Konrad Lorenz is the author of a marvellous theory in the field of animal psychology, which he calls “imprinting”. It implies that young animals have an inborn mechanism for jumping to unshakable conclusions. For example, a newly hatched gosling adopts as its “mother” the first moving thing it sets eyes on. This mechanism is well adapted to normal circumstances, though a bit risky for the gosling. (It may also be risky for the chosen foster parent, as we learn from Lorenz.) But it is a successful mechanism under normal circumstances; and also under some which are not quite normal.
The following points about Lorenz’s “imprinting” are important:
(1) It is a process—not the only one—of learning by observation.
(2) The problem solved under the stimulus of the observation is inborn; that is, the gosling is genetically conditioned to look out for its mother: it expects to see its mother.
(3) The theory or expectation which solves the problem is also to some extent inborn, or genetically conditioned: it goes far beyond the actual observation, which merely (so to speak) releases or triggers the adoption of a theory which is largely preformed in the organism.
(4) The learning process is nonrepetitive, though it may take a certain amount of time (a short time),42 and involve often some activity or “effort” on the part of the organism; it therefore may involve a situation not too far removed from that normally encountered. I shall say of such nonrepetitive learning processes that they are “noninductive”, taking repetition as the characteristic of “induction”. (The theory of nonrepetitive learning may be described as selective or Darwinian, while the theory of inductive or repetitive learning is a theory of instructive learning; it is Lamarckian.) Of course, this is purely terminological: should anybody insist on calling imprinting an inductive process I should just have to change my terminology.
(5) The observation itself works only like the turning of a key in a lock. Its role is important, but the highly complex result is almost completely preformed.
(6) Imprinting is an irreversible process of learning; that is, it is not subject to correction or revision.
Of course I knew nothing in 1922 of Konrad Lorenz’s theories (though I had known him as a boy in Altenberg, where we had close friends in common). I shall here use the theory of imprinting merely as a means of explaining my own conjecture, which was similar yet different. My conjecture was not about animals (though I was influenced by C. Lloyd Morgan and even more by H. S. Jennings43) but about human beings, especially young children. It was this.
Most (or perhaps all) learning processes consist in theory formation; that is, in the formation of expectations. The formation of a theory or conjecture has always a “dogmatic”, and often a “critical”, phase. This dogmatic phase shares, with imprinting, the characteristics (2) to (4), and sometimes also (1) and (5), but not normally (6). The critical phase consists in giving up the dogmatic theory under the pressure of disappointed expectations or refutations, and in trying out other dogmas. I noticed that sometimes the dogma was so strongly entrenched that no disappointment could shake it. It is clear that in this case—though only in this case—dogmatic theory formation comes very close to imprinting, of which (6) is characteristic.44 However, I was inclined to look on (6) as a kind of neurotic aberration (even though neuroses did not really interest me: it was the psychology of discovery I was trying to get at). This attitude towards (6) shows that what I had in mind was different from imprinting, though perhaps related to it.
I looked on this method of theory formation as a method of learning by trial and error. But when I called the formation of a theoretical dogma a “trial”, I did not mean a random trial.
It is of some interest to consider the problem of the randomness (or otherwise) of trials in a trial-and-error procedure. Take a simple arithmetical example: division by a number (say, 74856) whose multiplication table we do not know by heart is usually done by trial and error; but this does not mean that the trials are random, for we do know the multiplication tables for 7 and 8.45 Of course we could programme a computer to divide by a method of selecting at random one of the ten digits 0, 1,… 9, as a trial and, in case of error, one of the remaining nine (the erroneous digit having been excluded) by the same random procedure. But this would obviously be inferior to a more systematic procedure: at the very least we should make the computer notice whether its first trial was in error because the selected digit was too small or because it was too big, thus reducing the range of digits for the second selection.
To this example the idea of randomness is in principle applicable, because in every step in long division there is a selection to be made from a well-defined set of possibilities (the digits). But in most zoological examples of learning by trial and error the range or set of possible reactions (movements of any degree of complexity) is not given in advance; and since we do not know the elements of this range we cannot attribute probabilities to them, which we should have to do in order to speak of randomness in any clear sense.
