HUMAN

EVOLUTION

H.C. Plotkin University College London

Key words: intelligence,

evolutionaryepistemology, hierarchy theory, adaptational theory

Vol.

3 - N.

6 (437-448) - 1988

Intelligence and Evolutionary Epistemology InteLligenceis considered as one level in a multiple level model of evolution, each level being defined by its own units of selection and by the capacity to furnish adaptations to environmental events that occur within particular frequencybands. The specific implications for intelligenceare that it is never an across-species generalist capacity but always focussed around species-typical adaptive behaviours; and that it interacts with evolution at other levels, notably at the socioculturallevel in the case of Humans.

The principal claim of evolutionary epistemology is that knowledge is a problem in biology. As the science of knowledge, it makes several subsidiary claims with respect to individual intelligence. These include the following: that intelligence is a subset of the processes by which animals gain and utilize knowledge about themselves and their worlds; that intelligence is related to other processes of knowledge acquisition and utilization, including individual development and culture, by a complex hierarchical structuring; that all forms of knowledge gain and utilization, including intelligence, share an identical selectional process; and that intelligence is not only a consequence of evolution but also an active cause in the evolutionary process. The following is a brief exposition of this view and some of its implications.

Some Basic Evolutionary Epistemology CAMPBELL'S (1974) authoritative review has led to a burgeoning literature in evolutionary epistemology (see for examples BRADm, 1986; CALLEBAUT& PmXTEN, 1987; RIEDL & WUKETITS, 1987). There is a strong philosophical contingent working in this area. What will be described here, however, is restricted to evolutionary epistemology as an approach to biology, and specifically to issues that bear on the problem of intelligence. The following is a necessarily restricted outline of the position that is detailed elsewhere (PLOTKIN, 1982, 1987; PLOTKIN & ODLING-SMEE,1979, 1981, 1982). Consider first the place of evolutionary epistemology in the context of other approaches to evolutionary theory. The theory of evolution has changed significantly a number of times over the last two centuries. There is no reason to believe that the currently most widely accepted version, neo-Darwinism, which is based on the notion that evolution occurs through changes in gene frequencies in breeding populations wrought by natural selection, is going to prove any more resistant to change than previous accounts. There is no one <~correct~>evolutionary theory. The nature of the theory that wins out at any time in the history of any science, or within a restricted group of scientists at a particular moment, is determined by its scope and the extent to which it encompasses phenomena that are held to be crucial at that time or by that group, including newly discovered or reinterpreted empirical findings. Evolutionary epistemology is an approach to evolution that has been fostered by biologists with a particular interest in cognition, intelligence and culture, and who feel that these are important phenomena for which no proper place has yet been found in

9 Editrice I1 Sedicesimo - Firenze

ISSN 0393-9375

438

PLOTKIN

evolutionary theory. It is cognitive science's way of placing individual mentation within a wider evolutionary framework. The best possible outcome that might be expected is an enrichment both of evolutionary theory and our understanding of cognition and culture. For the purposes of this essay, I will argue the position through two main points. The first, already mentioned, is that the process by which knowledge or information is gained is universal and is a process of evolution. The second is that living things are complex, hierarchically organized systems, each level of which is defined by the presence of a knowledge gaining or evolutionary process.

