Where learning provides
the repertoire of responses than an organism might make, motivation is one
of the main variables that determine which responses will be made and with
what intensity. Motivation is difficult to define since, like learning, it
is not directly observable but can only be indirectly measured through performance.
However, basically motivation
refers to temporary states
that tend to activate behaviors. Two main categories of motivation are drives
and incentives. Drives refer to general energizers of behaviors. A
food-deprived rat is often said to have a hunger drive that energizes those
behaviors which lead to food. Incentives
refer to expectations of
rewards following specific behaviors. Based on past experiences in a maze,
a rat develops an “expectation” that he will find food pellets if he goes
to a particular part of the maze.
Drives can be divided
into specific drives and non-specific drives.
Specific drives energize the organism toward a small class of
goals or objectives. A thirst drive energizes the organism toward the goal
of something to drink, not toward some other goal such as sex. A common trap
when studying human behavior is explaining the behavior by postulating an
associated drive. Thus a social psychologist or anthropologist might try to
explain man’s warlike behavior in terms of a drive or instinct to defend and
expand his territory. Or a personality theorist might try to account for large
parts of human behavior in terms of drives such as power-striving. Such approaches
that explain complex learned behavior by postulating a few drives have very
limited scientific usefulness because of their oversimplification. Also, labeling
a phenomenon does not explain it. Most learning theorists try to minimize
the number of drives they must postulate (e.g., thirst, hunger, release of
sexual tension, needs in terms of stimulus complexity, avoiding pain) and
explain most behavior in terms of what the organism learns to do in various
situations associated with basic drives. Most unlearned drives are based on
biological needs such as food.
A non-specific drive receives input from a number of different drive
sources such as hunger and thirst, and theoretically energizes all behaviors
the organism is involved in. However, specific drives and the drives feeding
into a non-specific drive have associated with them drive stimuli— stimuli specific to the individual drives—as
stomach contractions may be associated with a hunger drive. These drive stimuli
help cue the animal to the appropriate goal (e.g., food if hungry), whereas
the nonspecific drive is assumed to energize whatever behaviors are selected
or cued.
Some theorists deal
only with specific drives, and some, often under the influence of B. F. Skinner,
do no more than describe the deprivation procedures (e.g., “the pigeon is
23-hours food deprived”); they see no need to postulate any types of drives.
Other theorists, often under the influence of Clark Hull, describe behavior
in terms of a non-specific drive with the behaviors guided by the specific
drive stimuli. Neal Miller has suggested that deprivation procedures produce
drive stimuli which, if intense enough, result in a general drive (Miller
& Dollard, 1941).
The relationships
between learning and motivation are very complex (Bolles, 1970; Brown, 1961;
Kimble, 1961, Chap. 13), but to a large extent performance is the interaction
of learning and motivation. Theorists differ in terms of the relative weight
they give to motivational variables and learning variables when explaining
performance. For example, in clinical practice the Freudian approach accounts
for performance primarily in terms of motivational constructs, while behavior
modification includes a clinical approach that emphasizes the role of learning
in explaining performance.
Hull was the first
major theorist to give equal importance to learning and motivation (Bolles,
1970, p. 5). This can be seen in the following formulation from Spence (1960),
who expanded on Hull’s basic model:
R= f[(D+ K) H]
D stands for the
non-specific drive, K stands for the effects of incentives, and H is the learned
habit. Although Spence’s theory is more complex than the above formulation,
and varies with the type of learning situation, we can see that Spence considers
performance (R) to be some function (f) of the product of motivation (D +
K) and learning (H).
Figure 1-1 shows
an often found inverted-U relationship between motivation and performance.
That is, performance is usually best at some intermediate level of motivation
and decreases as motivation is increased or decreased. For example, consider
the effects on test-taking of general arousal, a motivational variable that
corresponds to the amount of excitement within the person. If arousal is too
low, as when the person is tired and uninterested in the test material, performance
on the test will probably be below optimum. Performance will similarly be
poor when arousal is too high, as when the person has excessive test anxiety.
Optimal performance requires some intermediate level of arousal.
A common finding
(Broadhurst, 1957) is that the point of optimal motivation varies with the
complexity of the task; within limits the more difficult the task, the lower
the optimal amount of motivation. This supports the Yerkes-Dodson law proposed 70 years ago (Yerkes & Dodson,
1908). This law suggested that there is an optimal intermediate level of motivation
for learning, which decreases as problem difficulty increases.