Thus we have to reject the idea that the method of trial and error operates in general, or normally, with trials which are random, even though we may, with some ingenuity, construct highly artificial conditions (such as a maze for rats) to which the idea of randomness may be applicable. But its mere applicability does not, of course, establish that the trials are in fact random: our computer may adopt with advantage a more systematic method of selecting the digits; and a rat running a maze may also operate on principles which are not random.
On the other hand, in any case in which the method of trial and error is applied to the solution of such a problem as the problem of adaptation (to a maze, say), the trials are as a rule not determined, or not completely determined, by the problem; nor can they anticipate its (unknown) solution otherwise than by a fortunate accident. In the terminology of D. T. Campbell, we may say that the trials must be “blind” (I should perhaps prefer to say they must be “blind to the solution of the problem”).46 It is not from the trial but only from the critical method, the method of error elimination, that we find, after the trial—which corresponds to the dogma—whether or not it was a lucky guess; that is, whether it was sufficiently successful in solving the problem in hand to avoid being eliminated for the time being.
Yet the trials are not always quite blind to the demands of the problem: the problem often determines the range from which the trials are selected (such as the range of the digits). This is well described by David Katz: “A hungry animal divides the environment into edible and inedible things. An animal in flight sees roads of escape and hiding places.”47 Moreover, the problem may change somewhat with the successive trials; for example, the range may narrow. But there may also be quite different cases, especially on the human level; cases in which everything depends upon an ability to break through the limits of the assumed range. These cases show that the selection of the range itself may be a trial (an unconscious conjecture), and that critical thinking may consist not only in a rejection of any particular trial or conjecture, but also in a rejection of what may be described as a deeper conjecture—the assumption of the range of “all possible trials”. This, I suggest, is what happens in many cases of “creative” thinking.
What characterizes creative thinking, apart from the intensity of the interest in the problem, seems to me often the ability to break through the limits of the range—or to vary the range—from which a less creative thinker selects his trials. This ability, which clearly is a critical ability, may be described as critical imagination. It is often the result of culture clash, that is, a clash between ideas, or frameworks of ideas. Such a clash may help us to break through the ordinary bounds of our imagination.
Remarks like this, however, would hardly satisfy those who seek for a psychological theory of creative thinking, and especially of scientific discovery. For what they are after is a theory of successful thinking.
I think that the demand for a theory of successful thinking cannot be satisfied, and that it is not the same as the demand for a theory of creative thinking. Success depends on many things—for example on luck. It may depend on meeting with a promising problem. It depends on not being anticipated. It depends on such things as a fortunate division of one’s time between trying to keep up-to-date and concentrating on working out one’s own ideas.
But it seems to me that what is essential to “creative” or “inventive” thinking is a combination of intense interest in some problem (and thus a readiness to try again and again) with highly critical thinking; with a readiness to attack even those presuppositions which for less critical thought determine the limits of the range from which trials (conjectures) are selected; with an imaginative freedom that allows us to see so far unsuspected sources of error: possible prejudices in need of critical examination.
(It is my opinion that most investigations into the psychology of creative thought are pretty barren—or else more logical than psychological.47a For critical thought, or error elimination, can be better characterized in logical terms than in psychological terms.)
A “trial” or a newly formed “dogma” or a new “expectation” is largely the result of inborn needs that give rise to specific problems. But it is also the result of the inborn need to form expectations (in certain specific fields, which in their turn are related to some other needs); and it may also be partly the result of disappointed earlier expectations. I do not of course deny that there may also be an element of personal ingenuity present in the formation of trials or dogmas, but I think that ingenuity and imagination play their main part in the critical process of error elimination. Most of the great theories which are among the supreme achievements of the human mind are the offspring of earlier dogmas, plus criticism.
What became clear to me first, in connection with dogma-formation, was that children—especially small children—urgently need discoverable regularities around them; there was an inborn need not only for food and for being loved but also for discoverable structural invariants of the environment (“things” are such discoverable invariants), for a settled routine, for settled expectations. This infantile dogmatism has been observed by Jane Austen: “Henry and John were still asking every day for the story of Harriet and the gipsies, and still tenaciously setting [Emma]… right if she varied in the slightest particular from the original recital.”48 There was, especially in older children, enjoyment in variation, but mainly within a limited range or framework of expectations. Games, for example, were of this kind; and the rules (the invariants) of the game were often almost impossible to learn by mere observation.49
My main point was that the dogmatic way of thinking was due to an inborn need for regularities, and to inborn mechanisms of discovery; mechanisms which make us search for regularities. And one of my theses was that if we speak glibly of “heredity and environment” we are liable to underrate the overwhelming role of heredity—which, among other things, largely determines what aspects of its objective environment (the ecological niche) do or do not belong to an animal’s subjective, or biologically significant, environment.
I distinguished three main types of learning process, of which the first was the fundamental one:
(1) Learning in the sense of discovery: (dogmatic) formation of theories or expectations, or regular behaviour, checked by (critical) error elimination.
(2) Learning by imitation. This can be interpreted as a special case of (1).
(3) Learning by “repetition” or “practising”, as in learning to play an instrument or to drive a car. Here my thesis is that (a) there is no genuine “repetition”50 but rather (b) change through error elimination (following theory formation) and (c) a process which helps to make certain actions or reactions automatic, thereby allowing them to sink to a merely physiological level, and to be performed without attention.
The significance of inborn dispositions or needs for discovering regularities and rules may be seen in the child’s learning to speak a language, a process that has been much studied. It is, of course, a kind of learning by imitation; and the most astonishing thing is that this very early process is one of trial and critical error elimination, in which the critical error elimination plays a very important role. The power of innate dispositions and needs in this development can best be seen in children who, owing to their deafness, do not participate in the speech situations of their social environment in the normal way. The most convincing cases are perhaps children who are deaf and blind like Laura Bridgman—or Helen Keller, of whom I heard only at a later date. Admittedly, even in these cases we find social contacts—Helen Keller’s contact with her teacher—and we also find imitation. But Helen Keller’s imitation of her teacher’s spelling into her hand is far removed from the ordinary child’s imitation of sounds heard over a long period, sounds whose communicative function can be understood, and responded to, even by a dog.
The great differences between human languages show that there must be an important environmental component in language learning. Moreover, the child’s learning of a language is almost entirely an instance of learning by imitation. Yet reflection on various biological aspects of language shows that the genetic factors are much more important. Thus I agree with the statement of Joseph Church: “While some part of the change that occurs in infancy can be accounted for in terms of physical maturation, we know that maturation stands in a circular, feedback relationship to experience—the things the organism does, feels, and has done to it. This is not to disparage the role of maturation; it is only to insist that we cannot view it as a simple blossoming of predestined biological characteristics.”51 Yet I differ from Church in contending that the genetically founded maturation process is much more complex and has much greater influence than the releasing signals and the experience of receiving them; though no doubt a certain minimum of this is needed to stimulate the “blossoming”. Helen Keller’s grasping (not mentioned by Church) that the spelled word “water” meant the thing which she could feel with her hand and which she knew so well had, I think, some similarity with “imprinting”; but there are also many dissimilarities. The similarity was the ineradicable impression made on her, and the way in which a single experience released pent-up dispositions and needs. An obvious dissimilarity was the tremendous range of variation which the experience opened up for her, and which led in time to her mastery of language.
In the light of this I doubt the aptness of Church’s comment: “The baby does not walk because his ‘walking mechanisms’ have come into flower, but because he has achieved a kind of orientation to space whereby walking becomes a possible mode of action.”52 It seems to me that in Helen Keller’s case there was no orientation in linguistic space or, at any rate, extremely little, prior to her discovery that the touch of her teacher’s fingers denoted water, and her jumping to the conclusion that certain touches may have denotational or referential significance. What must have been there was a readiness, a disposition, a need, to interpret signals; and a need, a readiness, to learn to use these signals by imitation, by the method of trial and error (by non-random trials and the critical elimination of spelling errors).
