What has been said above has cleared the ground for the consideration of the next level or logical type of "learning" which we shall here call Learning II. Various terms have been proposed in the literature for various phenomena of this order. "Deutero-learning,"104 "set learning,"105 "learning to learn," and "transfer of learning" may be mentioned.
We recapitulate and extend the definitions so far given:
Zero learning is characterized by specificity of response, which—right or wrong — is not subject to correction.
Learning I is change in specificity of response by correction of errors of choice within a set of alternatives.
Learning II is change in the process of Learning I, e.g., a corrective change in the set of alternatives from which choice is made, or it is a change in how the sequence of experience is punctuated.
Learning III is change in the process of Learning II, e.g., a corrective change in the system of sets of alternatives from which choice is made. (We shall see later that to demand this level of performance of some men and some mammals is sometimes pathogenic.)
Learning IV would be change in Learning III, but probably does not occur in any adult living organism on this earth. Evolutionary process has, however, created organisms whose ontogeny brings them to Level III. The combination of phylogenesis with ontogenesis, in fact, achieves Level IV.
Our immediate task is to give substance to the definition of Learning II as "change in Learning I," and it is for this that the ground has been prepared. Briefly, I believe that the phenomena of Learning II can all be included under the rubric of changes in the manner in which the stream of action and experience is segmented or punctuated into contexts together with changes in the use of context markers.
The list of phenomena classified under Learning I includes a considerable (but not exhaustive) set of differently structured contexts. In classical Pavlovian contexts, the contingency pattern which describes the relation between "stimulus" (CS), animal's action (CR), and reinforcement. (UCS ) is profoundly different from the contingency pattern characteristic of instrumental contexts of learning.
In the Pavlovian case: If stimulus and a certain lapse of time: then reinforcement.
In the Instrumental Reward case: If stimulus and a particular item of behavior: then reinforcement.
In the Pavlovian case, the reinforcement is not contingent upon the animal's behavior, whereas in the instrumental case, it is. Using this contrast as an example,
104 G. Bateson, "Social Planning and the Concept of Deutero-Learning," Conference on Science, Philosophy and Religion, Second Symposium, New York, Harper, 1942.
105 H. E. Harlow, "The Formation of Learning Sets," Psycho!. Review, 1949, 56: 51-65.
we say that Learning II has occurred if it can be shown that experience of one or more contexts of the Pavlovian type results in the animal's acting in some later context as though this, too, had the Pavlovian contingency pattern. Similarly, if past experience of instrumental sequences leads an animal to act in some later context as though expecting this also to be an instrumental context, we shall again say that Learning II has occurred.
When so defined, Learning II is adaptive only if the animal happens to be right in its expectation of a given contingency pattern, and in such a case we shall expect to see a measurable learning to learn. It should require fewer trials in the new context to establish "correct" behavior. If, on the other hand, the animal is wrong in his identification of the later contingency pattern, then we shall expect a delay of Learning I in the new context. The animal who has had prolonged experience of Pavlovian contexts might never get around to the particular sort of trial-and-error behavior necessary to discover a correct instrumental response.
There are at least four fields of experimentation where Learning II has been carefully recorded:
(a) In human rote learning. Hull106 carried out very careful quantitative studies which revealed this phenomenon, and constructed a mathematical model which would simulate or explain the curves of Learning I which he recorded. He also observed a second-order phenomenon which we may call "learning to rote learn" and published the curves for this phenomenon in the Appendix to his book. These curves were separated from the main body of the book because, as he states, his mathematical model (of Rote Learning I) did not cover this aspect of the data.
It is a corollary of the theoretical position which we here take that no amount of rigorous discourse of a given logical type can "explain" phenomena of a higher type. Hull's model acts as a touchstone of logical typing, automatically excluding from explanation phenomena beyond its logical scope. That this was so—and that Hull perceived it—is testimonial both to his rigor and to his perspicacity.
What the data show is that for any given subject, there is an improvement in rote learning with successive sessions, asymptotically approaching a degree of skill which varied from subject to subject.
The context for this rote learning was quite complex and no doubt appeared subjectively different to each learner. Some may have been more motivated by fear of being wrong, while others looked rather for the satisfactions of being right. Some would be more influenced to put up a good record as compared with the other subjects; others would be fascinated to compete in each session with their own previous showing, and so on. All must have had ideas (correct or incorrect) about the nature of the experimental setting, all must have had "levels of aspiration," and all must have had previous experience of memorizing various sorts of material. Not one of Hull's subjects could have come into the learning context uninfluenced by previous Learning II.
In spite of all this previous Learning II, and in spite of genetic differences which might operate at this level, all showed improvement over several sessions. This improvement cannot have been due to Learning I because any recall of the specific sequence of syllables learned in the previous session would not be of use in dealing
106 E. L. Hull, et al, Mathematico-deductive Theory of Rote Learning, New Haven, Yale University, Institute of Human Relations, 1940
with the new sequence. Such recall would more probably be a hindrance. I submit, therefore, that the improvement from session to session can only be accounted for by some sort of adaptation to the context which Hull provided for rote learning.
It is also worth noting that educators have strong opinions about the value (positive or negative) of training in rote learning. "Progressive" educators insist on training in "insight," while the more conservative insist on rote and drilled recall.
(b) The second type of Learning II which has been experimentally studied is called "set learning." The concept and term are derived from Harlow and apply to a rather special case of Learning II. Broadly, what Harlow did was to present rhesus monkeys with more or less complex gestalten or "problems." These the monkey had to solve to get a food reward. Harlow showed that if these problems were of similar "set," i.e., contained similar types of logical complexity, there was a carry-over of learning from one problem to the next. There were, in fact, two orders of contingency patterns involved in Harlow's experiments: first the overall pattern of instrumentalism (if the monkey solves the problem, then reinforcement); and second, the contingency patterns of logic within the specific problems.
(c) Bitterman and others have recently set a fashion in experimentation with "reversal learning." Typically in these experiments the subject is first taught a binary discrimination. When this has been learned to criterion, the meaning of the stimuli is reversed. If X initially "meant" R1, and Y initially meant R2, then after reversal X comes to mean R2, and, Y comes to mean R1. Again the trials are run to criterion when again the meanings are reversed. In these experiments, the crucial question is: Does the subject learn about the reversal? I.e., after a series of reversals, does the subject reach criterion in fewer trials than he did at the beginning of the series?
In these experiments, it is conspicuously clear that the question asked is of logical type higher than that of questions about simple learning. If simple learning is based upon a set of trials, then reversal learning is based upon a set of such sets. The parallelism between this relation and Russell's relation between "class" and "class of classes" is direct.
(d) Learning II is also exemplified in the well-known phenomena of "experimental neurosis." Typically an animal is trained, either in a Pavlovian or instrumental learning con-text, to discriminate between some X and some Y; e.g., between an ellipse and a circle. When this discrimination has been learned, the task is made more difficult: the ellipse is made progressively fatter and the circle is flattened. Finally a stage is reached at which discrimination is impossible. At this stage the animal starts to show symptoms of severe disturbance.
Notably, (a) a naive animal, presented with a situation in which some X may (on some random basis) mean either A or B, does not show disturbance; and (b) the disturbance does not occur in absence of the many context markers characteristic of the laboratory situation.107
It appears, then, that Learning II is a necessary preparation for the behavioral disturbance. The information, "This is a context for discrimination," is communicated at the beginning of the sequence and underlined in the series of stages in which discrimination is made progressively more difficult. But when discrimination becomes impossible, the structure of the context is totally changed. The context
107 H. S. Liddell, "Reflex Method and Experimental Neurosis," Personality and Behavior Disorders, New York, Ronald Press, 1944
markers (e.g., the smell of the laboratory and the experimental harness) now become misleading because the animal is in a situation which demands guesswork or gambling, not discrimination. The en-tire experimental sequence is, in fact, a procedure for putting the animal in the wrong at the level of Learning 11.
In my phrase, the animal is placed in a typical "double bind," which is expectably schizophrenogenic.108
In the strange world outside the psychological laboratory, phenomena which belong to the category Learning II are a major preoccupation of anthropologists, educators, psychiatrists, animal trainers, human parents, and children. All who think about the processes which determine the character of the individual or the processes of change in human (or animal) relationship must use in their thinking a variety of assumptions about Learning II. From time to time, these people call in the laboratory psychologist as a consultant, and then are confronted with a linguistic barrier. Such barriers must always result when, for example, the psychiatrist is talking about Learning II, the psychologist is talking about Learning I, and neither recognizes the logical structure of the difference.
Of the multitudinous ways in which Learning II emerges in human affairs, only three will be discussed in this essay:
(a) In describing individual human beings, both the scientist and the layman commonly resort to adjectives descriptive of "character." It is said that Mr. Jones is dependent, hostile, fey, finicky, anxious, exhibitionistic, narcissistic, passive, competitive, energetic, bold, cowardly, fatalistic, humorous, playful, canny, optimistic, perfectionist, careless, careful, casual, etc. In the light of what has already been said, the reader will be able to assign all these adjectives to their appropriate logical type. All are descriptive of (possible) results of Learning II, and if we would define these words more carefully, our definition will consist in laying down the contingency pattern of that context of Learning I which would expectably bring about that Learning II which would make the adjective applicable.
We might say of the "fatalistic" man that the pattern of his transactions with the environment is such as he might have acquired by prolonged or repeated experience as subject of Pavlovian experiment; and note that this definition of "fatalism" is specific and precise. There are many other forms of "fatalism" besides that which is defined in terms of this particular context of learning. There is, for example, the more complex type characteristic of classical Greek tragedy where a man's own action is felt to aid the inevitable working of fate.
(b) In the punctuation of human interaction. The critical reader will have observed that the adjectives above which purport to describe individual character are really not strictly applicable to the individual but rather describe transactions between the individual and his material and human environment. No man is "resourceful" or "dependent" or "fatalistic" in a vacuum. His characteristic, whatever it be, is not his but is rather a characteristic of what goes on between him and something (or somebody) else.
This being so, it is natural to look into what goes on between people, there to find contexts of Learning I which are likely to lend their shape to processes of Learning II. In such systems, involving two or more persons, where most of the important events
108 G. Bateson, et al., "Toward a Theory of Schizophrenia," Behavioral Science, 1956, 1: 251-64.
are postures, actions, or utterances of the living creatures, we note immediately that the stream of events is commonly punctuated into contexts of learning by a tacit agreement between the persons regarding the nature of their relationship—or by context markers and tacit agreement that these context markers shall "mean" the same for both parties. It is instructive to attempt analysis of an ongoing interchange between A and B. We ask about any particular item of A's behavior: Is this item a stimulus for B? Or is it a response of A to something B said earlier? Or is it a reinforcement of some item provided by B? Or is A, in this item, consummating a reinforcement for himself? Etc.
Such questions will reveal at once that for many items of A's behavior the answer is often quite unclear. Or if there be a clear answer, the clarity is due only to a tacit (rarely fully explicit) agreement between A and B as to the nature of their mutual roles, i.e., as to the nature of the contextual structure which they will expect of each other.
If we look at such an exchange in the abstract:
a1b1a2b2a3b3a4b4a5b5 where the a's refer to items of A's behavior, and the b's to items of B's behavior, we can take any ai and construct around it three simple contexts of learning. These will be:
(a, b, a,+ 1) , in which ai is the stimulus for b,.
(bi_1 a b) , in which ai is the response to b-j, which response B reinforces with b.
(a_1 bi _1 ai) , in which at is now A's reinforcement of B's bi-1, which was response to at_1.
It follows that ai may be a stimulus for B or it may be A's response to B, or it may be A's reinforcement of B.
Beyond this, however, if we consider the ambiguity of the notions "stimulus" and "response," "afferent" and "efferent" — as discussed above — we note that any ai may also be a stimulus for A; it may be A's reinforcement of self; or it may be A's response to some previous behavior of his own, as is the case in sequences of rote behavior.
This general ambiguity means in fact that the ongoing sequence of interchange between two persons is structured only by the person's own perception of the sequence as a series of contexts, each context leading into the next. The particular manner in which the sequence is structured by any particular person will be determined by that person's previous Learning II (or possibly by his genetics).
In such a system, words like "dominant" and "submissive," "succoring" and "dependent" will take on definable meaning as descriptions of segments of interchange. We shall say that "A dominates B" if A and B show by their behavior that they see their relationship as characterized by sequences of the type aiba, where a1 is seen (by A and B) as a signal defining conditions of instrumental reward or punishment; b1 as a signal or act obeying these conditions; and a2 as a signal reinforcing b1.
Similarly we shall say that "A is dependent on B" if their relationship is characterized by sequences a\b\a2,, where al is seen as a signal of weakness; b\ as a helping act; and a2 as an acknowledgement of b1.
But it is up to A and B to distinguish (consciously or unconsciously or not at all) between "dominance" and "dependence." A "command" can closely resemble a cry for "help."
(c) In psychotherapy, Learning II is exemplified most conspicuously by the phenomena of "transference." Orthodox Freudian theory asserts that the patient will inevitably bring to the therapy room inappropriate notions about his relation-ship to the therapist. These notions (conscious or unconscious) will be such that he will act and talk in a way which would press the therapist to respond in ways which would resemble the patient's picture of how some important other person (usually a parent) treated the patient in the near or distant past. In the language of the present paper, the patient will try to shape his interchange with the therapist according to the premises of his (the patient's) former Learning II.
It is commonly observed that much of the Learning II which determines a patient's transference patterns and, in-deed, determines much of the relational life of all human beings, (a) dates from early infancy, and (b) is unconscious. Both of these generalizations seem to be correct and both need some explanation.
It seems probable that these two generalizations are true because of the very nature of the phenomena which we are discussing. We suggest that what is learned in Learning II is a way of punctuating events. But a way of punctuating is not true or false. There is nothing contained in the propositions of this learning that can be tested against reality. It is like a picture seen in an inkblot; it has neither correctness nor incorrectness. It is only a way of seeing the inkblot.
Consider the instrumental view of life. An organism with this view of life in a new situation will engage in trial-and-error behavior in order to make the situation provide a positive reinforcement. If he fails to get this reinforcement, his purposive philosophy is not thereby negated. His trial-and-error behavior will simply continue. The premises of "purpose" are simply not of the same logical type as the material facts of life, and therefore cannot easily be contradicted by them.
The practitioner of magic does not unlearn his magical view of events when the magic does not work. In fact, the propositions which govern punctuation have the general characteristic of being self-validating.109 What we term "con-text" includes the subject's behavior as well as the external events. But this behavior is controlled by former Learning II and therefore it will be of such a kind as to mold the total context to fit the expected punctuation. In sum, this self-validating characteristic of the content of Learning II has the effect that such learning is almost ineradicable. It follows that Learning II acquired in infancy is likely to persist through life. Conversely, we must expect many of the important characteristics of an adult's punctuation to have their roots in early infancy.
In regard to the unconsciousness of these habits of punctuation, we observe that the "unconscious" includes not only repressed material but also most of the processes and habits of gestalt perception. Subjectively we are aware of our "dependency" but unable to say clearly how this pattern was constructed nor what cues were used in our creation of it.
109 J. Ruesch and G. Bateson, Communication: The Social Matrix of Psychiatry, New York, Norton, 1951.
Was this article helpful?