__________________________________________________________________________
|  COPYRIGHT NOTICE:                                                       |
|                                                                          |
|  You may forward this document to anyone you think might be interested.  |
|                                                                          |
|  The only limitations are:                                               |
|  A)  You must copy this document IN ITS ENTIRETY, WITHOUT MODIFICATIONS, |
|  including this copyright notice.                                        |
|  You do NOT have permission to change the contents or make extracts.     |
|  B)  You do NOT have permission to copy this document for commercial     |
|  purposes.                                                               |
|                                                                          |
|  The contents of this document are copyright (c) 1966 by Journal of      |
|  the American Society for Psychical Research.                            |
|                                                                          |
|  It was posted on the University of California at Davis ftp server by    |
|  permission of the copyright holder.  This ftp server contains ASCII     |
|  files of published articles by Professor Charles T. Tart.  Individuals  |
|  wishing to obtain other documents there (which are added to from time   |
|  to time) should                                                         |
|	Connect to ftp server, "ftp.ucdavis.edu".                          |
|	Log in as username "anonymous".  Send your e-mail address          |
|	  as the ident/password string.                                    |
|	cd to /pub/fztart.                                                 |
|	A "dir" command will show you what is available.                   |
|	A "get" command will retrieve documents.                           |
|	The file "currentcontents" will be updated regularly, showing      |
|	  what papers are available, perhaps with an abstract of each.     |
|__________________________________________________________________________|
 
 
                   Card Guessing Tests: Learning Paradigm
                         or Extinction Paradigm?
 
                            Charles T. Tart
 
Originally published in the Journal of the American Society for Psychical
Research, 1966, Vol. 60, No. 1, pp 46-55.
 
 
ABSTRACT:  The standard psychological procedure for teaching skilled 
behavior to an organism (whether human or animal) consists of 
administering immediate reward when the correct response is emitted by 
the organism, and/or administering punishment when the incorrect 
response is emitted. If reward and punishment cannot be clearly 
associated with correct and incorrect responses by the organism,
the desired behavior is not learned or, if already present, is 
extinguished.  An analysis of the typical card guessing test situation 
in parapsychology suggests that it is an extinction paradigm. Empirical 
support of this hypothesis is provided by the fact that the scores 
of even the highest-scoring subjects who have come into the laboratory 
have finally fallen off to chance in almost all cases, as would be
expected if they were subjected to an extinction procedure. This 
article makes a plea that the basic principles of learning theory be 
applied to parapsychological tests in an attempt to have subjects 
learn to use their psi abilities more effectively rather than 
extinguishing whatever abilities they have.
 
 
 
 
	In an excellent article (17), Rhea White has pointed out some
striking differences between the conditions in modern card guessing
tests of ESP and in older, but often more significant, experi-
ments in which the subjects had time to use and analyze their
internal imagery and other psychological processes. Her plea that
we should evaluate and study the conditions of these older experi-
ments in order to understand the means whereby ESP impressions
can come into consciousness is one that should be heeded by all
workers in this area.
	White's article reflects a steadily growing disillusionment among
workers in the field about the value of the standard card guessing
tests (note 1) of ESP. This disillusionment is realistic in many ways,
for we seem to have about exhausted this technique: despite many
interesting and minor studies which remain to be done, one can be
legitimately skeptical about the use of a technique where marginally
significant results are the norm, where we are dealing with very
weak manifestations of the underlying phenomena, and where the
magnitude of this manifestation has not increased over decades of
experimentation.
 
-----
Note 1: The term "card guessing tests" will be used broadly in this 
context to include all tests in which the subject chooses between 
several alternative responses on each trial.
endnote------
 
