For the fourth session I am asking you to read the accompanying section called "Intelligence" and Kleinig's chapter 11 which corresponds to it.

The question to ponder is: Why do we value intelligence?

The problems

One set of questions arises directly out of "our" concept of intelligence, or perhaps better of intelligent activity. (Note that thinking is a kind of activity.) Several others arise out of the notion of IQ and the psychometric industry. As Kleinig suggests, that industry has been implicated in much nasty business: racialist and reactionary for the most part. It does, however, raise certain issues we should examine. One concerns what psychometricians call the "validity" of IQ tests, or what we might say is the question how well those tests capture what we are interested in when we talk about intelligence. Another, highly controversial set of issues relate to the determinants of IQ — crudely, nature or nurture — but we shall not look very far into these somewhat technical questions.

A generic/specific ambiguity

Kleinig begins his account of "our" concept of intelligence by noting two overlapping sets of contrasts: intelligent vs. non-intelligent and intelligent vs. unintelligent. Once the issue is clearly formulated it is clear that educational interest centres on the second pair. But the point is worth noting explicitly since it is an instance of a fairly frequent kind of ambiguity. In general we tend to think in the following way: for each property, there are things that have it, there are things that don't have it but might have had, and there are things that simply couldn't have it, where the question of having it or not doesn't arise. So roses are red; daffodils are not red, but might have been; sound waves just aren't the sort of thing to be any colour. In this case, we have a word "coloured" to pick out the wide genus of coloured things and various words like "red", "yellow" for the specific sorts of colour that there are. But in some other cases, we use the same word both for the genus and for one of the species within it. (Actually that is true of "coloured" when applied to people, so you may find the illustration helpful, or it may just be more confusing.) So some activities raise moral questions: abortion, rape, gifts to charity,.... But within any such moral question, we often think that the moral position to adopt is just one of the range we find people actually taking up. Similarly we think of ourselves as rational animals, in contrast with non-rational flat worms, but focussing on our own actions we think some of them are rational in a more specific sense while others are more or less irrational.

It is not often that this generic/specific ambiguity is more than a nuisance for our attempts to think clearly about the issues, but it is interesting to ask yourself how far the author of the extract at the end of this section is being taken in by it.

Adverbs rule, but so what?

A common philosophical claim is that one way of talking about X is more fundamental than another, meaning roughly that the fundamental way of talking gets closest to what things are really like, while the less fundamental erects some sort of linguistic smokescreen between us and the basic facts. In Argument Analysis, section 10, I made this sort of claim about "truth" and "true" — suggesting that the noun "truth" could mislead us.

In the case of intelligence, Kleinig echoes Ryle on the tremendous range of things that can be done (un)intelligently, and accepts his claim that intelligent behaviour in one field need not be correlated with intelligent behaviour elsewhere. One could state this position as saying that the fundamental way of talking here is by using the adverb "intelligently". Adverbs modify verbs, and are often "parasitic" on them in that what they convey will depend crucially upon, and vary with, the different verbs they modify.

What Kleinig and Ryle both stress is the highly generic content they think attaches to "intelligent" and its cognates. Kleinig includes a lengthy quotation from Ryle that I shall reproduce to give you a glimpse of a notable philosopher exemplifying his own style of philosophizing:

'Intelligence' is so generic a noun, and such variations exist in the species of intelligence, in the degrees and mutabilities of each of these species and in the conditions of their acquisition and exercise, that the How much? question, which would fit pools, reservoirs and other static containers (even some metaphorical ones) cannot begin to fit the notion of a person's intelligence ... How much Efficiency is there in British Railways? There are thousands of specific ways in which we can indeed sensibly and often numerically compare today's operations, etc., of British Railways with last year's, or with those of the Belgian Railway system, e.g. in respect of train- punctuality, accident-rates, timetable legibility, etc., etc. But these do not congregate into a stock or pool such that we can then nominate an absolute or relative size for it.... With this point there goes another. We can in principle often count the tributaries from which a river or reservoir gets its water; and we can compare the sizes of the water-contributions made by these different tributaries. For these contributions are all contributions of water. But if asked what fraction of British Rail efficiency derives from its nationalised status, and what fraction from electricity, we can only gasp.

The Ryle-Kleinig line is, then, that there is no sensible way of measuring overall general intelligence and a fortiori no way of parcelling it up between different sources (nature and environment). Kleinig does not deny "some necessarily cognitive dimension" (p. 134) to all cases of intelligent action, but he says precious little more to characterize it in any particular type of case.

There are several issues here. We may grant that the adverbial construction is fundamental. But does this mean that a general account is unavailable? Not necessarily. But we may have to be careful. To take a simple analogy: travelling quickly from Kingston to Spanish Town in Jamaica will take a lot less time than travelling quickly from earth to the nearest galaxy, but no one ought to think that "quickly" refers just to a period of time. It seems to refer to a period of time significantly shorter than is normal for activity of the kind in question. A quick marathon lasts a lot longer than a slow 1000 metres, but "quick" means the same applied to a marathon or a 100 metres dash. As a consequence, if you wanted to add "quickness" scores for different races, you would be foolish simply to add the times they took; you would have to do something akin to "standardizing" scores, or in some other way take account of the complexity of the meaning of "quick."

Another point that perhaps needs to be made explicitly is that if you don't know how to F then you can hardly F intelligently, however intelligent you may be elsewhere. Part of the difficulty of the topic is that knowing how to F seems to shade into Fing well, though when we want to we can keep these two distinct, and both separate from Fing intelligently and Fing successfully.

I think we can get something that fits "our" everyday concept of intelligence by adopting an account offered by Schwartz. He claims that intelligent activity manifests effective and efficient use of skills and abilities, a flexible plan of action. As he says, the common belief that intellectual competences are central to intelligence is grounded in the fact that "some skills, in particular some intellectual skills, enable us to use our other abilities and competences more intelligently. Although these intellectual skills are not identical with intelligence and although they themselves may be employed stupidly or intelligently, they can and do play an important role when it comes to planning, adjusting, and adapting other activities" (pp. 262-3).

Schwartz's account squares with the fact that Kleinig notes but tries to explain away that while intelligent behaviour can coexist with unintelligent behaviour we do yet speak of intelligent people simpliciter. And we may note that many educators have thought that their activity might promote some such generalized intelligence. (The considerable body of evidence showing lack of transference of what is taught might just reflect unmotivated "learning".)

Psychological interlude — Sternberg's triarchic theory of intelligence

Schwartz' account is at the level of "our" everyday concept of intelligence and it gives a central role to particular intellectual factors. Without claiming that Schwartz would himself endorse the suggestion, I would like to present you a very brief sketch of a particular psychological conception of intelligence which seems to me to come close to what Schwartz is saying. The following paragraphs are taken from Sternberg's contribution to a discussion on whether intelligence can be improved; he sets out the triarchic theory behind the experiment I mentioned in an earlier section to bear out my preference for explicit teaching over mere practice in learning second order features:

The three processes in the triarchic theory are: selective encoding, selective combination, and selective comparison. The selective encoding process involves distinguishing relevant from irrelevant information. For example, if one is studying for a test or trying to learn something from a book or lecture, one cannot memorize or learn everything; one has to selectively decide what information is relevant, and focus on encoding that in preference to the rest. The second process is selective combination. Once the selective encoding is complete, the information must be integrated in an optimal way so that it fits together. For example, people who study for an exam on difficult and complex material often find that all of a sudden they see the way the subject fits together into a coherent whole. The third process is selective comparison. This process concerns the relationship between the new knowledge and what you already know.

Consider an application of the triarchic theory to verbal comprehension training. We picked that topic because psychologists have found vocabulary to be one of the best single indicators, if not the very best indicator, of a person's overall level of intelligence. Why is vocabulary such a good measure of intelligence? I suggest that it is because vocabulary tests measure indirectly people's ability to learn meanings of words as they encounter the words in context, and this ability to acquire information from context is essential to function intelligently in a wide variety of domains....

According to the theory [applied to learning words from context], three general processes are involved: knowledge-acquisition components, use of these components on context cues, and effective handling of mediating variables that render application of the components to the cues more or less difficult.

The training starts with an exposition regarding three processes of knowledge acquisition: (1) selective encoding, by which information relevant to deciphering the meaning of the word is separated from information irrelevant to that purpose; (2) selective combination, by which relevant information is combined; and (3) selective comparison, by which old information is brought to bear upon learning of the new word.

Context cues are kinds of information to which the knowledge-acquisition components can be applied. There are eight types of context cues: (1) setting cues, (2) value—affect cues, (3) static property cues, (4) active property cues, (5) causal—functional cues, (6) class membership cues, (7) antonymic cues, and (8) equivalence cues. Students receive exercises on the application of the three processes to each of these kinds of cues.

Mediators are the variables that make application of the knowledge-acquisition processes to the contextual cues either easier or more difficult. Seven mediators that have been identified as particularly important in learning words from context are (1) the number of occurrences of the unknown word, (2) the variability of the contexts in which multiple occurrences of the unknown word appear, (3) the importance of the unknown word to understanding the context in which it is embedded, (4) the helpfulness of the surrounding context to understanding the meaning of the word, (5) the density of unknown words in the passages you are reading, (6) the concreteness of the unknown word and of the surrounding context, and (7) the usefulness of previously known information in understanding the passage or in understanding the meaning of the unknown word.

Each of the concepts in this theory is explained and illustrated through the use of passages that require learning of words from context. Consider an example of one of the 17 passages used to illustrate these concepts:

Although the others were having a marvellous time at the party, the couple on the blind date was not enjoying the merry-making in the least. A pococurante, he was dismayed by her earnestness. Meanwhile, she, who delighted in men with full heads of hair, eyed his substantial phalacrosis with disdain. When he failed to suppress an eructation, her disdain turned to disgust. He, in turn, was equally appalled by her noticeable podobromhidrosis. Although they both loved to dance, the disco beat of the music did not lessen either their ennui or their mutual discomfort. Both silently vowed that they would never again accept a blind date.

The meaning of the unknown words are:

• pococurante – a nonchalant or indifferent person;
• phalacrosis – baldness; a bald spot;
• eructation – a belch;
• podobromhidrosis – smelly feet.

Through the use of aspects of the theory, it becomes possible to improve one's ability to learn meanings of words from context, and thereby to improve one's vocabulary.

R.J. Sternberg in Thinking: the Second International Conference, eds. D.N. Perkins, J. Lochhead, and J. Bishop (Hillsdale, NJ: Lawrence Erlbaum, 1987), pp. 55-57.

Sternberg here seems to pick on intellectual skills or strategies that can be used to produce more intelligent behaviour, in the way Schwartz describes. There is something of catching your own tail as well, in that, say, following an injunction to distinguish relevant from irrelevant information intelligently requires you to know which is which. But luckily many people do, and they can be encouraged to be more successful overall by being asked to monitor their intellectual activity more carefully.

Abilities and content specificity

We have noted that in everyday contexts a person might know how to do various things, might exhibit intelligence in doing some of them, but be hopelessly unintelligent in doing others. Schwartz' story does not need to deny any of this. It simply says that a person we call "intelligent" without specifying some limited kind of activity is one who tends to do different things intelligently, just as a prudent person tends to do things prudently. But a prudent person may occasionally do something rashly, just as a generally intelligent person may do some things stupidly.

A similar picture greets us when we turn simply to the things people can do. However we divide up abilities, a person will do some things better than others, and many things not at all.