_dir%5CtemD082seg431.jpg)
Practically, it is
often very difficult to divide variables into those that affect learning and
those that affect motivation. Many variables probably affect both learning
and motivation to various degrees. Those variables that might affect only
learning or only motivation are hard to discriminate since we can only measure
performance, which is some unknown interaction of learning and motivation.
Another problem in
separating learning and motivation is that learning affects motivation and
motivation may affect learning. The primary way in which learning affects
motivation is through learned drives and learned incentives (to be discussed
in Chapter 5). Basically what is meant here is that much of motivation is
acquired. Man’s striving for approval and recognition is not innate, but is
based on the learned value of approval and recognition.
The effects of motivation
on learning are not always clear. Although motivation affects when learning
takes place, it is debatable whether motivation affects the memory itself.
There are some suggestions that motivation may affect the availability and
retrieval of memories (Weiner, 1966). For example, Weiner has done a series
of experiments (Weiner, 1967; Weiner & Walker, 1966) in which subjects
learn sets of consonants. The consonants (e.g., 3 consonant letters called
“trigrams”) are presented with a background color which tells what the subject
will later receive for correct recall. Thus one color might mean the subject
will get one cent for later correct recall, while another color means five
cents. By controlling for variables such as how much the subject can rehearse
(review his learning) between trials, Weiner argues that later better recall
of high incentive consonants (e.g., 5 cents) over lower incentive consonants
(e.g., 1 cent) is due to the effect of the incentive on the retention of the
memory, rather than on the original learning. There needs to be much more
research in this area.
In Chapter 4 we will
suggest a way that motivation might influence learning by affecting the neural
activity responsible for formation of the memories. If there is some active
process (called “consolidation”) responsible for whatever physiological change
underlies the storage of learned material, then perhaps motivational variables
might affect this process.
Occasionally an organism,
simply as a result of his responding to a specific class of stimuli, will
have a change in his tendency to respond to these stimuli. The change may
be an increment (sensitization) or a decrement (habituation). The changes
are non-associative; they do not result from any learned association, but
occur simply as a result of reacting to the stimuli. That is, the changes
do not require the animal to learn associations between stimuli or associations
between stimuli and responses, as is implied in many forms of learning. The
issue is whether we wish to include such phenomena under our concept of learning.
Some theorists would not include them, since they want all learning to be
associative. However, following Razran (1971), these will be included as simple
forms of learning, there being no reason to restrict learning to associative
processes alone. In the discussion that follows the reader should keep in
mind that there is no concensus on the definitions of sensitization and habituation,
with many theorists meaning significantly different things by the same terms.
According to Razran
(1971, p. 58) sensitization
can be defined as “a more or less permanent
increment in an innate reaction upon repeated stimulation.” This increased
reactivity “is manifested in two behavioral modes: (a) increases in incidence
and magnitude and decreases in latency and threshold of reactions, and (b)
pseudoconditioning, or new reactions to originally inadequate stimuli” (Razran,
1971, p. 78). These are described below.
The following is
an example of the first type of sensitization: a dog is given electric shocks
in the hind leg. These shocks produce a muscular response in the leg. Although
we keep the intensity of the shock constant, we might find that after a few
shocks the dog’s response is more pronounced (increase in magnitude) and occurs
faster (decrease in latency) after the shock comes on. This increment in the
response is sensitization. Similarly, after a few shocks the number of times
the dog actually responds to the shock might increase (increase in incidence)
and the minimum intensity of the shock necessary to produce the response might
decrease (decrease in threshold).
Pseudoconditioning,
the second type of sensitization,
refers to the strengthening of a response to a previously neutral stimulus
through the repeated elicitation of the response by a non-associated stimulus. That is, the stimulus which elicits the response is not paired
(non-associated) with the neutral stimulus. Between trials of shocking the
dog we might present to the dog a tone. At first the tone does not elicit
the shock- produced response; it is a neutral stimulus. But after eliciting
the leg- response by shock a number of times, we might find that the dog now
occasionally responds to the tone, even though the tone and shock were never
systematically paired. This increasing of the leg-response to the tone is
an example of pseudoconditioning.
We could have produced
a similar effect by pairing the tone and shock, presenting first the tone
and shortly thereafter the shock. After a few such pairings, the dog would
respond to the tone. But this is an example of associative learning (respondent
conditioning to be discussed later), based on systematic pairing of stimuli.
For true pseudoconditioning we must assume that the effect is not due to any
associative learning. An interesting question is whether many of the types
of learning we assume to be associative (because we experimentally present
stimuli in what appears to be an associative context) can be more readily
reduced to pseudoconditioning.