	Important as the need for totally new approaches to studying
ESP is, however, we should be doing more card tests at the same
time because the card guessing tests have, by and large, never been
used in a way which would bring out their possible potential. This
paper will point out a basic flaw in our card guessing test proced-
ures and discuss ways of rectifying it.
	The assumption behind almost all ESP testing has been that we
are trying to detect an extant capacity. It may be more profitable,
however, to assume that whatever this capacity is, it is latent in
the subject (S) and he must learn to use this capacity within the
context of the experimental situation. Let us now consider some
basic facts about learning.
	Learning refers to a hypothetical change within an organism
(whether animal or human) which is reflected or manifested as a
change (improvement) in performance during the course of practice
at some task. Almost all learning takes place in situations where the
correct response is rewarded on each trial and incorrect responses
are not rewarded, or may even be punished. Thus, someone
attempting to learn to play a scale on the piano is rewarded by a
smile from his teacher (and, perhaps, the satisfaction of the
harmony) if he runs through it correctly, but by disapproval
(and disharmony) if he is incorrect. Reward can, especially with
human Ss, also be conceived of as feedback of information on the
correctness or incorrectness of performance, and insofar as the S is
motivated to perform correctly, knowledge that his response was
correct is rewarding.
	Two typical laboratory learning situations will illustrate some
important facts about the learning process. As a case of animal
learning, we have a pigeon inside a soundproof box. In one corner
is a trough where pellets of food may appear as they are released
by an automatic mechanism. As the pigeon is hungry, food is
rewarding. On one wall of the box is a key that the pigeon may
press with his beak, and over the key is a red light. We want the
pigeon to press the key whenever the red light is on, but not when
it is off, so a circuit is set up such that key presses while the light
is on drop food pellets into the trough, while key presses while the
light is off have no effect.
	When first put in the box, the pigeon will be agitated. After
calming down, the pigeon will sooner or later press the key while
the light is on, by "accident." He will be rewarded immediately by
a food pellet. Over the course of a few hours, we will find that the
pigeon pecks rapidly at the key whenever the light is on, and not
when it is off. He has been rewarded for the correct response.
Because his behavior is now correct, we infer that he has learned
the proper response to the red light, viz., key pressing.
	Take a human S and have him extend his hand behind a curtain
to hold a metal stylus. Put the tip of the stylus in a simple maze
constructed so that whenever the stylus goes off the correct path
of the maze it will sound an electric buzzer. Otherwise the S has no
way of knowing when he is on or off the correct pathway. Now
instruct the S to try to trace his way through the maze (which
he cannot see) without going off the path. On his early trials he
will make many deviations from the path, but whenever he hears
the buzzer he will draw back the stylus from that direction. Eventu-
ally (assuming the maze is not too complex) he will be able to trace
through the maze without any mistakes. The reward here is the
lack of the buzzer sounding, for the buzzer is the feedback signal
that he has made a mistake. Behaviorally we may analyze the
responses of the human S and the pigeon in the same way, viz., how
many trials needed to reach a criterion of perfect performance.
Introspectively, for the human S, he might report that the task
was difficult, that he had to coordinate his hand movements with
some sort of image he was developing in his mind, and that while
he might not be able to describe just what he did to get through
the maze without activating the buzzer, he nevertheless learned to
do it. This latter point is particularly important, for there are
many things we learn to do in life which we cannot verbalize to
others, or even adequately conceive of ourselves-try describing
just how you ride a bicycle, for instance. The operation of ESP
is probably no exception to this, i.e., it is possible for a person
to use it without being able to understand or explain just how it
operates.
	Now a number of factors affect the course of learning over and
above the simple presence or absence of reward or feedback, such as
the S's motivation to learn and his state of health. A very important
factor is the time relationship between the S's response and the
reward or feedback. Almost all studies of learning show that
learning is slower and less effective as the interval between response
and reward increases. With lower organisms, particularly, a fairly
lengthy interval between response and reward results in no learning
at all, i.e., the organism never emits the correct response with
greater than chance frequency. In general, intervals between
response and reward or feedback are optimal if they are on the
order of less than a second, and learning falls off rapidly in many
instances if these intervals reach even a few seconds.
	The opposite of learning is extinction, i.e., the correct response
in a situation appearing less and less frequently, and finally failing
to appear altogether. The typical laboratory procedure for pro-
ducing extinction of a learned response is simply to stop rewarding
each such response as it is emitted by the organism. Or one can give
the reward, but give it in such a way that it is ineffective for the
particular organism, e.g., by making the response-reward interval
so long that the organism no longer "associates" the reward with the
correct response.	
 
	In this light, let us examine the typical card guessing situation
as used in almost all parapsychological experimentation. The S
comes in, with some motivation to do well on the test (whether
"doing well" means scoring positively for sheep or negatively for
goats). He is then required to give a large number of responses
(guesses), usually twenty-five, and some of these responses are
correct, some are incorrect. The correct responses may occur with
greater than chance frequency; in fact, they frequently do on the
initial run. After he has emitted a large number of responses,
the experimenter usually tells him which were correct and which
incorrect. There has been no reward or feedback immediately after
each response. Indeed, the feedback coming after such a large
number of trials is probably completely ineffective. (note 2) What little
reward there is (feeling gratified at scoring above chance) tends to be
associated with the entire run rather than with the individual
responses. This paradigm, then, is basically an extinction paradigm,
and is well suited to eliminate correct responses occurring with a
greater than chance frequency.
 
