[From Bill Powers (951220.0930 MST)]
Ed Ford Dec. 19, 1995 --
I liked your post about the meeting and how you develop new ideas with
your associates. Your words are reassuring.
I also agree with you about criticism, if you mean criticizing people
rather than ideas or presentations. In many of our interchanges on the
net, I tried to make it clear that I was concerned mainly with your
words having unintended meanings which could lead people to think that
this was simply another reward-and-punishment scheme. This did not mean
I think it IS such a scheme; only that the way you were representing it
left that interpretation wide open. The cure for this is to change the
way you describe the program.
In short, for those on my net to make intelligent comments, they
must have worked in the trenches.
So does this mean that you will accept postings only from teachers,
administrators, etc. in school or correctional systems?
Those who tend to shoot us down are not the top flight, secure,
competent teachers. They tend to be the highly insecure.
Isn't this arguing _ad hominem_ -- that is, questioning the motives of
the people instead of considering the substance of what they say? It
might be advantageous to extend your way of interacting with colleagues
to those who disagree with your program. Even an insecure person might
have something worthwhile to contribute.
Criticism kills initiative, whether in adults or children. And if
there is no suggestion offered for a better way to improve what is
being criticized, it leaves the person being criticized frustrated.
It's perceived as a disturbance, rather than as an alternative way
to close or reduce and error. I think my total process, which
eliminates criticism, is better.
I'm sure it is.
···
-----------------------------------------------------------------------
Chris Cherpas (951219.1212 PT) -
Now, you want me to prove with experimental evidence that food is
indeed delivered after x number of bar presses, not just logically
or appealling to common sense? Is this one of those zen trick
questions? I suppose we could experimentally prove this by
disconnecting the wires attached to the control mechanisms which
deliver reinforcement, change the computer program from FR to EXT,
etc.; of course, you don't need a rat to test whether the
contingency is FR -- the experimenter can directly test out the
equipment.
No tricks. I was trying to bring out the fact that investigating the
effect of pressing on reinforcements involves a straightfoward way of
establishing cause and effect, because you can experiment directly with
the apparatus, take it apart, check the wiring, etc.. There can be no
doubt that the pattern of reinforcements is completely dependent on the
pattern of behavior, via the apparatus. As you say, you don't even need
the rat's presence to determine how the apparatus is set up.
After the first pellet delivery or so, the rate of bar pressing is
likely to skyrocket. In terms of your question, it means the
behavior will increase in probability -- i.e., relative to the
probability prior to receiving the first pellet.
"Probability" is not measurable directly; you can measure only frequency
of pressing. Whether this frequency is the result of a stochastic
process (making probability an appropriate term) or a regular process is
a question of theory. Notice that you are not talking, now, the way you
did in talking about the apparatus.
Most everything after that is likely to be "maintenance," which you
can prove by changing the contingency -- for example, to EXT.
I would like to suggest a distinction between acquisition and
maintenance. During acquisition, the lever gets pressed because it is
among many physical objects being contacted by the animal. Acquisition
involves gradually focusing efforts on the right object or objects. This
creates the same appearance, in terms of contact closures, that we would
see if the animal were pressing only on the lever, but doing so at a
gradually increasing rate. The measured rate of contact closures can't
distinguish between acquisition and maintenance, as I said recently to
Bruce.
I think in either case, the experimental design would be a version
of A-B-A: take away what looks to have caused the change,
reintroduce it, take it away, reintroduce it, etc. If not
convinced, you could then introduce something other than a food
pellet, like a pellet of shit, and see if it has the same effect --
probably not.
The problem here is that you aren't seeing the effect of the reinforcer
on the animal's behavior; you're seeing the effect of changing the
apparatus setting on the closed loop consisting of the animal AND the
apparatus.
Contrast this with the way you investigated the apparatus. As you said,
you didn't even need the rat. In fact, if the rat had been present and
was also pressing the lever, your experiment to determine the ratio
would be confused, because you no longer have exclusive control over the
presses. You need to be able to vary the pressing in a known way, as an
independent variable, while observing the output of the apparatus as a
dependent variable.
To do a parallel experiment with the rat, you would have to take control
of the reinforcers, and administer them according to some pattern while
observing the pattern of behavior. You wouldn't need the apparatus. As
you know, this would not work very well. The reinforcers would not be
contingent on the behavior but would be truly independent variables.
Most likely, the rat would just eat the food and not press the lever (at
least not in the same relationship to the reinforcers as when the loop
is closed).
Did you need experimental evidence for this or just a method for
how one could get experimental evidence?
The method is enough. But the question for which we need evidence is
"How does the rat's behavior depend on the reinforcement rate?" Your
procedure does not answer that question. Instead, it answers the
question "With the loop closed, how does the rat's behavior change when
the schedule is changed?" That does not reveal behavior as a function of
reinforcement rate. The relationship we observe between behavior rate
and reinforcement rate is always determined completely by the apparatus,
by the schedule. On FR-10, there is always one reinforcement for every
10 presses of the lever. That is the apparatus function. With the loop
closed, we have no way of independently observing the organism function;
what we are seeing is the solution to a pair of equations, one
describing the apparatus and the other describing the organism. But the
function describing the organism has not been independently observed.