Habituation can be defined as “the decrement and disappearance
of innate reactions through nonassociative repeated reacting” (Razran, 1971,
p. 56). If suddenly a cement mixer starts going outside your window, you may
startle and orient to the sound. After a while your reactions to the sound
will decrease. This is an example of habituation. Razran sees habituation
as the lowest level of learning, manifested fully in protozoa and in reactions
mediated by just two neurons (nerve cells). For example, many protozoa will
contract to mechanical stimulation (being touched), but will habituate if
the stimulation is presented immediately after the animals have expanded.
Similar to sensitization
and habitnation is adaptation, the adjustment
of a sense organ to the stimulus environment. When you first go into a darkened
movie theater you have trouble seeing in the dark. But after a short time
your visual system adjusts to the dark (dark adaptation) and you can see better.
Similarly when coming out of the dark theater into bright lights your visual
system must readjust (light adaptation). Although adaptation involves changes
in responses (visual responses in this example), the changes are not specific
to a small class of stimuli as in sensitization and habituation, but rather
are sensory changes generally involving an entire sense modality. Therefore
adaptation will not be included as an example of learning.
OTHER
NON-LEARNING PERFORMANCE VARIABLES
Here we will consider
four other variables that affect performance, but which are not included under
learning: fatigue, maturation, senescence, and stimulus change.
Fatigue is a decreased capacity to respond due to previous
responding. It is a physiological condition of cells and organs. A rat might
learn to run a complex maze for a food pellet. After a number of such trials,
the rat’s running speed may decrease or the rat may stop running altogether.
This may simply be the result of the rat’s being tired, i.e., fatigue. We
wouldn’t want to say that the rat has forgotten the route through the maze
or that it has learned not to run. Some visual perception phenomena are also
explained in terms of fatigue of retinal cells, owing to chemical changes.
Maturation is change which occurs through the biological
processes of growth. Much of human motor development, such as learning to
walk, requires a certain amount of maturation. A child can’t be taught to
walk until maturation provides the necessary development of muscles and nerves.
The type and sequence of pre-walking behaviors the child goes through are
primarily due to maturation processes. Maturation and learning obviously interact,
but it is important to recognize them as separate processes. The fact that
we suddenly see a new behavior in an organism, such as some type of sexual
behavior, does not mean that the organism just learned it. Rather the organism
may have just reached the stage of maturation necessary for this behavior.
At the other end
of the developmental continuum from maturation is senescence—deterioration
with old age. In humans this is generally accompanied by a reduced capacity
to process and retain new information and to recall recent memories. Such
deficits appear to have a physiological basis correlated with aging and should
not be attributed to learning phenomena.
When something is
learned, it is often learned to a specific set of stimuli. The behavior will
occur to other stimuli (stimulus
generalization), but only
to the extent that the new stimuli are similar, from the organism’s point
of view, to the original stimulus. The more similar the stimulus is to the
original, the more probable and the stronger the behavior. Thus if an organism
learns a behavior in one situation and we change this situation (stimulus change), decrements in the behavior may be due to generalization
phenomena, and not to learning variables. Thus stimulus change may produce
changes in performance that do not necessarily reflect changes in learning.
The stimuli to which
the behavior is learned include not only external stimuli but also various
internal states of the organism, such as level of deprivation or chemical
states as might result from drugs. This is referred to as state-dependent learning (Overton, 1969).
In the earliest state-dependent
learning experiment, Girden and Culler (1937) gave dogs a curare type drug
which to a large extent paralyzes muscles. They found that conditioned autonomic
and muscular responses learned while the dog was drugged disappeared in the
non- drugged state, but reappeared when the dog was drugged again. That is,
the response was learned to a set of stimuli including those produced by the
drug. They could even train a dog to make one response to an external stimulus
while drugged and a different response to the same stimulus when not drugged.
Thus it is possible
that a student who uses amphetamines to stay up all night studying for an
exam might find much of his learning is tied to drug-produced stimuli. When
not on the drug much of the learned information might be difficult to recall.
Similarly, a mental hospital which provides therapy to the patients while
the patients are on drugs such as tranquilizers may find poor generalization
of some of the therapeutic effects to the non-drugged state.
Stimulus change is
often useful when trying to establish new behaviors. A married couple might
have a number of interpersonal problems. These problem behaviors then become
associated with the stimuli of the home so that now the stimuli tend to produce
the behaviors. A marriage counselor, after establishing a program of desirable
behaviors with the couple (perhaps while they live in a place other than home),
could utilize stimulus change in the home. The couple might be instructed
to rearrange their furniture, buy some colored lights, change their time and
place of eating, dress differently, buy new paintings, and so forth. This
way, it is hoped, the home stimuli become associated with the new desired
behaviors.
