

They might be feeling less shocked because they integrate an average rate over time, or because there is less accumulated error in control of some metabolic state or states. The question was formed by the debate, which was formed by theoretic presuppositions, and the results are interpreted as answers to the question. But what is the controlled variable? If you have played the coin game, or if you have witnessed expert practitioners of experimental psychology postulating what is going on in the rubber band demos, you know that presuppositions about the controlled variable are not very likely to be correct, and each guess has to be tested by applying disturbances and seeing if they are resisted.

But all of this is mooted by your answer, apparently "no", to my first question:

>If the rat is controlling a rate, then the interval between presses is
>regular and the appearance of regular periodicity is not a product of
>averaging. Is that the case?

In all of your examples, a "rate" is the product of averaging a range of varying intervals per minute (or some other unit) over time.

There is an easy way to test whether your experiment is measuring control of a rate (constant interval between events) or something else. If the rat is controlling a rate, when a disturbance lengthens an interval or two and then is removed, the rat will simply resume at the controlled rate. If in order to accomplish the desired result the rat has to diminish successive intervals to compensate for the delay and then resume at something like the average pace, the rat is not controlling a rate (constant interval between events).

Bruce A.

         Bruce Nevin

···

At 02:21 PM 11/28/2000 -0500, Abbott, Bruce wrote:

16

[From Bruce Nevin (2000.11.28.2104 EST)]

Bruce Abbott (2000.11.28.1715 EST)--

>Exactly why is it incorrect to say that temperature changes reinforce the
>thermostat's behavior?

Whether the thermostat will continue to actuate the furnace when the
temperature falls below reference does not depend on the past success or
failure of this action in raising the temperature toward the reference level.

More succinctly: reinforcement has to do with learning, and the thermostat doesn't learn.

A PCT critique of EAB will be interpreted as a critique of a theory of learning. A critique of the underlying theory of behavior passes unnoticed, because behavior itself is taken for granted, it is the learning of new "behaviors" that is in focus.

         Bruce Nevin

···

At 05:16 PM 11/28/2000 -0500, Abbott, Bruce wrote:

17

[From Bill Powers (2000.11.28.0157 MST)]

Bruce Nevin (2000.11.28.2102 EDT)--

Good post, but:

an interval between events bears no necessary relation to a rate.

Interval = 1/rate. Two events that occur 1 second apart are occurring at 1
event per second. If the next event occurs 0.5 seconds later, the rate has
changed to 2 events per second.

Best,

Bill P.

18

[From Bruce Nevin (2000.11.29.1119 EDT)]

Bill Powers (2000.11.28.0157 MST)--

> an interval between events bears no necessary relation to a rate.

Interval = 1/rate.

Thanks, Bill, I stand corrected. Technically, a single pair of events occurs at a rate specified by the interval between them. This is the basis for what I mean by a rate as a "constant momentary speed (interval between successive pairs of events)". In any normally meaningful sense -- in particular in the EAB sense of integration over time -- "rate" refers to an extended series of events.

The essential thing is that an interval is a more primitive entity than a rate (as normally understood): in order to be able to perceive and control an event rate you must be able to perceive and control intervals between events, but in order to perceive and control an interval you do not need to be able to perceive a rate--unless you collapse the two notions, and if you do that you exclude the EAB question of integration over time.

         Bruce Nevin

···

At 02:01 AM 11/29/2000 -0700, Bill Powers wrote:

19

[From Bruce Abbott (2000.11.28.1750 EST)]

Bruce Nevin (2000.11.28.2102 EDT)]

To do it faster does not necessarily require the rat to perceive a rate. If
the rate of pellet delivery increases the quantity of food available
increases, so there is no need for the rat to perceive the rate of pellet
delivery. Just because these are the things measured does not mean that
they are relevant -- like Nasrudin looking for his house key under the
streetlamp, where there's light to see by, rather than in the alleyway
where he lost it.

There's a fair amount of research attempting to determine what the rats (or
pigeons) are actually perceiving and acting on in these situations, but
results have been less clear-cut than one would hope.

Is the rat controlling the rate or the interval between events? Hard to
distinguish the two cases: one variable is just the inverse of the other.

It looks to me as though Ockam's razor shaves the rate and leaves the
interval, since the former is built upon the latter. A steady rate of
discrete events is necessarily a succession of like intervals between
events; an interval between events bears no necessary relation to a rate.

To average rate.

This issue has been at the core of a continuing debate in EAB referred to as
the "molar" versus "molecular" views. The molar view says that the subjects
are able to integrate events over time, the other says that the are
sensitive only to relatively immediate experiences, such as the interval
between two events.

I once carried out a study in which rats had to choose between two
conditions that were identical except for a difference in the average length
of a series of variable-length foot-shocks. The rats had little trouble
distinguishing which condition offered the shocks of shorter average
duration, down to differences as small as 0.1 seconds. In another study,
rats had to choose between two variable-time shock schedules differing only
in the average frequency of shock (or if you prefer, the average time
betweens shocks). They had no trouble choosing the condition having the
lower shock density.

The rats are getting shocked less. The rate is how you are measuring it,
but you have no idea how they perceive that.

No, but it is significant that they can tell that they are getting shocked less.

I believe that such performances require the ability to integrate over time.

They might be feeling less shocked because they integrate an average rate
over time, or because there is less accumulated error in control of some
metabolic state or states.

Accumulated error is error integrated over time.