This variation can cause difficulties when you try to arrive at a psychological theory of what is going on. For instance, Piagetian theories of cognitive development tend to present a picture in which a child thinks one way at one time and in a different way at a later time, in both cases with respect to all the child's thoughts. But what we find is that a child may think one way about X and a different way (in Piagetian terms) about Y at the very same moment of its life.

Of course, you can go in the other direction and set up a theory in which different contents are dealt with by different cognitive subsystems, and you can get some very interesting clues from people with exceptional abilities and also from what happens to people with various sorts of brain damage. Such people seem to show that what appear to us as unified mental happenings are in fact the resultant of various separate subsystems, any one or more of which can be "switched off." People can lose colour vision, or the ability to see things in motion, or again they can keep colour vision but forget the words for colours. (For these and other cases, see the September 1992 special issue of Scientific American.)

Given the complexity of human beings, what I am adverting to are difficulties for theories, not necessarily decisive refutations of them. There are lots of adjustments that different theories can make to accommodate, including often reinterpreting the "data."

Second psychological interlude — Gardner's "multiple intelligences"

As an example of a theory that postulates different subsystems, I shall give you a few quotations from a discussion of a recent and popular redaction: Howard Gardner's theory of multiple intelligences. He considered various kinds of evidence about how the mind is organized (brain damage; prodigies; psychometrics; transferability of learning;...) and the different uses different cultures make of cognitive skills:

I eyeballed the data for a long time, testing for plausibility various ways of classifying "intelligences," and thus arrived at a provisional list of seven different kinds:

Linguistic intelligence: that is the intelligence of a poet or an orator;

Logical Mathematical Intelligence: that is the intelligence found in a mathematician or a scientist.

I always mention these two intelligences first, because they are the ones most prized in the west; the ones most important for schooling, and that, by and large, standardized tests exist to assess. I accept that these tests do a reasonable job in assessing linguistic and logical mathematical intelligence. But where I begin to depart from my colleagues is in my claim that there are five other intelligences that deserve to be placed on their own pedastals along with the other two:

Musical Intelligence: found in the composer;

Bodily Kinesthetic Intelligence: the ability of the dancer, and of the athlete; the person who works with the hands, the craftsperson or surgeon;

Spatial Intelligence: found in the sculptor, surveyor, and the person skilled in topology.

Many psychologists, swallowing hard, can accept the five intelligences so far given; it is the last two, which are the `personal' forms of intelligence, which are the most controversial, except, interestingly enough, amongst clinicians:

Interpersonal Intelligence: this concerns knowledge of other people, how they work, and how to work with them: salesmen and women, politicians and religious people usually have high interpersonal intelligence;

Intrapersonal Intelligence: this is the skill correlative to the previous one, and turned inward: it has to do with self-knowledge, and culminates in a developed sense of self, that includes an accurate notion of what you can and cannot do, an ability to plan and the like.

H. Gardner in Thinking: the Second International Conference, p. 81.

Intelligence as ethics

Gardner is offering a theory of mental structure; intelligence only gets in because, as Schwartz is happy to agree, you can do many different things intelligently (and possibly via the generic/specific ambiguity). To show the variation among card-carrying psychologists, it may be worth noting a view which is not far removed from Schwartz' level of analysis and which also brings us back to another range of educational concerns.

Jonathan Baron, in the same discussion from which I have quoted Sternberg, contrasts skills, strategies, and styles. He says skills cannot be trained so that they generalize;¹ strategies can be trained in general but there are not many which are sufficiently broad for his taste; styles, however, are already general and can be fruitfully trained. "I have in mind things like thoroughness in searching for evidence, willingness to consider alternative possibilities, and fairness in the way one goes about searching for evidence and using it. To some extent, these styles may be taught as habits, the way one teaches good manners. But I think a more productive way to teach them is by instilling appropriate goals and beliefs. Just as we may teach good manners by instilling a concern for others, we may teach good thinking by instilling a concern for the truth and a belief that it is possible to get to the bottom of things through our own efforts. In essence, the teaching of intelligence, like the teaching of moral behaviour, involves the enforcement of certain standards of conduct" (op. cit., pp. 60-61).

It is not a coincidence that traditionally people have spoken of the intellectual virtues alongside the moral ones. They are different, but their possible connections might provide an answer to those who think that a stress on academic subjects must ignore moral development.

Aspects of IQ

We have seen something of what we are ordinarily concerned with in thinking about intelligence, and some of what psychology has offered to help elucidate these things. We have so far largely ignored the enormous psychometric industry and its central notion of IQ. But it is this branch of study that has had the most extensive consequences for schooling, not merely in the US but also for us in, say, some of the types of test used in "Common Entrance" examinations. Kleinig gives a good account of the problems raised by IQ, and one of his sources, the long two-part article by Block and Dworkin, is also worth reading if you can find it. I shall just note a few of the main points.

• The absence of theory. I have already tried to indicate the variety of "theories" in current psychology that claim to address intelligence. What is remarkable about the IQ movement is its atheoreticity, its lack of any theoretical guidance in working out how to measure intelligence.² The most it has used is the commonsense or the prejudices of various investigators: do the results on these tests rank people the way we think they ought to be ranked? We have already seen how difficult it is to answer that sort of question in everyday life; but psychometricians have not in general sought any more refined guidance. Crudely, Binet got something that fitted teachers' judgments and ever since new tests are judged by how well they fit earlier ones, with a little adjustment when people make a lot of fuss about cultural or gender differences. For an activity with pretensions to science, this is a remarkably slapdash way of proceeding. Block and Dworkin offer an extensive discussion of the way scientists have refined our notion of temperature in the light of theoretical considerations; and this is typical — we need theory to revise our conceptions of the world, not merely to enshrine them. The absence of theory is of course the more worrying in that what IQ operators pick upon (speed; verbal or mathematical puzzles; mental manipulation of shapes;...) is not often obviously the sort of thing we value in everyday intelligent action.
• Innateness. Given that we have no theoretical grasp of what is measured by IQ tests, we have little reason to suppose that whatever it is is an unalterable given. The immense coaching industry that grows up around institutions that exploit IQ testing testifies to a general belief that ability here can be influenced by practice or even insight; though of course people might just be wasting their money.
• Explaining the correlations. Years of psychometric testing have yielded a lot of correlational data: basically people who do well on test A tend to do well on test B and show all sorts of abilities on test C. In the absence of theoretical guidance on what might explain these correlations, psychometricians have used various techniques (mostly factor analysis of various sorts) to try to simplify the picture. Kleinig suggests that there is an apparently geographical divergence of opinion over the existence of a "general" factor of intelligence, Spearman's g. Now while the tests may not have much to do with the richer conception of intelligence that we use and which educators at least tend to value, one can say that there are in fact a number of fairly well established correlations between the scores people get on batteries of such tests,³ and it should be a scientifically decidable question how best to account for the patterns of correlations so revealed. Given my geographical background perhaps, I am inclined to believe the authorities who claim that the evidence favours a general factor but the point here against Kleinig is that there is something to be explained. It may lack educational significance, however, inasmuch as the tests may be measuring trivial accomplishments. It may also be urged against me that the only decent way to resolve the question is by providing an independent theory: try analysing particle physics data without a theory of particles.
• Heritability. Heritability and the arguments surrounding the heritability of IQ are a complex area. An estimate of the heritability of a trait requires a reliable estimate of its genetically based variance, but in the case of IQ it is virtually impossible to disentangle genetic from environmental sources. Secondly, heritability is a characteristic of a population in an environment; in general if you change the environment you change the heritability of a trait. Thirdly, you may make significant changes in the manifestation of a trait by changing the environment without altering its heritability: a striking instance is height — over the last century people in the developed world have grown taller because of nutritional improvements, while height has remained highly determined genetically. Unless IQ (or more importantly intelligence) is an exceptional case, these facts undermine most arguments for not intervening in educational or other ways to boost the trait in question. (This is to grossly oversimplify, though I hope not to misrepresent, some difficult issues - consult Block & Dworkin, pp. 40-55, for the details.)