The preceding discussion
covered a number of variables that affect performance, but which are not to
be included in the concept of learning. Our final definition of learning was
as follows: Learning is a relatively permanent change in behavior potential
which occurs as a result of practice, and does not include behavioral changes due to motivation, sensory adaptation, fatigue,
maturation, senescence, or stimulus change.
MEASURES
OF LEARNING AND RETENTION
Remembering that
we cannot measure learning, but only performance, the question is what measures
of performance should we use. The following are a few of the common response
measures that are presumed to reflect learning:
1. Response probability.
How likely is it that the
response will occur in a given situation? (How probable is it that the elementary
student will know the sum of 6 plus 7 after covering this sum once in class?)
2. Response latency.
How long does it take the
response to occur to the stimulus situation? (How quickly can you remember
the definition of “pseudoconditioning” given earlier?)
3. Response duration.
How long does the response
keep going? (How long does the rabbit keep his eyelid closed when being trained
to blink his eye to a tone?)
4. Response amplitude.
How strong is the response?
(How hard does the dog turn the wheel to avoid shock?)
5. Trials to extinction.
How many times does the
response occur after the learning contingency has been removed? (How many
times does a rat continue to press a bar which it learned to press for food
after pressing of the bar no longer yields food?)
Unfortunately there
is little correlation among the various response measures, all of which are
supposedly measures of the same “learning.” The reason for this is not clear.
Perhaps other variables that affect performance cloud the actual correlations,
or perhaps only a couple of the measures actually reflect learning. Probably
the error is in assuming that one type of learning underlies all the measures.
It might be more profitable to think of learning within each separate measure.
Logan (1956) has suggested a micromolar
theory of learning in which
even different response values of the same response system are treated as
being entirely different responses. That is, we can think of an organism learning
to make a response of a certain range of rates, a certain range of amplitudes,
and so forth. If this is the case, then the rate and amplitude are also learned
and are not independent measures of learning. For example, a rat might learn
to press a bar at a certain rate and with a certain amplitude. The rate and
amplitude then are learned and cannot be taken as measures of learning itself.
Given the response
measures listed above, some are used more with some response systems than
others. For many autonomic responses, such as heart rate, amplitude and latency
are common measures. For some skeletal responses, such as pressing a bar,
probability and trials to extinction are common measures, while for others,
such as running a maze, latency is often used.
In studying human
verbal learning three popular measures of retention are recall, recognition,
and savings. Recall is the subject’s ability to reproduce previously
learned material without additional cues. (What is the German word for train
station?) Recognition is
the subject’s ability to identify the correct response from a limited set
of alternatives. (Which of the following five German words means train station?)
Savings refers to how much faster the subject can learn material for the second
time as opposed to the amount of time it took the first time. All of these
are assumed to be measures of memory. But a subject can recognize correctly
when he can’t recall and relearn faster when he can’t recognize. What does
this mean for the concept of memory? Are these measures of different types
of memories, measures of different sensitivities for the same memory, or measures
requiring different types of retrieval processes?
In applied situations
it is important to recognize the independence of response measures that appear
to measure the same learning. Consider a therapy program in which a person
is to learn to overcome some particular fear. The fear can be measured in
terms of the person’s verbal report (what he says about his fear), his overt
motor fear behavior (how he acts in the feared situation), and relevant physiological
measures such as heart rate. The problem is that all of these measures can
be controlled independently of the others (Lang, 1969). Thus the therapist
might conclude, on the basis of physiological measures, that the subject has
learned to overcome the fear, when in fact there is no change in the overt
motor behavior.
Unfortunately, many
change agents, particularly therapists and counselors, rely almost exclusively
on the subject’s verbal report as a measure of learning. But verbal report
is very easy to change without affecting other response systems.
Given that we have
some idea about what learning is, the next question is what is capable of learning. The answer to this question depends to a large
extent on the definition of learning. Some theorists in their definition of
learning put in the requirement of a central nervous system. With such a requirement,
one-celled animals by definition are not capable of learning. But it is probably
desirable not to build such biases into definitions. There is no question
that complex animals such as cats and rats are capable of learning. So let
us ask what is the simplest organism that appears capable of learning according
to the definition of learning given earlier.
The simplest animals,
of course, are protozoans, one-celled animals. Most of the research has involved
paramecia, and most of the controversy has centered on the experiments by
Beatrice Gelber (see discussion in McConnell & Shelby, 1970). Paramecia
generally avoid platinum wires lowered into their cultures. However, Gelber
reported that if she baited the wire with bacteria (food for paramecia), the
paramecia would approach and cling to the wire. Learning seemed to occur in
that, after a number of such bacteria trials, the paramecia would then cling
to the bare wire, even 10 hours after the last bacteria trial.
Critics of these
experiments generally try to explain the effect as being an artifact of the
bacteria introduced into the paramecian culture. For example, in the presence
of the bacteria food in the fluid surrounding the wire, the paramecia will
cling to any nearby structure, which in this case just happens to be the wire.
There is also some question as to whether paramecia can even perceive the
wire except through chemical contamination of the surrounding fluid. The issue
of whether paramecia actually learn in this situation is far from settled.
Earlier we quoted
Razran (1971) as arguing that habituation is manifested fully in protozoa
(e.g., habituation to contraction-producing mechanical stimuli). To the extent
that these results hold up, and since we include habituation as learning,
it appears that protozoa are capable of simple learning and perhaps also of
the more complex associative learning described by Gelber.
An even more controversial
issue is whether plants can learn. Enough people consider such an idea absurd
that most studies on plant learning remain unpublished. Unfortunately, because
of such biases and the lack of a series of well-designed and replicated studies,
little can be said about plant learning at this time. However, one representative,
unpublished study will be given as an example.
Armus conditioned
the plant Mimosa pudica, a sensitive plant that closes its leaves and droops
its stem on contact. The plant will close its leaves when put in darkness,
but not as fast as when the stem of the plant is struck. The conditioning,
then, consisted of turning off the illumination lamp and two minutes later
striking the main stem of the plant. The measure of learning was the latency
of the folding up of the leaves. It was reported that the plant “learned”
to close up its leaves faster to the offset of the lamp, even on test trials
when the stem was not struck. Control plants which had their stems struck
one-half hour before light offset did not show such learning. During extinction,
when the light offset was no longer paired with stem-striking, the latency
of the conditioned plants returned to the level of the control plants. Rather
than decide in advance whether something can learn we should merely see if
it fits into our definition of learning. This procedure, however, often yields
unsettling results. For example, we all assume that linseed oil can’t be capable
of learning. But consider the following: linseed oil when exposed to light
will turn gummy. If we expose it to some illumination, but not enough for
there to be a noticeable effect, we find upon later illumination that it turns
gummy faster than if it hadn’t been previously exposed. Did the linseed oil
“remember” its first exposure and did it “learn” to turn gummy faster? It
might be argued that this fits our definition of learning, unless we pass
it off as something akin to sensory adaptation or maturation. This raises
an important general point: There does not exist a concept or definition of
learning that includes all phenomena generally held to be learning and that
excludes all non-learning phenomena. This, of course, reflects theoretical
differences among researchers, but it also reflects the basic ambiguity surrounding
a concept as basic and broad as “learning.”
Unlearned behaviors
basically can be divided into two groups: reflexes and instincts. A reflex is a simple, unlearned, and immediate response to specific stimulation.
For example, the patellar reflex is the knee kick that occurs when the tendon
just below the knee is tapped. Reflexes depend on maturation. Many of them
are absent in the newborn human and only occur later with maturation.
An instinct, on the other hand, is a more complex, unlearned sequence of responses.
A phenomenon which illustrates instinctive variables is imprinting, the attachment of a behavior pattern to a specific stimulus purely as
a function of exposure to the stimulus at a critical time. For example, newly
hatched ducks will follow whatever moving object (within certain dimensions)
they happen to see during the imprinting time, the maximum time being between
13 and 16 hours after birth (Hess, 1959). This moving object generally is
the mother duck, which is why you often see a mother duck followed by a row
of ducklings. But it could be a human if the mother duck is absent. Naturalist
Konrad Lorenz had ducks imprint on him and follow him about.
For our purposes,
the importance of understanding unlearned behaviors is to keep from confusing
them with learned behaviors. A lot of effort could have been wasted trying
to determine how ducklings learn to follow their mother, when the behavior
is primarily instinctual. Unfortunately, behaviors don’t neatly divide into
the categories of learned and unlearned; rather, there is a complex interaction
between the two. For example, the feeding behavior of a sea gull chick is
primarily instinctual but requires a certain amount of learning. Hailman (1969)
summarizes it as follows: