Grice on "The Criteria of Intelligence"

Item in the Grice Archive.

From an online source by R. Sternberg:

"Innumerable tests are available for measuring intelligence (see intelligence: measurement), yet no one is quite certain of what intelligence is, or even of just what it is that the available tests are measuring. There have been any number of attempts to resolve these uncertainties, attempts that have differed in their approach to the problem, and in the outcome of applying each given approach."

"One time-honoured approach to discovering the meaning of a construct is to seek expert opinion regarding its definition. This is exactly what the editors of a major psychological journal did in 1921, when they sought the opinions of experts in the field of intelligence regarding what they 'conceive "intelligence" to be, and by what means it can best be measured by group tests' (Intelligence and its Measurement, 1921, p. 123). Fourteen experts replied, with definitions of intelligence such as the following: (i) the power of good responses from the point of view of truth or fact (E. L. Thorndike); (ii) the ability to carry on abstract thinking (L. M. Terman); (iii) having learned or having the ability to learn to adjust oneself to the environment (S. S. Colvin); (iv) the ability to adapt oneself adequately to relatively new situations in life (R. Pintner); (v) the capacity for knowledge, and knowledge possessed (V. A. C. Henmon); (vi) a biological mechanism by which the effects of a complexity of stimuli are brought together and given a somewhat unified effect in behaviour (J. Peterson); (vii) the capacity to inhibit an instinctive adjustment, the capacity to redefine the inhibited instinctive adjustment in the light of imaginally experienced trial and error, and the volitional capacity to realize the modified instinctive adjustment into overt behaviour to the advantage of the individual as a social animal (L. L. Thurstone); (viii) the capacity to acquire capacity (H. Woodrow); and (ix) the capacity to learn or to profit by experience (W. F. Dearborn). The other experts did not answer the question directly.

Viewed narrowly, there seem to be almost as many definitions of intelligence as there were experts asked to define it. Viewed broadly, however, two themes seem to run through at least several of these definitions: the capacity to learn from experience and adaptation to one's environment. Indeed, an earlier definition often cited by these experts viewed intelligence as general adaptability to new problems and conditions of life. (For an update of the 1921 symposium, see Sternberg and Detterman 1986.)

If one is dissatisfied with the heterogeneity in these definitions, one can attempt to answer the question of what intelligence is by begging the question. Edwin Boring (1923) did just that when he defined intelligence as whatever it is that the tests measure. This definition tells us no more than we knew when we started, and it may tell us less: no two tests measure exactly the same thing, so that one is left with as many definitions of intelligence as there are tests (which is certainly a number greater even than that of experts in the field!).

A more recent and sophisticated version of the definitional approach to discovering what intelligence is has been suggested by Ulric Neisser (1979). According to Neisser, the concept of intelligence is organized around a 'prototype', or ideal case. One is intelligent to the extent that one resembles this ideal case:
There are no definitive criteria of intelligence, just as there are none for chairness; it is a fuzzy-edged concept to which many features are relevant. Two people may both be quite intelligent and yet have very few traits in common — they resemble the prototype along different dimensions. ... [Intelligence] is a resemblance between two individuals, one real and the other prototypical. (p. 185)

If there is a single prototype, or ideal case, Neisser's notion will give us a concept of intelligence validated by consensus, if not a concrete definition. There are at least two problems with Neisser's approach, however. First, there exist multiple prototypes, or ideal cases, not just a single one. Different groups of people have somewhat different prototypes. Which one do we use? If we use all of them, including those of various groups of experts and laymen, we end up with as many ideal concepts of intelligence as there are different prototypes, and we are no better off than we were when appealing to experts' definitions. Secondly, the 'ideal case' approach seems to be an excellent way of discovering what people mean by 'intelligence', but not of discovering what 'intelligence' means. Neisser would argue that the two are indistinguishable, but I suspect they are not. Suppose, for example, that people in some culture view their ideal case as able to harmonize with Kron, the god of nature. This description tells us what these people think intelligence is, but it does not tell us much about the nature of intelligence: we still have to find out what it means to harmonize with Kron. In our culture, an analogous notion might be the ability to adapt to natural events. Again, we still need to find out what kinds of mental events and physical behaviours result in the ability to adapt. What is it that people who are able to adapt well do that people who are not able to adapt well do not do?

Questions such as this have led some theorists of intelligence to seek the nature of intelligence by the analysis of individual differences. The question asked here, as above, is what aspects of mental functioning distinguish more intelligent people from less intelligent ones. The nub of this individual-differences approach is to have people perform a large number of tasks that seem to predict intelligent performance (in school, on the job, or wherever), and to analyse patterns of individual differences in task performance. These patterns of individual differences have usually been analysed through the use of a method of statistical analysis called 'factor analysis'. The idea is to identify the 'factors' of human intellect.

The earliest factorial theory of the nature of intelligence was formulated by the inventor of factor analysis, Charles Spearman. Spearman's (1927) analysis of relations among the kinds of mental tests he and other psychologists had been administering led him to propose a 'two-factor' theory of intelligence. According to this theory, intelligence comprises two kinds of factors — a general factor and specific factors. General ability, or g, as measured by the general factor, is required for performance of mental tests of all kinds. Each specific ability, as measured by each specific factor, is required for performance of just one kind of mental test. Thus, there are as many specific factors as there are tests, but only a single general factor. Spearman suggested that the ability underlying the general factor could best be understood as a kind of 'mental energy'.

Godfrey Thomson's (1939) reassessment of Spearman's individual-differences data led him to accept Spearman's hypothesis of a general factor running through the range of mental ability tests; however, it led him to reject Spearman's interpretation of this factor. Thomson disputed Spearman's claim that the general factor represented a single underlying source of individual differences. Instead, he proposed that the appearance of a general factor was due to the workings of a multitude of mental 'bonds', including reflexes, learned associations between stimuli, and the like. Performance of any particular task activates large numbers of these bonds. Some bonds will be required for the performance of virtually any task requiring mental effort, and these bonds will in combination give rise to the appearance of a general factor.

L. L. Thurstone (1938), like Thomson, accepted Spearman's hypothesis of a general factor. But he disputed the importance of this factor. He argued that it is a 'second-order' factor or phenomenon, one which arises only because the primary or 'first-order' factors are related to each other. What are these primary factors, or, as Thurstone called them, 'primary mental abilities'? Thurstone suggested that they include verbal comprehension (measured by tests such as knowledge of vocabulary), word fluency (measured by tests requiring rapid word production — for example, a listing of as many words as a person can think of that have c as their third letter), number (measured by tests of arithmetical reasoning and computation), spatial visualization (measured by tests requiring mental manipulation of geometric forms), perceptual speed (measured by tests requiring rapid visual scanning, for example, proofreading), memory (measured by tests of recall and recognition of previously presented information), and reasoning (measured by tests such as number series, which require people to say which of several numbers should come next in a given series).

J. P. Guilford (1967) parted company with the majority of factorial theorists by refusing to acknowledge the existence of any general factor at all in human intelligence. Instead, he proposed that intelligence comprises 120 elementary abilities, each of which involves the action of some operation upon some content to produce some product. An example of an ability in Guilford's system is 'cognition of verbal relations'. This ability involves recognition of a conceptual connection between two words: for example, recognition that a caboose is often the last car in a train.