-----
Note 2: We can ignore for the purposes of this discussion the use of 
intermittent reinforcement in psychology, for this is used only after 
some degree of learning has been brought about by constant reinforcement 
of correct responses.
endnote-----
 
	Recall the learning situation with the pigeon, where a food
pellet was produced immediately after each correct response. The
same pigeon, superimposing the card guessing paradigm, would be
put in the box and after it had emitted a large number of responses,
correct and incorrect, the experimenter would give it a handful of
food! No psychologist would expect the pigeon to learn the correct
response; in fact, the key pecking response would remain at an
extremely low frequency as this response would not become
associated with the reward.
	In the maze learning situation with the human, superimposing
the card guessing paradigm, we would disconnect the buzzer,
but tell the S to trace the maze and not make any mistakes. After
he had tried this a number of times we would inform him that he
had made mistakes, and to try again! As before, no learning is
likely to result.
	Looking at the typical card guessing experiment introspectively,
on each guessing trial the S is responding to a host of internal cues,
many of them probably not clearly represented in consciousness
and many of them probably extremely transient. In going over his
results with him at the end of twenty-five trials, we are asking
him to do a rather heroic task, viz., to recall the particular set of
amorphous feelings and sensations associated with each one of
the twenty-five trials and to retrospectively associate these amor-
phous, transient feelings with this late knowledge of results.
Moreover, as White (17) has pointed out, the interval between trials
has typically been much too short for the S to attempt to clarify his
internal feelings and perceptions in the first place.
	Any psychologist, if asked to have any organism learn under
conditions of massed, unrewarded trials, followed by occasional
rewards which cannot be associated with particular responses by the
organism, will throw up his hands in disgust and wonder where the
idea for such a bizarre joke came from.
	Not only does the typical card guessing paradigm fit this theo-
retical model of the extinction paradigm, but the empirical results
hear it out. Almost all Ss, no matter how much above chance
expectancy they are at first, eventually, with repeated testing,
come down to chance expectancy (another of the factors which is
leading to discouragement with card guessing tests). We have,
unknowingly yet systematically, been extinguishing the operation
of ESP in our Ss. Indeed, one might cite as an argument for the    ~
existence and lawfulness of ESP the fact that we are able to extin-
guish it by conventional procedures!
	What can be done about it?
	What is needed is an experimental procedure in which (a) the
S's guesses have virtually immediate consequences, i.e., knowledge
of results and/or reward and/or punishment on every trial; (b) the
testing situation is intrinsically motivating enough to the S so
that some ESP is operative in the first place; and (c) the mechanics
of target selection, recording, and presentation of feedback, reward,
or punishment are unobtrusive so as not to distract the S or the
agent. Note that requirement (b) brings out an assumption basic to
the argument of this paper, viz., that the S will utilize ESP in
conjunction with some of his responses; otherwise there is nothing
to reinforce! If an S is simply guessing, immediate reinforcement
of correct responses amounts to reinforcing randomly varying
factors of no value and there will be no learning to use ESP. If, on
the other hand, the S is using ESP in conjunction with some of his
responses, this is a constant factor that will be reinforced and we
would expect learning to occur.
	The situation is somewhat complicated by the fact that even
with an S who is utilizing ESP in conjunction with some of his
guesses, he is also being reinforced for some responses that are
pure guesses, but are correct by chance alone. One might think of
this as "noise," and this consideration leads us to predict three
general outcomes for experiments using immediate reinforcement:
(a) For an S who shows no ESP at first (indicated by chance-level
scoring), there is nothing to be reinforced, so he will continue to
score at a chance level no matter how long the experiment is
continued; (b) for an S who shows only a little ESP at first, the
infrequent reinforcement of ESP responses and the more frequent
reinforcement 0# chance responses may not allow learning to begin
before extinction has started, i.e., there is far more reinforcement of
"noise" than of "signal," so he will soon start to score at a
chance level; (c) for an S who shows a large number of ESP
responses at first, their systematic reinforcement should outweigh
the reward of chance responses, and learning should take place as
manifested by an over-all increase in scoring level with further
trials. What the exact dividing line is between (b) and (c) constitutes
an empirical problem that future research must solve.
	There have been some experimental setups in the past which have
come close to getting away from this extinction paradigm. The
procedure of allowing Ss to check the calls they felt sure were
correct, as in the research of Humphrey and Nicol (5), was one
approach, but here in many cases the feedback of correctness or
incorrectness did not come until the end of the run; here it would
have been difficult for the S to remember just what his feelings
were that made him check a particular call, such that he could
learn to recognize them clearly in the future. Some experiments
have been done, using open decks, where the experimenter tells the
S whether he is right or wrong after each call. These experiments
would seem to fit a learning rather than an extinction paradigm,
yet in retrospect it is questionable whether the feedback of correct-
ness or incorrectness was very rapid-in dealing with such a
nebulous and poorly understood phenomenon it may be that the
difference between half a second and one second intervals between
response and feedback is crucial. Moreover, the mechanics of the
experimental procedure in these studies (randomizing, recording
responses in duplicate, etc.) may have been a factor detracting from
the opportunity for learning.
	After reading an earlier draft of this paper, Laura Dale was
kind enough to call my attention to a series of experiments carried
out at the A.S.P.R. in which there was quick knowledge of results
for the S. The first of these (7) was one in which the agent pressed
a switch to give a signal to the S as to whether he had been right
or wrong on each trial. Considering the mechanics of this procedure,
however, this feedback was delayed and variable, and thus not
well suited for an initial investigation of the effect of knowledge of
results on learning to use ESP. In three later experiments the
apparatus was modified so that a bell rang immediately if the S
pushed the switch corresponding to a correct guess, thus giving
immediate reinforcement of correctness. In the first study (12)
the authors reported a marked decline effect rather than any
learning. This result does not, however, constitute a demonstration
that immediate knowledge of results or reinforcement fails to help
an S learn to use ESP. As pointed out above, there is probably some
critical ratio of correct responses due to ESP versus correct re-
sponses due to chance which must be reached or exceeded early in
the guessing so that learning can begin before the ESP responses
begin to extinguish. Apparently this ratio was not reached in the
Taves and Dale study (12). Nor was it reached in the two later
experiments (3, 13), as the authors reported there were no significant
results either in over-all scoring or in terms of decline effects: thus
there was probably no ESP to be reinforced.
	A number of mechanical devices have been proposed in
the past which produced random targets and automatically scored
responses (8, 9, 11, 16). Unfortunately, most of these devices
never saw any use to speak of and many of them were actually
rather awkward to operate, so that a quick reward of responses
would have been difficult to accomplish. (note 3)  A modern device which
would easily allow the use of quick reinforcement (2) has been 
proposed, but no one has backed its construction. Another modern
device (10) designed to allow all the techniques of reinforcement
used in modern psychology to be applied has also failed to receive
backing for construction. Apparently the reaction against card
guessing tests has dampened enthusiasm for such testing aids, but
they are absolutely necessary if we are to turn card guessing tests
into a learning situation (using animal or human Ss) instead of an
extinction procedure because manual procedures are probably too
slow, cumbersome, and distracting to both experimenter and S.
 