The question was formed by the debate, which was
formed by theoretic presuppositions, and the results are interpreted as
answers to the question.

Call me dense, but I have no idea what that means.

There is an easy way to test whether your experiment is measuring control
of a rate (constant interval between events) or something else. If the rat
is controlling a rate, when a disturbance lengthens an interval or two and
then is removed, the rat will simply resume at the controlled rate. If in
order to accomplish the desired result the rat has to diminish successive
intervals to compensate for the delay and then resume at something like the
average pace, the rat is not controlling a rate (constant interval between
events).

Wouldn't that depend on how long an interval the average rate is being
integrated over? And the subject's ability to discriminate the short-run
changes?

Bruce A.

20

[From Bruce Nevin (2000.11.29.1933 EDT)]

Bruce Abbott (2000.11.28.1750 EST)--

>It looks to me as though Ockam's razor shaves the rate and leaves the
>interval, since the former is built upon the latter. A steady rate of
>discrete events is necessarily a succession of like intervals between
>events; an interval between events bears no necessary relation to a rate.

To average rate.

The notion that a rat is computing an average rate is extremely implausible to me. But there is an easy way to test whether your experiment is measuring control of a rate (constant interval between events) or an average rate. If the rat is controlling a rate, when a disturbance lengthens an interval or two and then is removed, the rat will simply resume at the controlled rate. If in order to accomplish the desired result the rat has to diminish successive intervals (press faster) to compensate for the delay and then revert to something like the average pace, the rat is not controlling a rate (constant interval between events) and might be controlling an average rate. Which do you see?

Wouldn't that depend on how long an interval the average rate is being
integrated over? And the subject's ability to discriminate the short-run
changes?

Are you saying that with sufficiently long an interval the degree of slop in your averaging would be so great that there would be no need for the rat to catch up after being impeded? I think we have to limit your experiment to one where you might hope to demonstrate *something*.

>The rats are getting shocked less. The rate is how you are measuring it,
>but you have no idea how they perceive that.

No, but it is significant that they can tell that they are getting shocked less.

If my muscle aches less I can tell I have got punched less. That does not mean I am perceiving or controlling an average rate of punches.

>>I believe that such performances require the ability to integrate over time.
>
>They might be feeling less shocked because they integrate an average rate
>over time, or because there is less accumulated error in control of some
>metabolic state or states.

Accumulated error is error integrated over time.

The ache in my muscle is an equally sophisticated mechanism for integrating error over time. In neither case is there a control process calculating the integration of error over time. But in any case, now we are talking about integration of error, not integration of an average rate. This has no bearing on your claim that the rat is perceiving and controlling average rates and then controlling a relationship between one average rate and another.

>The question was formed by the debate, which was
>formed by theoretic presuppositions, and the results are interpreted as
>answers to the question.

[...] I have no idea what that means.

Sorry, I'll unpack it. It has four parts.

1. The debate: "This issue has been at the core of a continuing debate in EAB referred to as the "molar" versus "molecular" views. The molar view says that the subjects are able to integrate events over time, the other says that the are sensitive only to relatively immediate experiences, such as the interval between two events." (your 2000.11.28.1420 EST)

2. The question: the experiments were designed to answer the "molar or molecular" question posed in this debate.

3. The debate itself was in accord with presuppositions about what behavior is. In particular, the debate was uninformed by any notion of controlled perceptions, or any recognition that behavior is significant to an organism (and therefore to a psychologist) only insofar as it helps to make the organism's perceptions be as it wants them to be. An investigator who is so informed would never consider administering shocks as a sensible research technique. It overwhelms control and very likely induces reorganization.

4. The results are interpreted as answering the question. Alternative interpretations are not considered, and indeed cannot arise within the unquestioned conceptual presuppositions about what behavior is.

What is the controlled variable? If you have played the coin game, or if you have witnessed expert practitioners of experimental psychology postulating what is going on in the rubber band demos, you know that presuppositions about the controlled variable are not very likely to be correct, and each guess has to be tested by applying disturbances to the putative controlled variable and seeing if they are resisted.

I still think that your answer to my first question was "no": Is the interval between actions regular? In the manner in which when we say that a motor is turning at 3600 rpm we expect that to mean that it is turning 360 times every 6 seconds, 60 times every second, throughout the time, perhaps several minutes, that it is running at that speed. Or does the interval between presses vary irregularly, so that a statement of a "rate" is possible only by averaging over an extended period of time? As if we were to rev the motor between idling and 5000 rpm several times over the course of a minute and then calculate that it was running at a "rate" of 3600 rpm.

If the rat is controlling a rate, as in 3600 rpm as we normally understand that, then when a disturbance slows the rat down the rat will counter the disturbance merely by resuming at the controlled rate, much as when a load slows a motor from 3600 rpm and then is removed the motor simply reverts to 3600 rpm (other things being equal). If the rat is controlling an average rate, then the rat will know that it must speed up in order to counter the disturbance and get back on track to meet its quota for the averaging period, and when it is confident of meeting its quota it will reduce the pace to something near the average rate again, rather like a rally driver.

The third possibility, of course, is that the rat is controlling neither. This opens a prospect of testing to find out what the rat in fact is controlling. That may be uncomfortably open-ended. But that's life. Something that we're only beginning to nibble at the near edge.

         Bruce

···

At 05:53 PM 11/29/2000 -0500, Abbott, Bruce wrote: