The Making of Intelligence
By K. Richardson. Columbia University Press, New York, 2000.
Review snippets on the cover of my copy describe this book as a "whodunnit", a "quietly passionate polemic", and a "thought-provoking view". In what follows, I can only belabor what one reads between those lines: the frustrating absence of a balanced, scholarly treatment of intelligence.
Richardson has two central theses. The first is that human intelligence, that which allows us to build airplanes, cure some diseases, and make movie sequels, is best viewed as lying not inside our heads but in the interaction between our minds and the environment. Intelligence is an emergent property of person-in-society, not an inborn capability or an epigenetically developed trait of individuals. There is certainly a sense in which this is true (see, e.g., Hutchins 1995), and there are several recent accounts of the role of a "cultural ratchet" in the evolution of human intelligence (e.g., Tomasello 2001).
The role of biology in Richardson's take on these views is unclear. He is explicit about belief in intellectually progressive evolution: "I assume we can take it for granted that the evolution of living things has consisted, more than anything else, of an increased ability to inhabit more complex and changeable conditions, and that the evolution of the brain and intelligence reflects this" (p. 123). He is wrong; leaving aside tax law, in what sense is the Holocene more complex than the Miocene? Given that hominid brain size tripled in only three million years, "it takes time" just wonÕt wash.
Having established a ladder of progress, he sometimes emphasises the distance humans have come, as when he rejects the relevance of nonhuman primates for understanding human intelligence (we are at "a distinct level of intelligence, integrating but opening up epigenetic regulations"; p. 147). Elsewhere he focuses on the complexity of the lower rungs: for example, he introduces the concepts of "information hyperstructures" or "nested [embedded] covariations" (p. 128) to explain organisms' ability to recognise objects seen from novel angles or fragmented views. He emphasises that such abilities demonstrate that animals cannot be working from a hard-wired (genetically specified) image database or (early) computer-style rigid match-to-sample recognition system. "It is easy to find myriad examples ... in which constantly novel current images have to be rapidly interpreted from representations of informational structure built up from past experience. Adaptation depends neither on innate programs nor empirical copies, but on representational structures of increasing depth and sophistication developing on a lifelong basis" (p. 130). This sounds impressive, but one of the examples my ellipsis hides is "birds flying through trees". At some level, of course, he's correct, but is this level of any interest to those concerned with the misuse of IQ testing?
This brings me to the second thesis, that this view of (human) intelligence as being distributed through space and time in culture and in patterns of social interaction somehow negates the possibility of measuring the individual's contribution. This clearly derives from Richardson's antipathy to the misuses of intelligence testing as a social practice (which I share), but he fails to distinguish among comparing individuals, comparing groups, and defining that which is being compared ("intelligence"). This leads to an overall message of "if we can't measure it and don't fully understand it, it is hypercomplex and so meaningful variation in it doesn't exist". He does not really believe that on the inside we are all intellectually identical in ability, but by denying the possibility of studying differences, we are left with this as a de facto position.
There is no question that the assumptions underlying many uses of intelligence tests are badly flawed. Concerns about cultural bias in IQ testing were initially based on a view of "the environment" as "that to which one is exposed". The unfairness of using the word "regatta" to test urban kids outside the yacht club set had to be explained to test-writers, and much effort has gone into creating "culturally fair" testing vocabularies and switching to matrix manipulations or other non-linguistic tests. Richardson indicates that the problem is deeper, though: Learning is not only about what we know, but provides the basis for mental models (schema) for how to manipulate knowledge, how concepts interrelate, and how those manipulations and interrelations are expressed. The implications for intelligence testing are far more subtle than simply changing vocabulary, and they render unbiased intergroup comparisons of "innate intelligence" of the Bell Curve sort (nearly?) impossible. This is an important point. However, it does not negate the existence of individual variation in the facility with which efficient mental tools are acquired, or the ability to measure such variation within groups, or the possibility that there is a significant genetic contribution to that variation.
Readers of this journal are likely to be most interested in Chapter 3, "Does biology hold the key? Searching biology for intelligence". Here, Richardson takes on heritability of intelligence, studies of brain size, and evolutionary psychological "mental modules" (among others). The technique (similar to that of creationists discussing evolution) is to question conclusions and create doubt in the reader's mind, without offering any complete, balanced analysis of the problem or suggesting a plausible alternative hypothesis for the observations under discussion. If we don't know the answer, we've no business asking the question.
One of his arguments against a genetic basis for observed variation in intelligence is that if "intelligence" were important there would be no remaining underlying genetic variation; since intelligence is important, heritability studies must be flawed (p. 63-4). In a section titled "Keeping quiet about interactions" he asserts that genetic models derived from twin and adoption studies are "of dubious validity" and that "we do not know what the expected correlations should be" (p. 67). He goes on to argue that alternatives to "simple genetic models" are "never considered" and that we need more work on "the nature of ... intelligence ... and how it is affected by the environment. To dash into declarations about genetic variation without doing this seems to be scientifically quite illicit". That some (e.g., an American Psychological Association task force) 'dash in' anyway "seems to bespeak some deeper imperative, perhaps that of simply using science to 'prove' what is already socially assumed or accepted" (p. 68).
Further evidence against genetic reductionism lies in the complexity of regulatory genes. Chapter 5 ("Intelligent systems") introduces their role in development, concluding that epigenesis is really, really complicated. This is important reading for anyone who believes in a simple "gene for intelligence" model, but the complexity of the process has no necessary philosophical connection to the nature of the result. Richardson's discussion of epigenesis is not useful; he presents development as an either/or nature/nurture proposition. For example, "the differentiation of the cerebral cortex ... arises not from a "protomap" laid down in genetic or genomic regulations, but as a result of specific forms of stimulation from the outside world. ... rather than being predetermined by a genetic code for cortical functions, these functions emerge from other regulations operating at critical times in the course of development" (pp. 116-117). That those regulations stem from regulatory genes seems lost on him.
This is a great pity because he loses an opportunity to clarify the distinction between biological and genetical bases for traits, one that is critical for evaluating policy with respect to group test score differences. If a population is systematically subjected to conditions which interact with developmental "programs" in a way that influences brain structure or function, resulting group differences might stem from "biological" differences in individuals (potentially irreversible, physically observable, pervasive in effects) without there being any genetic difference between the groups. Compensatory education may not "cure" the effects of environmental terratogens (the familiar example is the effect of lead on neurogenesis; see Muir & Zegarac 2001 for the societal costs of others), which are 'biological' and which are not distributed randomly with respect to class/ethnicity. However, if policy-makers don't recoil in horror from "biological explanations", confusing them with genetic determinism, other strategies obviously can work.
In Chapter 4, "Computations and connections", Richardson takes aim at connectionism in particular and the use of computer models and metaphors in general. The problem for Richardson is that because a computer's architecture and programming can be specified, a reasonable connectionist simulation of learning/intelligence would necessarily support the logical possibility that genes could analogously specify the architecture and programming of a brain (whether such an analogy would be persuasive is a matter of opinion). Opposed to any hint of genetic determinism, he has no choice but to reject computational approaches to the field. He does this by citing critiques published in 1990 or earlier; connectionism, the main approach in use today, was developed in the mid-1980s. If you see no changes in your software since 1989, he may convince you. The one exception is his citation of Geoffrey Hinton (page 99): "As one of the leaders in the field, Geoffrey Hinton of Carnegie-Mellon University, put it in an article in 1998: 'I am disappointed that we still haven't got a clue what learning algorithms the brain uses'". Though Richardson ends his paragraph with that quotation, he is truncating Hinton who in the source goes on to say: "... but let me say one more encouraging thing"; this is followed by a lengthy, enthusiastic discussion of a promising new approach. For Richardson, because connectionists have had nearly 20 years to figure out how the brain works, and haven't, it's time to give up (p. 100). Unfortunately, all Richardson offers as an alternative is a set of metaphors based on "hypernetworks" and the conclusion that intelligence is a terribly complex emergent property. He eventually may prove to be correct, but it is a bit early to concede.
His writing is clear enough, but the treatment of sources is fuzzy beyond belief. Studies are mentioned with or without citation, and if an author is identified, the work in question may or may not show up in the short, briefly-annotated chapter bibliographies. Oddly, given that a number of works donÕt make it to the bibliography, at least three are included in two or more. One has the impression of chapters written rapidly, perhaps out of sequence, with remembered publications cited and no effort made to track down the rest.
In sum, if you believe that "intelligence" is clearly defined, can be accurately and unambiguously measured using standard techniques, and derives from explicit, invariant mental modules that are rigidly specified by structural genes impervious to environmental effects, then read this book; it will give you a much-needed jolt. Otherwise, Gould (1996) remains definitive regarding the policy side of IQ testing, and Elman et al. (1996) provides a far better account of how complex behaviors can arise through epigenesis.
REFERENCES