-----
Note 3: Tyrrell's device (14, 13) seems a noteworthy exception. Here 
the S tried to guess which box among several had a light on inside it. 
On opening the box, the S saw immediately whether she was right or 
wrong.  Tyrrell's tests were some of the most successful in the field, 
despite the drawbacks due to lack of randomicity in a number of his 
experiments.
endnote-----
 
	A properly designed testing aid, which automatically generated
random targets and scored the S's responses, could easily be set
up to do all of the following: (a) allow the S to respond as slowly
as he wishes, thus giving him a chance to clarify his internal
feelings and imagery, or to work rapidly, almost automatically;
(b) reward the S for correct responses, with fixed or variable
intervals between response and reward, and on a constant or
variable reward schedule; (c) provide reward as straight information
feedback (a buzzer for correct responses, say) or provide something
like coins falling from a dispenser on each correct response, or
(d) punish the S for incorrect responses, either in an informational
feedback way or by something like electric shock or monetary
fines. Other techniques could be programmed in, but basically the
point is that by the use of modern apparatus all the highly developed
techniques of learning psychology and operant conditioning could
be applied to guessing situations; and quite possibly we would find
that Ss could learn to perform at more and more significant levels
over time instead of dropping off to chance.
	Considering the literature reviewed above, then, Ss have simply
never been given an adequate chance to learn to use their ESP
abilities, especially those high scoring Ss where the use of immediate
reinforcement techniques would be most profitable. Undoubtedly
there are some other experiments in the scattered literature of
parapsychology which come closer to a learning paradigm that I
(a relative newcomer to the field) know of; what is most amazing to
me as a psychologist, however, is this well-nigh universal use of
an extinction paradigm in parapsychology. The main point of this
paper is a plea to workers in this field to give the learning paradigm
a fair try before abandoning guessing tests entirely.
	It should be noted that these comments are not a sophisticated
analysis of the card guessing method as a learning situation; rather,
they are based on knowledge that can be picked up in elementary
textbooks on psychology (4,6). Because we have been absorbed in the
idea of detection instead of learning, we have actually been working
against ourselves insofar as producing the phenomena we want to
study goes. It is to be hoped that the application of these basic
principles of learning will be carried out, for the crucial problem in
parapsychology today is to produce the phenomena we want to
study at a much higher level than the marginal one we are used
to, and the proper application of the psychology of learning may
be one way of accomplishing this.
 