A non-conclusion

A couple of final suggestions. I have endorsed Schwartz's concern to preserve the intellectual focus of our educational wish to promote intelligence. It might be worth saying explicitly that, as I understand this, it requires a disciplined use of the imagination: flexibility typically involves seeing what may happen, rather than waiting for it actually to come about. Secondly, we may well find it advantageous to swap our unreflective though intelligent ways of doing something for a more "mechanical" routine method. A lot of what one learns in mathematics at school or in elementary logic consists of such algorithms or procedures for getting a reliable answer. While application may still require some intelligence, the routine nature of such methods means that people's initial intellectual endowments need not restrict their taught achievement. Put like that it may seem too trivially true to be worth saying, but in the context of the IQ industry I suspect otherwise.

This section has opened out into several large questions. I leave you with an extract taken from an educational magazine which may throw yet another light on our topic — it certainly gets into some very murky waters at the end.

It is not the habit of researchers in artificial intelligence to define "intelligence" itself when asked to say what they do for work. Marvin Minsky responds typically to the question "what is artificial intelligence?" in the preface to Semantic Information Processing, saying simply that AI is the project of constructing machines to do things which would require intelligence if done by people. The occurrence of the term "intelligence" in the definition signals an appeal to an intuitive notion of intelligence. More importantly, the appeal singles out human intelligence. So although AI may in principle investigate the whole range of possible intelligences, in practice it begins by addressing itself only to the human sector. Its practitioners appear simply to have no interest in the myriad possible intelligences at levels other than the human one. Human intelligence is both the model and the goal. In this sense AI is thoroughly anthropo-centric. How has this influenced the direction of research?

Anthropocentrism has deep roots in Western culture, but these inherited tendencies in AI were reinforced by two developments. One was the failure of the behaviourist paradigm in psychology to account for human behaviour, despite early successes with rats. The study of infra-human intelligence thus appeared to be an unproductive detour. The second development was the computer conceived of as a symbol cruncher. Computers are instinctively good at precisely what behaviourism seemed incapable of explaining, that is, the sort of abstract formal reasoning prominent in logic, grammar and mathematics. So for the first time it seemed that human level artificial intelligence was not only attainable, but that research could start at that level without tedious detours through infrahuman levels.

So landmark systems in early AI exhibited fragments of behaviour clearly identifiable as human. They played chess, solved cryptarithmetic problems, high-school algebra word problems, and so on. Underlying assumptions about human intelligence are apparent here. Your average cat neither plays chess nor solves algebra word problems. This betrays a second overlay of anthropocentrism in AI. Not only is human level intelligence regarded as central, but the features which differentiate it most sharply from infra-human intelligence are regarded as its most significant features.

But these AI systems cannot deal — were never designed to deal — with real-life problems like getting the cat to come out from under the bed or planning a trip to Arizona. The difference between these problems and algebra word problems is that solving the latter does not require that you know anything at all about the things mentioned — cities and distances, say. It only requires you to consider abstract entities with some formal properties and relations specified within the problem itself. But real-life problems require that you have reams of concrete knowledge about, for example, cats or travel. The gulf separating abstract reasoning ability from this ability to deal with real-life situations became known as the commonsense knowledge problem; and AI researchers immediately set about formalising commonsense understanding.

For example, in The Naive Physics Manifesto Patrick Hayes proposed that everyday notions of how physical stuff behaves consist of a rich array of densely interconnected concepts, "clustered" around key concepts. Examples include: a "history" cluster concerning processes like trajectories, accumulations, disintegrations, etc; a "support" cluster involving stability, piling, sliding, hanging, floating, etc.

