![]()
CHAPTER 1
GOING BEYOND THE INFORMATION GIVEN
H. Gruber et al. (eds), Contemporary Approaches to Cognition: A Symposium held at the University of Colorado (1957), Cambridge: Harvard University Press
More than thirty years ago, Charles Spearman (1923) undertook the ambitious task of characterizing the basic cognitive processes whose operations might account for the existence of intelligence. He emerged with a triad of non-genetic principles, as he called them, the first of these being simply an affirmation that organisms are capable of apprehending the world they live in. The second and third principles provide us with our starting point. One of these, called, as you know, “the education of relations,” holds that there is an immediate evocation of a sense of relation given the mental presentation of two or more things. “White” and “black” evoke “opposite” or “different.” The third principle, the “education of correlates,” states that in the presence of a thing and a relation one immediately educes another thing. “White” and “opposite of” evokes “black.” I think that Spearman was trying to say that the most characteristic thing about mental life, over and beyond the fact that one apprehends the events of the world around one, is that one constantly goes beyond the information given. With this observation I find myself in full agreement, and it is here that my difficulties start. For, as Professor Bartlett (1951, p. 1) put it in a recent paper, “…whenever anybody interprets evidence from any source, and his interpretation contains characteristics that cannot be referred wholly to direct sensory observation or perception, this person thinks. The bother is that nobody has ever been able to find any case of the human use of evidence which does not include characters that run beyond what is directly observed by the senses. So, according to this, people think whenever they do anything at all with evidence. If we adopt that view we very soon find ourselves looking out upon a boundless and turbulent ocean of problems.” Bother though it be, there is little else than to plunge right in.
Some instances of going beyond the information given
It may help to begin with some rather commonplace examples of the different ways in which people go beyond information that is given to them. The first of these represents the simplest form of utilizing inference. It consists of learning the defining properties of a class of functionally equivalent objects and using the presence of these defining properties as a basis of inferring that a new object encountered is or is not an exemplar of the class. The first form of “going beyond,” then, is to go beyond sense data to the class identity of the object being perceived. This is the more remarkable an achievement when the new object encountered differs from in more respects than it resembles other exemplars of the class that have been previously encountered. A speck on the horizon surmounted by a plume of smoke is identified as a ship, so too a towering transatlantic liner at its dock, so too a few schematic lines in a drawing. Given the presence of a few defining properties or cues, we go beyond them to the inference of identity. Having done so, we infer that the instance so categorized or identified has the other properties characteristic of membership in a category. Given the presence of certain cues of shape, size, and texture, we infer that the thing before us is an apple: ergo, it can be eaten, cut with a knife, it relates by certain principles of classification to other kinds of fruits, and so on. The act of rendering some given event equivalent with a class of other things, placing it in an identity class, provides then one of the most primitive forms of going beyond information given.
William James (1890) wrote picturesquely of this process, remarking that cognitive life begins when one is able to exclaim, “Hollo! Thingumbob again.” The adaptive significance of this capacity for equivalence grouping is, of course, enormous. If we were to respond to each event as unique and to learn anew what to do about it or even what to call it, we would soon be swamped by the complexity of our environment. By last count, there were some 7.5 million discriminable differences in the color solid alone. Yet for most purposes we get by treating them as if there were only a dozen or two classes of colors. No two individuals are alike, yet we get by with perhaps a dozen or so “types” into which we class others. Equivalence categories or concepts are the most basic currency one can utilize in going beyond the sensory given. They are the first steps toward rendering the environment generic.
Consider a second form of going beyond the information given, one that involves learning the redundancy of the environment. I present the word, P*YC*OL*GY, and with no difficulty at all you recognize that the word is PSYCHOLOGY. Or the finding of Miller, Heise, and Lichten (1951) that words masked by noise are more easily recognized when they are in a meaningful or high-probability context than when they are presented in isolation. Indeed, the missing word in the sentence, “Dwight _________ is currently President of the United States” can be completely masked by noise and yet “recognized” correctly by anybody who knows the subject matter. Or we find that subjects in some experiments currently in progress check off about an average of thirty trait words from the Gough list as being characteristic of a person who is only described as being either “intelligent,” or “independent,” or “considerate.” Any one of these key traits has at least thirty possible avenues for going beyond it, based on learned probabilities of what things are likely to go with what in another person. Once one learns the probability texture of the environment, one can go beyond the given by predicting its likely concomitants.
We move one step beyond such probabilistic ways of going beyond the information given and come now to certain formal bases for doing so. Two propositions are presented:
and with very little difficulty most people can readily go beyond to the inference that
Or I present a series of numbers, with one missing one to be supplied:
and as soon as you are able to see that the numbers are powers of two, or that they represent successive doublings, you will be able to provide the missing number 16. Or in an experiment by Bruner, Matter, and O’Dowd, rats are taught to find their way through a four-unit T-maze by threading the path LRLR. Given the proper conditions (and to these we will return later), an animal readily transfers to the mirror-image pattern of RLRL – provided he has learned the path as an instance of single alternation and not as a set of specific turns.