Postscript: Although I cannot develop the idea here, it is obvious
that everything discussed above about the S learning to use ESP can
also be applied to the problem of the agent learning to "send." The
typical ESP experiment provides no immediate feedback of results
to the agent, so that he is in a poor position to learn to "send" more
effectively.
	As an example of what this sort of feedback might accomplish,
consider the old and remarkably successful experiment of Brugmans
and his colleagues (1, pp. 396-408), where the agent(s) were able
to continuously observe the movements of the S's hand as he
attempted to locate the square to which the agent(s) were trying to
direct him. Whenever the S moved his hand in the right direction
the agent(s) could intensify their "sending," but whenever it went in
the wrong direction they could try another technique and thus
continuously vary their "sending" behaviors in accordance with
what seemed to produce the best responses in the S. This sort of
experiment could easily be carried out today, and now that closed
circuit TV systems are reasonably priced, the agent(s) could be
miles away, totally eliminating problems of sensory cues. They
could jump about and shout if they thought it helped, and work up
a tremendous emotional involvement in their task!
	Giving both the S and the agent a chance to learn to use their
psi capacities should be more fruitful than either approach alone.
 
REFERENCES
 
1.	BRUGMANS, H. "Une communication sur des expe'riences
telepathiques au laboratoire de psychologie a' Groningue faites
par M. Heymans, Docteur Weinberg et Docteur H. I. F. W.
Brugmans." Le Compte Rendu Officiel du Premier Congres
International des Recherches Psychiques. Copenhague, 1922.
2.	CUTTEN, J. H. "ESPIAR" (ESP Indicator and Recorder).
Journal S.P.R., Vol. 41, March, 1961, 32-41.
3.	DALE, L. A., TAVES, E., AND MURPHY, G. "Research Notes."  
Journal A.S.P.R., Vol. 38, July, 1944, 160-170.
4.	HILGARD, E.R. Introduction to Psychology. New York: Harcourt
Brace & World, Inc., 1962.
5.	HUMPHREY, B. M., AND NICOL, J. F. "The Feeling of Success
in ESP." Journal A.S.P.R., Vol. 49, January, 1955, 3-37.
6.	MORGAN,  C.  T. Introduction to Psychology. New York:
McGraw-Hill, Inc., 1956.
7.	MURPHY, G., AND TAVES, E. "Current Plans for Investigations
in Psychical Research." Journal A.S.P.R., Vol. 36, January,
1942,15-28.
8.	SMITH, W. R., AND OTHERS. Testing for Extrasensory Perception
with a Machine. Bedford, Massachusetts: Air Force Cambridge
Research Laboratories, 1963.
9.	STEWART, W. C. "Three New ESP Test Machines and Some
Preliminary Results." Journal of Parapsychology, Vol. 23,
March, 1959, 44-48.
10.	TART, C. T. "ESPATESTER: An Automatic Testing Device
for Parapsychological Research." Awaiting publication in
Journal A.S.P.R.
11.	TAVES, E. "A Machine for Research in Extrasensory Percep-
tion." Journal of Parapsychology, Vol. 3, March, 1939, 11-16.
12.	TAVES, E., AND DALE, L. A. "The Midas Touch in Psychical
Research." Journal A.S.P.R., Vol. 37, April, 1943, 57-83.
13.	TAVES, E., DALE, L. A., AND MURPHY, G. "A Further Report
on the Midas Touch." Journal A.S.P.R., Vol. 37, July,
1943, 111-118.
14.	TYRRELL, G. N. M. "Further Research in Extrasensory
Perception." Proc. S.P.R., Vol. 44, 1937, 99-168.
15.	. "The Tyrrell Apparatus for Testing Extrasensory
Perception." Journal of Parapsychology, Vol. 2, June, 1938,
107-118.
16.	WEBSTER, D. D. "An Automatic Testing and Recording Device
for Experiments in Extrasensory Perception." Journal of
Parapsychology, Vol. 13, June, 1949, 107-117.
17.	WHITE, R. "A Comparison of Old and New Methods of
Response to Targets in ESP Experiments." Journal A.S.P.R.,
Vol. 58, January, 1964, 21-56.
 
Department of Neurology and Psychiatry
School of Medicine
University of Virginia
Charlottesville, Virginia 22901