Initially it was thought that abstract reasoning is the hard part, and that commonsense could be easily reconstructed as a mopping up exercise. But opinion shifted sharply as the enormous complexity and volume of the commonsense knowledge necessary for everyday activity became apparent, and there is now a virtual consensus that commonsense poses a much greater challenge to AI. So AI's anthropocentrism has been compromised to the extent that the more specifically human aspects of intelligence are no longer the focus of research.

But anthropocentrism is compromised here in a deeper way, too. Commonsense understanding is a sort of intelligence we share with non-human animals. Infra<-> human intelligence must include some sort of na<139>ve physics, for example. To leap from the table to the top of the refrigerator, your cat must have some conception of trajectories, relative stability of objects, and so on. So the emphasis on commonsense understanding constitutes a rejection of anthropocentrism, first, because it invites contemplation of the points of commonality between our intelligence and the intelligence of non-human animals, and second because it shows that these common abilities are much harder for AI to duplicate than, for example, the ability to play chess.

Another consequence of the initial emphasis in AI on abstract reasoning as the core of intelligence was that sensory abilities such as vision were regarded merely as supplying input, and not as intelligent (or interesting) in their own right. Indeed, legend has it that Marvin Minsky once assigned "solving the vision problem" to a bright MIT undergraduate as a summer project. The vision problem was not solved that summer, nor has it been since, although machine vision has become a major research area in AI. In fact, it is currently bogged down at a crucial point. Although research into early vision (which extracts information about edges, regions, depth, etc) has enjoyed considerable success, research into high level vision (which is responsible for object recognition) has made little progress. So despite considerable expenditure of effort AI has made little headway on the main function of vision, to let us know what is out there.

The problems involved in navigation and manipulation were similarly underestimated, and have proven equally difficult to solve. Thus apparently low level abilities like perception and locomotion — which may collectively be called periphral intelligence — have resisted simulation in a way that abstract reasoning has not. We have programs which routinely beat all but world class human chess players, but none that can take a simple stroll through the woods. Why should this be? Several researchers have pointed out that from an evolutionary point of view, peripheral intelligence is much more basic and has been under development for a much longer time. Thus these abilities constitute the core of intelligent behaviour, whereas the high level skills exemplified in abstract reasoning have a relatively minor role to play even in most human activity. From this point of view, the investigation of infra-human levels of intelligence is not only interesting and important in its own right, but ultimately necessary for understanding of the human level.

The ascendancy of commonsense knowledge intimates a divergence from anthropocentrism, and the growth of research into peripheral intelligence confirms it. The original anthropocentric premise is challenged again by the twin realisations that the great bulk of intelligent behaviour depends on aspects of intelligence common to us and infra-human intelligences, and that these shared aspects of intelligence are much more complex and difficult to understand and to reproduce than the uniquely human ones. Some practical implications of this emerging non-anthropocentric conception of intelligence may be easily imagined. But what are its ethical implications? These can best be seen against the background of Joseph Weizenbaum's ethically motivated critique of AI.

In Computer Power and Human Reason, Weizenbaum argues that scientific modes of thinking are dehumanising, and represents AI as their culmination. Dehumanisation occurs when someone is treated merely as an object rather than as a person. This objectification is caused by the "imperialism of instrumental reason" — reason which considers everything as a technical problem to be solved rather than as a conflict to be adjudicated. It is dehumanising because it reduces rationality to the calculation of how to attain pre-given goals, and covers up the human capacity to choose those goals.

Emphasis on instrumental reason fosters the view that science is objective and value neutral, and thus absolves scientists of accountability for its ends and results. This attitude is responsible for everything from the technologically mediated atrocities of the Holocaust to the anticipated global catastrophes of genetic engineering. AI promotes the imperialism of instrumental reason because it makes abstract reasoning and problem solving central to intelligence, thus locating instrumental reason at the heart of each individual's thinking, rather than merely in optional practices like experimental science.