The Evolutionary Analogy In 1880 William James wrote an essay on <~the remarkable parallel.. [that] ... obtains between the facts of social evolution and the mental growth of the race, on the one hand, and of zoological evolution, as expounded by Mr. Darwin, on the other>> (pp. 441). James provided a lengthy argument on the similarities between the Darwinian conception of evolution as the outcome of natural selection acting on variant forms of a species, and the creative thinking of <~great mere>, a manifestation, surely by any account, of intelligence. He concluded that <~...new conceptions, emotions and active tendencies which evolve are originally produced in the shape of random images, fancies, accidental outbursts of spontaneous variation in the functional activity of the excessively unstable human brain, which the outer environment...selects...just as it selects morphological and social variations due to molecular accidents of an analogous sort~> (pp. 456). The specific form of the analogy that James used is not at issue here. The point to be made is the general nature of the argument, which was that whatever the processes are that account for 'biological' evolution (adaptation and speciation), these same processes may help in understanding complex cognitive phenomena, such as creative thinking in man. This is what is meant by the phrase 'the evolutionary analogy'. James was not the first to use the analogy. Spencer had adopted a preDarwinian version of it before the publication of The Origin o/Species. The use of Darwinian concepts in the analogy by T.H.Huxley, who drew the parallel between the struggle for existence between organisms and the struggle for existence within an organism of its developing parts, appeared over a decade before the James essay. But it was James who first used the analogy in its Darwinian form as the basis for an extensive analysis of thought and problem solving. Subsequently, scores of writers have employed it in this same context, including such diverse figures as Simmel, Baldwin, Piaget, Popper, Lorenz, Skinner, Simon and Campbell. It is an error to consider the analogy a recondite intellectual thread running barely detectably through the writings of a restricted and obscure group of theorists. In general, the analogy can be used literally. It is then purely a heuristic device by which what is known of the process of 'biological' evolution is used to illuminate and further our understanding of cognitive processes and cultural evolution by pointing to similarities and conformities, and suggesting further lines of analysis and experimentation. Or the analogy can be hardened into an assumption of identity of process. For example, instead of saying that trial-and-error learning is like evolution in this and that respect, one says that trial-and- error learning and evolution are the product of identical processes. This, too, acts as a heuristic device by which analogies and disanalogies between the various applications of the process can be established. However, it is also an assertion that all knowledge gain, which is the establishment of matching relationships between features of environmental order and aspects of organismic organization, contains at its core certain

INTELLIGENCE AND EVOLUTIONARY EPISTEMOLOGY

439

nontrivial, identical features of process, no matter how different the mechanisms embodying these might be, and no matter how different the forms of knowledge might seem. It is this strong form of the analogy, the assumption of identity of process, that is followed here.

What then is this process? More correctly, it is a set of processes that can be described by a simple routine. Variants are generated (genes and genotypes in the case of 'biological' evolution), they are tested (by natural selection to stay with the case of 'biological' evolution), and the successful variants that have passed the test phase are regenerated (the further propagation of genes and genotypes). In the cognitive domain the variants that are generated are responses, associations, map references, inferences, solutions, hypotheses etc., depending upon the precise nature of the subset of cognitive processes being described. The test will be contiguity, excellence of fit to some deductive template, or some other form of match to an expectation or a forecast, depending again upon the nature of the cognition. Regeneration will take the form of the generation of further variants, some identical with those selected in the past and some entirely novel (cognitive mutations, in effect). The exact mix of old and new variants at the regenerate phase will vary again with the form of cognitive process being considered. In simple associative learning, for example, it will be very conservative with a low number of novel variants. Complex problem solving will have a much higher proportion of novel variants at the regenerate phase, which allows for what is commonly described as ~creativity~>: Not only can the analogy be used in this way in the cognitive domain of humans and other animals, but it can also be extended to artificial intelligence, and hence proffered as a general characteristic of intelligence. DENNETT (1981) makes this point well: <~... if AI is truly the study of all possible modes of intelligence, and if generate-and-test is truly a necessary feature of AI learning programs, then generate-and-test is a necessary feature of all modes of learning, and hence a necessary principle in any adequate psychological theory.~ (P 180) Dennett's <~generate-and-test~, of course, is his shorthand description of a form of the evolutionary analogy. It is clear from his use of the analogy, as well as from the previous paragraphs, that the evolutionary analogy is not just applicable to ~simple)) forms of learning such as trial and error. The processes of generate, test and regenerate are, it is assumed, essential to all forms of knowledge gain. There is one further point that must be made about the analogy in its strong form. It follows from the assumption of identity of process that the analogy can be run in either direction. We can not only understand more about cognition by applying what we know about evolution, but the reverse holds as well. That is, we should be able to use an increasing understanding of the complexities of cognition to further knowledge about 'biological' evolution. Piaget was a powerful exponent of the use of the analogy ~dn reverse>>.

A hierarchy of processes for gaining knowledge A criticism levelled from a number of different sources in recent years against orthodox neo-Darwinism is that it is limited in scope to explaining shifting gene frequencies in the gene pools of breeding populations as these are caused by the effects of natural selection on the adaptive attributes of phenotypes. Apart from obvious claims as to the adaptive significance of attributes such as those in the psychological or sociocultural realm, the theory gives no substantial guidance when it comes to giving a causal explanation of events at the psychological or sociocultural level. It must be acknowledged that evolutionists like WILLIAMS (1986) would argue that that is not the role of a theory of

440

PLOTKIN

evolution, but an alternative is to resist the limited claims of the current orthodoxy and somehow expand the theory. The most common way of doing this in recent years has been to use the concept of hierarchical organization (ELDRID6E& SALTHE, 1985). This use of hierachical notions by ecologists, systematists and developmentalists as a means of making evolutionary theory directly available to them accords well with evolutionary epistemology which, in its most recent form, has been cast in a hierarchical framework. CAMPBELL'S (1974) scheme, for example, envisages at least ten levels extending from the genetical to learning, thought and science. Here I will consider a four-level hierarchy. Both schemes are multiple-level models of evolution. The central notion of the model is that there is more than one evolutionary process; that each process defines a separate level in a hierarchical system; and that each is embodied in mechanisms that are sensitive to different frequencies of environmental change. The four levels of the model are the genetic, the developmental, the individual learner and the sociocultural level. The model is based on two fundamental assumptions. The first is that the hierarchy gains knowledge, and that that knowledge is an important causal component of the form that adaptations take. The second is that the original, and hence most fundamental, level of knowledge gain and storage is the genetic. However, the rate at which this level can furnish adaptations to the phenotype is restricted temporally by the rate of generation turnover. That is, it is sensitive to events of only low frequency relative to the rate of generation turnover time. Changes of higher frequency cannot be detected by this level; and if such changes are significant in detrimentally affecting the fitness of an organism, then they become the selection pressures for the evolution of the subsidiary levels of the hierarchy, each of which is sensitive to higher frequencies of change. Each level of the hierarchy is thus a tracking device that is tuned to ever higher frequencies of change. For example, level three is able to gain knowledge of events that occur rapidly. Level four operates even more quickly. It takes longer to learn through one's own direct interaction with the environment (a level three process) than it takes to acquire the same information from a conspecific (a level four process). The reason for these levels having the capacity for operating at increasing rates requires some amplification. The rate at which knowledge is gained in terms of the selection and propagation of adaptive phenotypic attributes that are wholly genetically determined is fixed absolutely by the parameter of generation time. Other variables may, and do, alter evolutionary rates, but none of these can make evolution at level one go faster than some upper limit that is set by that parameter. Levels two and three, by definition, comprise knowledge gaining and storing processes that, because they work by way of within-organism and not between-organism evolutionary processes, are freed from the limitation of generation time and they can furnish adaptations at higher rates. At level four the existence of non-genetic information transmission between individuals frees it from the genetic generation time limitations of level one. Furthermore, although individual and social learning likely occur at similar rates, if learning processes are evolutionary processes, the central tenet of the model, then the rate at which a process of evolution can establish adaptations is not just determined by the frequency of variant generation, but also by the variables that Wmson (1985) refers to as comprising the ~basic equation of evolutiom~. This is that <~the rate of evolution within a population equals the number of mutations arising per unit of time multiplied by the fraction of those mutations destined to be fixe&~ (p. 155). Wilson's assertion can he applied not only to level one evolution, but to that of'any level, including level four. At that level, knowledge is being gained and stored as a result of a dynamic interaction between more than one learner. Thus, in unit time, level four gives rise to more variant generation (and hence more mutations) than

INTELLIGENCE AND EVOLUTIONARY EPISTEMOLOGY

441

level three - - an animal that learns by observing the learned behaviour of another animal, is adding its own level three variants to that of another. Furthermore, the learning of the <~other)> animal has already been subjected to selection, and this selection of already selected variants will increase fixation rates as well. Thus, despite both levels requiring level three processes, the overall rates of evolution at level four are higher than those of level three because of increased variant generation and mutation fixation rates that arise when two individual learners are sharing information that has itself been acquired by level three learning. Given these four basic levels, within each level there are many possible forms of variants, and hence many possible ways for gaining and storing knowledge. This is a commonplace understanding. At the genetical level of a species, there are large numbers of genes and many genes occur in different allelic forms. At the developmental level, many possible developmental trajectories are potentially available to an organism, the ones being selected being a function of that organism's genotype and the nature of the environment in which development occurs. At level three, the cognitive level, there are many forms of learning and within each form, say simple associative learning, there are many kinds of separate associations that can be stored. And at level four, the socio-cultural level, the common assumption is of a very rich network of knowledge when this level is present in a species like man. Across these levels there exist many possible combinations of the variants of each level. I will refer to the vertical organization of the model as <~modules~. Each module is a hierarchically organized set of knowledge gaining processes. These modules may partially overlap, i.e. share variants, at some levels, and be independent at others. Thus no assumption is made about the independence of modules. All that this means, for example, is that two level three learning mechanisms (motor skills and spatial mapping, say) may share certain genetic and developmental bases. Not all modules extend over all levels, but as is implicit in the hierarchical nature of the model, all must cover at least two levels. Some will extend across three levels, and in rare instances all four levels are covered. Whether a hierarchically organized module of knowledge gaining processes can exist independently of the genetical level is an important point of difference between various approaches to cognition and intelligence: The model adopted here assumes that the genetical level is always implicated. The conception, then, is of many possible modules operating across various numbers of levels. The constant feature is that all variant-generating mechanisms at each level gain knowledge by the same generate-test-regenerate routine, albeit with differing mixes of old and new variants at the regenerate phase. Thus, for example, the developmen~ of orientation receptors in the visual system of a cat is the product of knowledge gaine/clat both first (genetic) and second (developmental) levels. That same cat, which learn~ the spatial position of a resource, will be using knowledge processes at three levels, th~third level taking the form of a spatial learning device. If and when that cat enters info social alliances with other cats such that it is able to gain knowledge of the world through the prior experience of those conspecifics, four levels are in action. The crucial difference between the third and fourth levels is that the former is a within-animal knowledge gaining device, and the precise knowledge gained cannot be transmitted to others. All that can be transmitted to others, genetically, is the adequacy of the device. The fourth level, on the other hand, is defined by the extragenetic transmission of knowledge between phenotypes - - they learn from each other and the knowledge gained may be quite specific. A hierarchy of evolutionary processes that gain and store knowledge about the organism and its world, has some interesting characteristics. For one, it directly brings into the purview of evolutionary theory classes of phenomena previously left out. Deve-