It appears that there must be inborn dispositions of great variety and complexity which cooperate in this field: the disposition to love, to sympathize, to emulate movements, to control and correct the emulated movements; the disposition to use them, and to communicate with their help; the disposition to react to language; to receive commands, requests, admonitions, warnings; the disposition to interpret descriptive statements, and to produce descriptive statements. In Helen Keller’s case (as opposed to that of normal children) most of her information about reality came through language. As a consequence she was unable for a time to distinguish clearly what we might call “hearsay” from experience, and even from her own imagination: all three came to her in terms of the same symbolic code.53
The example of language learning showed me that my schema of a natural sequence consisting of a dogmatic phase followed by a critical phase was too simple. In language learning there is clearly an inborn disposition to correct (that is, to be flexible and critical, to eliminate errors) which after a time peters out. When a child, having learned to say “mice” uses “hice” for the plural of “house”, then a disposition to find regularities is at work. The child will soon correct himself, perhaps under the influence of adult criticism. But there seems to be a phase in language learning when the language structure becomes rigid—perhaps under the influence of “automatization”, as explained in 3 (c) above.
I have used language learning merely as an example from which we can see that imitation is a special case of the method of trial and error-elimination.54 It is also an example of the cooperation between phases of dogmatic theory formation, expectation formation, or the formation of behavioural regularities, on the one hand, and phases of criticism on the other.
But although the theory of a dogmatic phase followed by a critical phase is too simple, it is true that there can be no critical phase without a preceding dogmatic phase, a phase in which something—an expectation, a regularity of behaviour—is formed, so that error elimination can begin to work on it.
This view made me reject the psychological theory of learning by induction, a theory to which Hume adhered even after he had rejected induction on logical grounds. (I do not wish to repeat what I have said in Conjectures and Refutations about Hume’s views on habit.)55 It also led me to see that there is no such thing as an unprejudiced observation. All observation is an activity with an aim (to find, or to check, some regularity which is at least vaguely conjectured); an activity guided by problems, and by the context of expectations (the “horizon of expectations” as I later called it). There is no such thing as passive experience; no passively impressed association of impressed ideas. Experience is the result of active exploration by the organism, of the search for regularities or invariants. There is no such thing as a perception except in the context of interests and expectations, and hence of regularities or “laws”.
All this led me to the view that conjecture or hypothesis must come before observation or perception: we have inborn expectations; we have latent inborn knowledge, in the form of latent expectations, to be activated by stimuli to which we react as a rule while engaged in active exploration. All learning is a modification (it may be a refutation) of some prior knowledge and thus, in the last analysis, of some inborn knowledge.56
It was this psychological theory which I elaborated, tentatively and in a clumsy terminology, between 1921 and 1926. It was this theory of the formation of our knowledge which engaged and distracted me during my apprenticeship as a cabinetmaker.
One of the strange things about my intellectual history is this. Although I was at the time interested in the contrast between dogmatic and critical thinking, and although I looked upon dogmatic thinking as prescientific (and, where it pretends to be scientific, as “unscientific”), and although I realized the link with the falsifiability criterion of demarcation between science and pseudoscience, I did not appreciate that there was a connection between all this and the problem of induction. For years these two problems lived in different (and it appears almost watertight) compartments of my mind, even though I believed that I had solved the problem of induction by the simple discovery that induction by repetition did not exist (any more than did learning something new by repetition): the alleged inductive method of science had to be replaced by the method of (dogmatic) trial and (critical) error elimination, which was the mode of discovery of all organisms from the amoeba to Einstein.
Of course I was aware that my solutions to both these problems—the problem of demarcation, the problem of induction—made use of the same idea: that of the separation of dogmatic and critical thinking. Nevertheless the two problems seemed to me quite different; demarcation had no similarity with Darwinian selection. Only after some years did I realize that there was a close link, and that the problem of induction arose essentially from a mistaken solution of the problem of demarcation—from the mistaken (positivist) belief that what elevated science over pseudoscience was the “scientific method” of finding true, secure, and justifiable knowledge, and that this method was the method of induction: a belief that erred in more ways than one.