Logan:  Basic Principles of Learning

                        DISCRIMINATION LEARNING
  
          If two (or more) different stimuli are simultaneously 
          presented to the organism, and only one of them is
          associated with reinforcement regardless of position, 
          the organism will learn to select the reinforced 
          stimulus on future occasions.

     The Principle of Discrimination Learning is distinguished from the
Principle of Differential Conditioning by the fact that the stimuli are
presented simultaneously and the organism is therefore permitted to 
choose between them.  In a typical laboratory study, a rat is presented 
with two doors, one of which is painted black and the other white.  One 
of the doors is arbitrarily designated as the correct (positive) one.  
The positions of the doors are varied irregularly so that the problem 
cannot be solved by consistently going to the left or to the right.  
It is very common under such conditions for rats to begin with a position
preference, tending to choose the left or the right door consistently.
This behavior results in 50% reinforcement as the correct door appears in
the preferred position half the time.  Eventually, the rat will refuse to
approach the negative stimulus in the preferred position and hesitantly
approaches the positive stimulus in the nonpreferred position.
Thereafter, the rat will make very few errors.

     The type of behavior that has just been described has led some
theorists to infer that rats solve discrimination learning problems by 
testing HYPOTHESES.  The presumption is that the rat begins with a 
"position hypothesis" that the correct choice is always left or right. 
After some number of nonreinforcements, however, the rat rejects that 
hypothesis and may try out some other incorrect hypothesis until 
ultimately hitting upon the solution.  This is called the NONCONTINUITY 
THEORY because learning is sudden.  In this type of theory, it is assumed
that the organism can only test one hypothesis at a time; accordingly,
nothing is learned about the relevant cue (black vs white) while testing
out various other hypotheses.  In contrast, the CONTINUITY THEORY
contends that the organism does indeed learn about the relevant cue even
while responding on the basis of a position preference.  For such
theorists, learning is a gradual (continuous) process.  The evidence
suggests that continuity theory is best when the discrimination problem
is a simple one involving stimuli (such as black/white) that dominate the
organism's perceptual field.

     Some theorists have attempted to understand discrimination learning 
without using any new principles; all that is needed is a "choice axiom".
According to such theorists, an excitatory tendency (E) is acquired with
respect to S+ as a result of reinforcement, and an inhibitory tendency
(I) is acquired with respect to S- as a result of nonreinforcement, and
both of these tendencies are subject to the Principle of Stimulus
Generalization.  The result is that S+ will have a greater net (E-I)
tendency, and assuming that the stimulus with the larger net tendency
will e the one chosen, discrimination learning follows automatically.
This approach has been called the ABSOLUTE THEORY because is assumes that
each stimulus has its own excitatory and inhibitory values.

     The contrasting theory is called the RELATIONAL THEORY because it 
assumes that organisms learn to choose on the basis of the relationship 
(e.g., larger, darker, heavier, etc.) Between the stimuli.  Proponents 
of relational theory argued that the phenomenon of transposition
supported their view.  Transposition occurs when an organism is presented
with a different pair of stimuli from that used in training, and selects
on the basis of the relationship between the stimuli.  However
transposition can also be deduced from absolute theory.

     Discrimination learning may be very specific to the context.  For 
example, a rat may learn to choose black over white in one location, and 
then treat it as an essentially new problem when encountered in another
location.  By training the discrimination task in a variety of contexts, 
the generality of the learning is increased.  Another rather surprising
corollary is called the over-training reversal affect (ORE).  Now as the 
name implies, a reversal is simply changing the contingencies so that the
previously reinforced stimulus in now incorrect, and reinforcement is
associated with the previously incorrect stimulus.  The ORE is the fact
that extended training on the original discrimination facilitates 
learning the reversed discrimination.

     An important corollary to the Principle of Discrimination Learning 
is the phenomenon of a learning set, or learning-to-learn.  If a monkey
or other primate is exposed to a large number of discrimination learning
problems, the rate at which these are learned gets better and better.  In
the limiting case, the subject will make at most one error.  This is 
because the organism has no better choice but to guess when first 
presented with a new discrimination problem.  However, once the outcome
of that choice is determined, so is the solution to the problem.  If the 
first choice was correct, the appropriate behavior is to stay with that 
stimulus on the future trials; if the first choice was incorrect, the
appropriate behavior is to switch to the other stimulus on future trials.
Apparently, primates can gradually learn this win-stay-lose-shift pattern
of behavior and accordingly become very proficient at learning
discrimination problems.

     A discrimination learning set is very specific to the type of 
problems experienced during training.  The advantage that adult humans 
have over other organisms is that we have, as a result of formal and 
informal education, learned to learn a very large number and variety 
of problems.  There is evidence that the human child behaves much like 
any naive animal, learning in gradual continuous steps with respect to 
absolute properties of the stimuli, whereas an older child and certainly 
an adult can bring many hypotheses to bear on any new problem, and can 
respond to complex relationships as readily as to absolute stimuli.  In 
effect, the basic Principle of Discrimination Learning is the primitive
beginning of the development of sophisticated, adult human learning 
processes.

TERMS: Stimulus discrimination, hypothesis, learning set, theory
(continuity, noncontinuity), transposition.