Probably the most widely accepted factorial description of intelligence is a hierarchical one. A good example of this class of descriptions was proposed by P. E. Vernon (1971). He proposed that intelligence can be described as comprising abilities at varying levels of generality: at the highest level of generality (the top of the hierarchy) is general ability as identified by Spearman; at the next level are 'major group' factors, such as verbal–educational ability (the kind of ability needed for successful performance in courses such as English, history, and social studies) and practical–mechanical ability (the kind of ability needed for successful performance in courses such as draughtsmanship and car mechanics); at the next level are 'minor group' factors, which can be obtained by subdividing the major group factors; and at the lowest level (the bottom of the hierarchy) are specific factors, again of the kind identified by Spearman. This description of intelligence may be viewed as filling in the gaps between the two extreme kinds of factors proposed by Spearman: in between the general and specific factors are group factors of intermediate levels of generality.

The factorial views of intelligence are unlike the definitional ones in that they are based on the analysis of intelligent functioning (on ability tests), rather than merely on the speculations of one or more psychologists or laymen. The factorial views are like the definitional ones, however, in their potential number and diversity. Is it the case that one of the factorial descriptions is right and the others wrong, or is it possible that a single entity or complex of entities, intelligence, can conform to all of these different descriptions? In other words, is there some level, or common denominator, at which these various descriptions all reduce to the same thing? It is here suggested that such a level of description exists, and that it can be found by analysing the ways in which people process information when solving problems of the kind found both on intelligence tests and in everyday life.

Information-processing psychologists have sought to understand general intelligence in terms of elementary components (or processes) used in the solution of various kinds of problems (Sternberg, 1979a, 1979b). Let us distinguish five kinds of components that people use in the processing of information. Metacomponents are higher-order control processes that are used for planning how a problem should be solved, for making decisions regarding alternative courses of action during problem solving, and for monitoring one's progress during the course of problem solution. Performance components are processes that are used in the actual execution of a problem-solving strategy. Acquisition components are processes used in learning, that is, in the acquisition of knowledge. Retention components are processes used in remembering — that is, in the retrieval of previously acquired information. Transfer components are used in generalization — that is, in the transfer of knowledge from one task, or task context, to another.

Consider, for example, how each of these five kinds of components might be applied in the solution of an arithmetical problem.
Mrs Smith decided to impress Mrs Jones. She went to a costume jewellery shop and bought three imitation diamonds of equal value. She received £4 in change from the £10 note she gave the assistant. (But as Mrs Smith was receiving her change, Mrs Jones walked into the shop!) How much did each imitation diamond cost?

Metacomponents would be used in setting up the equations for solving the problem, for example, in deciding that the problem can be solved by subtracting £4 from £10 and dividing the difference by three; the metacomponents must also decide what information is relevant to the problem at hand, and what information is irrelevant. Performance components would be used in the actual solution of these equations to obtain, first, £6 as the price of the three imitation diamonds and, then, £2 as the price of each item. Acquisition components were used in the problem solver's past to learn how to set up the equations, how to subtract, how to divide, and so on. Retention components are used to retrieve this information from memory at the time that it is needed. Transfer components are used to draw an analogy between this problem and previous ones: the problem solver has never learned how to solve this particular problem, and must generalize his or her learning from similar problems previously encountered to the problem presently being encountered.

How can this scheme account for the various factorial views of intelligence described earlier? According to this view, the general factor that appears in various theories of intelligence results from the operations of components that are general across the range of tasks represented on intelligence tests. For the most part, these are metacomponents — mental activities such as deciding upon the particular components to be used in the solution of problems, deciding upon a strategy for problem solution, monitoring whether the strategy that has been chosen is leading to a solution, deciding upon how quickly the strategy can be executed and still lead to a satisfactory result, and so on. Major group factors of the kind found in Vernon's theory, and primary factors of the kind found in Thurstone's theory, are obtained in factor analyses primarily as a result of the operations of performance, acquisition, retention, and transfer components. For example, verbal comprehension, as tested by vocabulary, is the product of past executions of acquisition components to learn new words, and of present executions of retention components to retrieve the meanings of these words. If vocabulary is tested by presenting the words in unfamiliar contexts, transfer components may also be involved in applying previously acquired information to the new contexts that are presented. Reasoning, as tested by problems such as numerical series completions (say, 3, 7, 12, 18, 25, ...), requires the execution of various performance components, such as encoding the terms of the problem, inferring the relations between the given pairs of numbers, applying these relations to the last given number to obtain the next number, and the production of a response.

This information-processing view of intelligence seems to unify what were formerly a number of disparate views regarding the nature of intelligence. A number of important questions still need to be answered, however, and it is possible to consider only a small number of them here.

First, is the meaning of intelligence the same across different societal and cultural groups? On the view proposed, the answer is both yes and no. On the one hand, the components that would be applied to the solution of a given problem in one culture or society probably overlap to a large degree, although perhaps not completely, those that would be applied to the solution of the same problem in a different culture or society. On the other hand, the kinds of problems that need to be solved may differ widely from one culture or society to another. The mental (and physical) processes needed to corner game in a hunt are very different from those needed to balance accounts. Hence, the kinds of persons who are considered intelligent may vary widely from one culture to another, as a function of the components that are important for adaptation to the requirements of living in the various cultures.

Secondly, if intelligence tests measure, in greater or lesser degree, the components of information processing, why are they so imperfectly predictive of real-world performance? One reason is that they are fallible as measuring instruments: they measure only imperfectly what they are supposed to measure. Another reason is that they do not necessarily weigh most heavily those aspects of intellectual functioning that are most important for intelligent functioning in a given environment. Metacomponential functioning is probably underemphasized, for example, in the measurements made by most of these tests. Yet another reason, and probably the most important one, is that there is a great deal more to everyday living than what the intelligence tests measure, or even than what can reasonably be called intelligence. The tests fail to take into account such important aspects of functioning as motivation, social skills, persistence in the face of adversity, and ability to set and achieve goals. The tests provide reasonably good measures of limited aspects of functioning for most people. But even here a qualification is necessary, since there are some people whose anxieties, or inability to take tests, render their test scores meaningless or even deceptive.

Finally, is intelligence largely inherited, as has been claimed by some (for example, Jensen 1969), or is it largely or exclusively determined by environment, as has been claimed by others (for example, Kamin 1974)? Few bodies of evidence are more confused and confusing than that dealing with the heritability of intelligence. The probability is that heredity, environment, and the interaction between heredity and environment all play some role in intelligence as it has traditionally been measured, but it is not at all clear what the relative extents of these roles are. Nor is it clear what it means, in practical terms, to assign proportional values to the influence of each. No matter what the proportions are, there is good evidence that at least some aspects of intelligence are trainable, and theoretical interest in the heritability of intelligence should not divert attention from questions about how intelligence can be modified in individuals of all levels of measured intelligence."

Bibliography
•Boring, E. G. (1923). 'Intelligence as the tests test it'. New Republic, 6 June.
•Gardner, H. (1999). Intelligence Reframed: Multiple Intelligences for the 21st Century.
•Guilford, J. P. (1967). The Nature of Human Intelligence.
•Jensen, A. R. (1969). 'How much can we boost IQ and scholastic achievement?' Harvard Educational Review, 39.
•Kamin, L. J. (1974). The Science and Politics of IQ.
•Neisser, U. (1979). 'The concept of intelligence'. In Sternberg, R. J., and Detterman, D. K. (eds.), Human Intelligence: Perspectives on its Theory and Measurement.
•Spearman, C. (1927). The Abilities of Man.
•Sternberg, R. J. (1979a). 'The nature of mental abilities'. American Psychologist, 34.
•— —  (1979b). 'Stalking the I.Q. quark'. Psychology Today, 13.
•— —  (2000). Handbook of Intelligence.
•— —  and Detterman, D. K. (eds.) (1986). What is Intelligence? Contemporary Viewpoints on its Nature and Definition.
•Thomson, G. H. (1939). The Factorial Analysis of Human Ability.
•Thurstone, L. L. (1938). Primary Mental Abilities.
•Vernon, P. E. (1971). The Structure of Human Abilities.