442

PLOTI~N

lopment is the classic example of a central area of biology that, until recently, has been conceptually separated from evolution by most evolutionary biologists. The model presented here is just one of a number of attempts that ~allows us to analyse a very simple array of separate but interacting causes constituting an hypothesis of a comprehensive evolutionary mecbanism~ that allows for exploration of <~the possibility that asymmetries in the introduction of variation at the focal level of individual phenotypes, arising from the inherent properties of developing systems, constitutes a powerful source of causation in evolutionary change~ (THOMSON, 1986, pp. 221 and 222). I go further in including level three (individual cognition) and level four (socioculture) as additional and significant sources of variation in evolving systems. A second feature of this approach is that it provides an explanation for behavioural and psychological predispositions which, whilst well recognized, have been difficult to understand in the light of other theories of learning and cognition. What a hierarchical model does is provide a framework for understanding how knowledge, which appears as a priori at one level, has been gained a posteriori at another level. Third, and closely related to the issue of psychological predispositions, it begins to resolve the age-old naturenurture issue, arguably the central problem for the social sciences. It does this by replacing ~dnteractionism)~ with the conceptualization of hierarchically nested processes. This eliminates the seeming paradox of the innate capacity to learn only certain things. Fourth, because the scheme allows adaptive load to be shared across levels, the burden of explanation for any and all adaptations need not be placed at just one level. Finally, evolutionary theory becomes more complicated, and more elegant as well, as an explanation suited to the complexity of living systems. The notion of cause is no longer restricted to just one (genetic) level with selection acting as a filter at only one other level. Each level is defined by the existence of an evolutionary process; each level has its own selection device and its own unit of selection; and each level has a degree of causal autonomy relative to all other levels.

Intelligence and Evolution The implications of the above for understanding the biology of intelligence require an additional, linking, assumption and then the spelling out of some of the rules by which the scheme might work. The assumption is that
Some Rules by which the model works The first rule is an optimizing assumption or trade-off rule common to many approaches to adaptation, viz., the maximizing of benefits and the minimizing of costs. We seek a mix of fast, reliable and cheap knowledge gain and utilization. For the hierarchy of knowledge gaining and storing processes just described, this takes the form of a trade-off between (t) levels tracking at the highest frequency necessary for a particular interaction between the organism and its environment, (2) those levels least prone to

INTELLIGENCE AND EVOLUTIONARY EPISTEMOLOGY

443

sampling errors and (3) those levels least subject to spontaneous changes of state of storage over the time span for which storage is required, and (4) those levels which are metabolicaUy least expensive. The levels that provide the maximum trade-off of these four factors will be the levels that gain the knowledge for the formation of the behavioural adaptation that the particular organism-environment interaction requires. The second rule is specific to the multiple level model. It is that the dimension of increasing frequency of change to which the levels are tuned corresponds to a dimension of increasing proneness to errors of sampling and to spontaneous errors of store, and also to increasing metabolic cost. This rule means that although the genetic level is the slowest level of knowledge gain, it is the least subject to errors, the nature of the store is reliable relative to storage-time requirements, and it is metabolically cheap. The higher levels in the hierarchy operate faster, are more prone to errors of sampling and storage, and they involve the metabolically expensive central nervous system. Also, the greater the amount of knowledge that must be gained at these high frequencies, the larger that organ system must be. These first two rules relate to the vertical distribution of knowledge gain across the levels of the hierarchy. The third rule is concerned with the horizontal or within-level, organization of the devices that generate the variants making up the modules of the hierarchy. It is a rule of inertia which asserts that the modules that have already evolved, and the variant generators of which they are composed, will be used for as many forms of knowledge gain as possible. New forms of variant generators and selectors, or new combinations of these comprising new modules, will evolve only if the costs of additional variance generators and selectors are consistently outweighed by benefits in the form of more efficient behavioural adaptations.

Some implications o/the rules [or intelligence Sharing knowledge across levels According to the model, intelligence is knowledge gain and utilization at the third level of the hierarchy. However, the interpretation of the model given here implicates the first and second levels in every instance of intelligent behaviour. More than that, rules one and two mean that the levels that will gains and stores the knowledge for the construction of adaptations, behavioural adaptations in this specific case, will always be the most fundamental possible. In other words, if an adaptation can be constructed by just the first and second levels, then it will be. This is, in principle, a firm empirical prediction, even if it is not original with this approach. G.C. WILLIAMS, working on <. The prediction is firm only in principle because, for the most part, we do not know how to partition the world, and hence changes in that world, relative to an organism that is interacting with that world, For the moment, therefore, the calculations must be made at an intuitive level, but they are not difficult to understand. Climatic changes, for example, if occurring at a low frequency (such as ice ages) will lead to adaptations resulting from knowledge gained at just the first, genetic, level with possible minor tuning by the second, developmental, level. This is because all organisms have a generational turnover time that is very much less than the periodicity of such major climatic events. The periodicity of lesser seasonal changes, however, raises the ratio of the frequency of environmental changes to generational turnover time, and hence spreads more evenly the burden of the load of knowledge gain across first and second levels. Many