Weizenbaum's proposal for counteracting this progressive dehumanisation is twofold. On the one hand, he says we must acknowledge the radical autonomy of the individual to choose goals which shape the world. On the other hand, he appeals to extra-scientific value systems, such as religion. There are internal difficulties with this proposal, but there is also a deeper reason for dissatisfaction — it is itself anthropocentric. Instrumental reason encourages us to think of people as objects rather than persons. So the salvation of the world, Weizenbaum says, depends on our treating other people as fully human — dehumanisation calls for re-humanisation. This betrays an ethical anthropocentrism on Weizenbaum's part, a counterpart to the epistemological anthropocentrism of early AI.

But anthropocentrism may not be a viable foundation for ethics. As the destruction of the natural world has proceeded, it has become evident to many that the fate of humanity is bound up with the fate of the planet and its other inhabitants. From this point of view, the problem is not primarily that we treat people as objects, but that we treat other animals in particular and nature in general as mere objects, as a sort of neutral stuff to be manipulated to satisfy human interests rather than as having value and existence in its own right. Treating other people as objects is only a special and extreme case. The failure (or refusal) to recognise this has been explicitly blamed on a deep- rooted anthropocentrism in Western culture, as pervasive in our religious and ethical thinking as it is in our science.

The result has been a new insistence by some ethical thinkers on a non-anthropocentric ethics, an insistence that we have direct ethical obligations not just to humans, but to members of other species and to nature. But we cannot stop treating non-human creatures as mere objects by re-humanising them, because they are not human in the first place. Instead, our ethical obligations to them must be grounded in a respect for the non-human rooted in a belief system which acknowledges our essential relatedness to it, and rejects the notion of inherent human superiority.

But with regard specifically to intelligence, this is precisely the set of beliefs promoted by the corresponding non-anthropocentric turn in AI. As we have seen, it emphasises the commonality of human and infra-human intelligence on the one hand, and on the other hand it inspires a respect for infra-human aspects of intelligence by illuminating their immense sophistication and complexity as against the relative simplicity and mechancial reproducibility of the uniquely human aspects. So this new conception of the status of the infra-human in the space of possible intelligences aids and abets recent ethical thinking about the status of the infra-human in the space of ethical obligations. The ethical import of AI may thus in the long run be positive rather than negative.

Beth Preston, The Higher Education Supplement, July 24th 1992, p. 17.

Footnotes

1. Once outside the psychologist's laboratory, this claim needs a little care. Baron refers to an experiment in which an undergraduate learnt to memorize strings of numbers; the subject's memory span increased by extensive practice from 7 to about 79 digits, but his span for letters remained unchanged. In the real world, however, a fairly specific skill like driving a motor car is learnt perhaps by driving a particular Toyota but generalizes to other versions of that model, to other model Toyotas, to several other makes of car, possibly to tractors, and other types of vehicle,.... But I agree it won't help much with a jumbo jet.

2. It ought not to need saying but the well known remark that "intelligence is what intelligence tests measure" should be recognized as an absurd claim. I can invent any crazy questions and call them a test of X and then say that X is what my X test measures — people ought to lock me up unless there is good reason to think the questions I find do in fact relate in some reliable way to what the rest of us mean by X. (Perhaps we should note here that a measuring instrument does not have to be a direct measuring instrument in the sense that it responds directly to changes in X if it is an X-measure; it may respond to changes in Y which we reliably believe to reflect changes in X.)

3. And, as Jensen argues in his contribution to the same discussion I have been quoting from, between such scores and apparently non-intellectual matters such as response speed in pressing a button; "high achievers" may just have a faster brain.

Return to Steps

URL http://www.uwichill.edu.bb/bnccde/epb/stepsintel.html

Return to list of publications.