What it is that one learns when one learns to do the sort of thing just described, whether it be learning to do syllogisms or learning the principle of single alternation, is not easily described. It amounts to the learning of certain formal schemata that may be fitted to or may be used to organize arrays of diverse information. We shall use the expression coding to describe what an organism does to information under such circumstances, leaving its closer examination until later. Thus, we can conceive of an organism capable of rendering things into equivalence classes, capable of learning the probabilistic relationships between events belonging to various classes, and capable of manipulating these classes by the utilization of certain formal coding systems.
We often combine formal codes and probability codes in making inferences beyond the data. Studies such as those by Wilkins (1928) provide instructive examples. One finds, for example, that a typical deduction made from the proposition “If all A are B” is that “All B are A,” and to the proposition “If some A are not B” a typical conclusion is that “Some B are not A.” Yet none of the subjects ever agrees with the proposition that “If all men are mammals, then all mammals are men,” or with the proposal that “If some men are not criminals, then some criminals are not men.” In sum, it may often be the case that “common sense” – the result of inductive learning of what is what and what goes with what in the environment – may often serve to correct less well learned formal methods of going beyond information given. In short, one may often have alternative modes of going beyond, sometimes in conflict with each other, sometimes operating to the same effect.
One final case before we turn to the difficult business of trying to specify what is involved in utilizing information in this soaring manner. This time we take a scientist, and we shall take him unprepared with a theory, which, as we know, is a rare state for both the scientist and the layman alike. He has, let us say, been working on the effects of sound sleep, and in pursuit of his inquiries has hit on the bright idea of giving his subjects a complete rest for five or six days – “just to see what happens.” To add to their rest, he places them on a soft bed, covers their eyes with translucent ground-glass goggles, lulls their ears with a soft but persistent homogeneous masking noise, and in general makes life as homogeneously restful as possible for them. At the end of this time, he tests them and finds, lo and behold, that they are incapable of doing simple arithmetic problems, that they cannot concentrate, that their perceptual constancies are impaired, and so on down the list of findings that have recently been reported from McGill by Bexton, Heron, Scott (1954), and their collaborators. (Please note that the McGill investigators started with a hypothesis about sensory deprivation; our example is a fiction, but it will serve us and may even relate to our Canadian colleagues before they are through.) Once one has got some data of this order, one is in a funk unless one can go beyond them. To do so requires a theory. A theory, of course, is something we invent. If it is a good theory – a good formal or probabilistic coding system – it should permit us to go beyond the present data both retrospectively and prospectively. We go backward – turn around on our own schemata – and order data that before seemed unrelated to each other. Old loose ends now become part of a new pattern. We go forward in the sense of having new hypotheses and predictions about other things that should be but that have not been tested. When we have finished the reorganizing by means of the new theoretical coding system, everything then seems obvious, if the thing fits. We mention theory construction as a final example of coding processes largely because it highlights several points that are too easily overlooked in the simpler examples given earlier. Coding may involve inventive behavior and we must be concerned with what is involved in the construction of coding systems. And coding systems may be effective or ineffective in permitting one to go beyond information. Later we shall inquire into the conditions that make for construction of new coding systems and what may lead to the construction of adequate ones.
On coding systems
A coding system may be defined as a set of contingently related, nonspecific categories. It is the person’s manner of grouping and relating information about his world, and it is constantly subject to change and reorganization. Bartlett’s memory schemata are close to what is intended here, and the early work of Piaget (1930) on the child’s conception of nature represents a naturalistic account of coding systems in the child.
Let it be clear that a coding system as I describe it here is a hypothetical construct. It is inferred from the nature of antecedent and consequent events. For example, in the rat experiment cited earlier, I teach an organism to wend a course that goes LRLR through a maze. I wish to discover how the event is coded. I transfer the animal to a maze that goes RLRL. He transfers with marked savings. I infer now that he has coded the situation as single alternation. But I must continue to test for the genericalness of the coding system used. Is it alternation in general or alternation only in spatial terms? To test this I set up a situation in the maze where the correct path is defined by taking alternate colors, now a black, now a white member of black-white pairs, without regard to their position. If there is saving here too, I assume that the original learning was coded not as positional alternation but as alternation in general. Of course, I use the appropriate control groups along the way. Note that the technique I am using is identical with the technique we use to discover whether children are learning proper codes in school. We provide training in addition, then we move on to numbers that the child has not yet added, then we move to abstract symbols like a + a + a and see whether 3a emerges as the answer. Then we test further to see whether the child has grasped the idea of repeated addition, which we fool him by calling multiplication. We devise techniques of instruction along the way to aid the child in building a generic code to use for all sorts of quantities. If we fail...