444

PLOTKIN

insect species, for example, show facultative polymorphisms as a function of environmental conditions during development (BARraNCTON, 1979). Such developmental effects occur because the conditions eliciting them are environmental fluctuations with a frequency higher than can be detected by the first level acting alone. All of this may seem remote from intelligence, but it leads to the empirically testable assertion that intelligence, even once evolved, will not gain knowledge necessary for the adaptive solution to a problem posed by a changing world if those changes occur at a frequency that is detectable by the first and second levels. But why, if the third, cognitive, level already exists, should it not operate to acquire knowledge about changes occurring at any frequencies lower than its upper operating limit? The answer lies partly in the rules given above as they operate in systems over long periods of time. More fundamental levels are cheaper and more reliable. However, there is also an answer in the immediate operation of a system that has evolved level three. This is that the third level cannot detect changes occurring at very low frequencies. The inability of a high frequency operating level to detect events occurring at frequencies lower than the operating characteristics of that level is probably a sufficiently general condition for it to be stated as a fourth rule governing the model - - in this case, a rule concerned specifically with the gaining of knowledge once level three has already evolved. Level four, however, may not be governed by this rule because its < in man provides the ability to detect changes of very low frequency. The successive nature of ice-ages is known to science alone. Climatic effects seem reasonably easy to handle at a common sense level within the framework of the rules by which the model is held to operate. There are, however, many other forms of change that present us with real difficulties as to how to partition organism-environment interactions in terms of rates of change and the levels that operate to gain knowledge about such change. Prominent here are events that are <~. Simple locomotion is a good example of such difficulty. An animal that alters its position in space in a non-uniform environment is generating change that may be significant to its survival. In some cases, level one tracking devices manifested as reflexes, kineses, etc., will supply the appropriate adaptive behaviours. In others, a memory for spatial location operating at level three seems to be required. Furthermore, interactions with other forms of change, e.g. changing climatic events induced by migration, may pose very difficult problems in terms of the frequencies at which the organism is experiencing change. These are technical problems for the model and present difficulties for its empirical testing. Is intelligence a general or a species-typical trait?

The model of a hierarchy of knowledge gaining and storing devices explains what is known as the constraints on learning at both a general (i.e., constraints will always exist) and specific (i.e. the nature of the constraints) levels. Consider the well-known example of nausea conditioning in the laboratory rat, a form of learning that shows certain characteristics that are different from other forms of conditioning in the same animal. One of the curious features of this learning is that it appears to be constrained by the a priori knowledge that it is the taste and smell of food that is to be associated with the illness, and not where the food is eaten. How can this be explained? In terms of our model,the relationship [food ~ poisoning ~ nausea and other symptoms of illness] is one that changes only very slowly. The knowledge of those relationships would be supplied by the first, genetic, level. The first level cannot, however, detect the relationships [which

INTELLIGENCE AND EVOLUTIONARY EPISTEMOLOGY

445

specific foods ~ what specific tastes --} illness]. Here the changes governing these relationships occur at a higher frequency than can be detected by level one, or even level two. If food poisoning contributes significantly to the fitness of certain kinds of animals, then knowledge of that specific, relatively high frequency relationship of taste, food and illness can only be acquired by a knowledge gaining device that can operate at the appropriate frequency, which in this instance is level three. We therefore expect the total behavioural adaptation to be partitioned across levels: [food may lead to illness] is knowledge gained and stored at the first level in the form of instructions for <>the central nervous system such that only the tastes of food can be easily associated with nausea; the wiring, however, is in the form of an < which means that it can acquire the more specific information [the salient characteristics of the particular foods that lead to illness] by the operation of level three. The multiple level model not only explains the constraints on learning within the framework of a general theory of adaptation, hut it predicts that all level three knowledge gain is nested under, and hence constrained by, levels one and two. According to the model, tabula rasa learning is impossible because the third level evolved out of a failure of the first and second levels to deal with changes occurring above a certain frequency; and having evolved out of this failure of levels one and two, it is rooted within and constrained by those more fundamental levels of the hierarchy. There is an obvious counter scenario on the evolution and operation of intelligence. This is that even if intelligence has its origins within the kind of explanatory framework presented here - - or however it might have evolved - - once in existence it may become detached not only from its original function but from any specific function at all: it is an all-purpose device for gaining knowledge of any events that occur with a frequency less than some upper, and more than some lower, limit. Furthermore, the argument goes, what then determines the differences between different species within major taxonomic groups all of whose constituent species share this free-floating intelligence, are certain ~contextual variables~ such as differences in perception, motor capacity, motivation and so on. This view is presented in extreme form by MhCPH~L (1985), who argues that there are no differences in intelligence between vertebrate species, with the exception of man. This is not an issue that is likely to be settled empirically because of the way the case has been framed. For this reason no attention will be paid to the literature that claims to present empirically validated differences in intelligence between different species of vertebrate. We need to consider the essence of the case itself. Take the extreme instance of a chimpanzee and goldfish comparison. In order to prove a difference in intelligence between these two species, we have to devise test procedures that do not advantage the chimpanzee and disadvantage the goldfish because ~many of the most impressive achievements of the chimpanzee have involved the use of its limbs, and this poses for comparative psychologists the challenge of devising for vertebrates without comparable limbs, tasks whose formal intellectual demands parallel those made inthe tasks mastered by chimpanzees~ (MACVHan, 1985, 46). This passage highlights the difference with the position adopted in this paper. The multiple level model of knowledge gaining processes considers the manual dexterity of chimpanzees not as a contextual variable but as one of the primary sources of chimpanzee intelligence. The argument goes as follows: chimpanzees have a high degree of manual dexterity. This allows them to manipulate and change their immediate environment at a high rate. The knowledge gaining processes of levels one and two, which gave rise to the manual dexterity in the first place, cannot track the rates of change which that selfsame trait generates. The chimpanzee, therefore, has evolved modules of knowledge gain that extend to the third level and hence can operate at the

446

PLOTKIN

required frequency - - but the changes that are being assimilated and accommodated to are specific to hand-use. Because of such dexterity, some of the intelligence of the chimpanzee is concentrated around that skill. In the language of part one of this paper, the chimpanzee has a nervous system with a high concentration of variance generation in the sphere of manual dexterity. Thus, if one removes the possibility of manual-dexterous-related variance generation from the chimpanzee, it may indeed appear to be less intelligent. This will be because it is less intelligent. But under normal circumstances manual dexterity, (more correctly manual-visual dexterity) is an important part of that animal's interaction with its world. It is not appropriate to label it a contextual variable. This line of reasoning applies also to man, whose intellectual differences from other vertebrates Macphail does allow. A considerable amount of human intelligence is related to language. This is presumably because the human brain is a prodigious generator of variants in the linguistic sphere. If this is a correct argument for humans, then it must also be correct for chimpanzees. Indeed, it must generalize to the specialized adaptations of all animals. The foci of intelligence in any intelligent species are the specialized forms of interaction with, and adaptation to, its environment that that species has evolved. This position is the polar opposite of the notion that intelligence is some kind of unattached adaptive potential in all intelligent species. This line of argument does not, however, confine intelligence only and directly to its original source of adaptation. A translation from the original source of variants to a different, perhaps higher order, knowledge form is not ruled out by the model. Thus the visual skills of the monkey might form the basis of volume conservation (e.g., PASNAK, 1979), or the manipulatory skills of the chimpanzee might lead to analogical reasoning (GmLAN, PRErvIACK& WOODRUFf, 1981). Nonetheless, the ~original~ variant sources will lie elsewhere, and this implication is again empirically accessible. Developmental studies are an obvious first choice for empirical verification. It is likely that synergistic relationships hold between all level one and two processes of knowledge gain and the level three processes of intelligence that they give rise to - - for example, manual dexterity and the intelligent skills directly related to such dexterity. This notion is less easily tested empirically. But on one point the model is absolutely clear. Level three is an evolutionary process requiring the generation of variants. Those variants cannot be generated in a vacuum. They must be variants of something. Nor can they be so multifaceted as to be adaptable to any adaptive behavioural demand made on an animal. To the contrary, variants are specific to pre-existing level one and level two adaptations. The model admits of no tabula rasa intelligence.