Reinforcement has been viewed as strengthening behavior in terms of
a monotonically increasing, negatively accelerated function since a
long time ago; your positive acceleration, runaway, interpretation
indicates that you do not actually study reinforcement before you
pontificate on the subject
This is a critical point in our investigation of the difference between
reinforcement theory and PCT, or systems analysis in general. Whether
the observed behavior accelerates positively or negatively is not the
question. The question is what organism function, when used in
conjunction with the apparatus function, will yield a solution in which
the observed behavior accelerates or declerates?
The organism function itself does not produce any acceleration or
deceleration; it simply states the way behavior rate depends on
reinforcement rate when the organism is considered in isolation from the
apparatus. The assumption you make about the organism function, taken
together with the known apparatus function, determines how the variables
will behave through time.
My "runaway interpretation" is not an interpretation; it is the solution
of a particular pair of equations. The assumption was that an increment
in reinforcement produces an increment in the rate of behavior, and that
with zero reinforcement, behavior rate will decline in a specific way
toward zero. The runway effect was an inevitable mathematical
consequence of those assumptions when the parameters had a certain range
of values. For the remaining range, no behavior at all would occur.
If the real behavior does not show that runaway pattern, that does not
show that the runway was incorrectly deduced; it shows that the
assumptions were incorrect. If you want the mathematics to yield a
different result, closer to what is observed, you are going to have to
change an assumption or add a new one. For example, as Bruce Abbott
proposed, you can introduce a cost function, so the net reinforcement
decreases nonlinearly as behavior rate increases. By adding the right
assumed new function, you can then show mathematically that the
solutions are negatively-accelerated curves, as they should be.
All this is part of the process of deducing the organism function from
observations of the visible variables and the known properties of the
feedback function, the apparatus. This process, which is an essential
part of systems analysis, is NOT part of the procedures of EAB, or for
that matter, psychology in general. If you're having difficulty
understanding my point, it's because this approach to analyzing closed-
loop systems is simply not taught in your profession. I never
encountered it once in my undergraduate psychology courses, or in the
year of graduate school in psychology that I had, nor have I ever seen
it used in JEAB or any other psychological journal. In physics it is
used all the time.
I've been looking for a place to insert a comment without interrupting
the argument, but haven't found one yet, so I'll put it in here.
The basic problem with determining the organism function is that if you
were to break the loop and simply manipulate the reinforcement rate
(analogously to the way you investigated the apparatus), the organism
would not continue behaving in the same way. When the loop is broken,
control is lost and other systems in the organism will come into play.
The only way to break the loop successfully is to do so very briefly in
the middle of a closed-loop behavior, returning immediately to the
closed-loop condition. Unfortunately this gives only a very brief
glimpse of the organism function, with an unknown amount of change of
organization commencing immediately when the loop is broken. It is
impossible to do any systematic variation of the reinforcement rate,
because before more than a few seconds or minutes (depending on the
schedule) have passed the organism will start turning to other
behaviors.
The method of varying the conditions with the loop closed can be used
rigorously to find the organism function WITHOUT BREAKING THE LOOP. In
some cases, analytical methods might exist that will work, but the main
practical method is to _propose_ organism functions and use them to
solve for a predicted pattern of behavior. By successive approximations,
an organism function can be found which yields a solution that matches
the actual behavior that is observed. This is basically the method of
modeling, which is used in PCT and also in many other places in the hard
sciences -- but not in psychology.
Bill, are you sincerely saying you don't know why anyone would
think that certain consequences of behavior make them more likely
in the future?
Yes. Well, I know why _I_ don't think it. What I'm questioning is the
word "make." I have no doubt that when behavior produces certain
consequences (when it did not do so before), both the behavior rate and
the consequence rate will increase, up to a point. If I find that
walking past the front door causes a dollar bill to pop through the mail
slot, I will quite likely walk past the door some more. But did the
dollar bill make me do that? Quite the contrary; it was my walking past
the door that caused the dollar bill to appear, and continues to do so
as long as the environment on the other side of the door maintains the
same organization. And more important than that, it is my desire for
dollar bills that drives the whole process. When I have enough dollar
bills, I will go do something more interesting, even though I could
continue to walk past the door and get more dollar bills. It is my own
desire for dollar bills, in relation to my desire for other things, that
leads me to use this method of getting dollar bills. The dollar bills
don't "make" me do anything. What the world on the other side of the
door has done is make it _possible_ for me to get dollar bills in this
way, if I want them.
What your formulation overlooks is that the consequences depend strictly
on the behavior. If for _any_ reason the behavior increases in
frequency, the consequences will increase in frequency. If a particular
consequence is desired, the organism will search for and often find the
action that will produce that consequence. If the consequence ceases to
be forthcoming, the organism will look for some other action to get it,
or turn to seeking some other consequence that is also lacking. In this
way of describing what happens, which is the control-theoretic way,
there is no hint that the consequence itself has any governing power
over the organism. Your formulation is of the form "post hoc, ergo
propter hoc."
-----------------------------------------------------------------------
Best to all,
Bill P.