Intelligence and human evolution Intelligence, as depicted here, is a form of evolution. It has, of course, also evolved. The relationship of level three to levels one and two is one of nested subordinacy - - level three (intelligence) must act in support of level one (genetical) evolution in the long term. Ir need not in the short term. But over long periods of time, maladaptive intelligence will be selected against. Because level three operates to gain knowledge of events occurring with a relatively high frequency and hence is the source of behavioural adaptations to rapid environmental fluctuations, the overall adaptive competence of intelligent creatures is increased, and this may extend the range of niches which they can occupy. Extended niches must mean that novel selection pressures are confronted, and such novel selection may exploit previously unexploited and unselected genetical variants. The net result of the nesting of these evolutionary processes is that the overall rates of evolution, both at levels

INTELLIGENCE AND EVOLUTIONARY EPISTEMOLOGY

447

one and three, of an intelligent species must be increased. This is a version of WADDINGexploitive system, i.e., the possession by certain species of animals of cognitive mechanisms whose causal status in evolution is co-equal with that of genetical, natural selective and developmental systems. It is also an empirically testable assertion, work on which is reported by WYLES et al. (1983). They demonstrated a correlation between brain size, which they took to be a measure of the capacity for hehavioural innovation together with the social propagation of acquired information, and rates of anatomical evolution in certain land vertebrates. The same reasoning should apply to man as to other intelligent species. This is especially so given that man is not merely an intelligent animal hut one that has evolved to a unique degree the fourth (sociocultural) level of evolution. As in the case of the other levels, the fourth is subordinate to more fundamental levels. But unlike level three, the rapidity with which knowledge is gained and utilized at level four is a consequence of a much greater degree of decoupling between level four and level one - - level four by-passes level one transmission routes. Just as other levels mutually enrich the potential for each to evolve further, so too does level four (sociocultural evolution). Its special characteristics, however, make a difference in two ways. The one is that whilst a synergistic relationship may well operate between levels one and four with level four r to new selection pressures previously unselected genetical variants, it may well evolve compensatory shifts at level four before level one is implicated. The other concerns the source of variants of level four, and which gives level four the high rate of evolution relative to level three ascribed to it in the first part of this paper. Consider again the notion that the environment can be banded in terms of frequency of change into s e t s of features, E a, E~ and E v, where E ~ are features of environmental order that change very slowly, Ea are features that change more rapidly, and Ev are events with high rates of change (remember that these frequencies are relative to certain important aspects of the life histories of individual phenotypes that characterize any species; notably longevity, generational turnover time and duration of ontogeny). Variant sets V~, Va , and Vv are generated at matching rates by level one (genetic), level two (developmental) and level three (intelligence) evolutionary processes such that some limited number of variants are selected from each level and enter into adaptive relationships E~-V% E~-V~, and E~-V~. In each case, the variants that are generated do not become features of the environment for other levels in the hierarchy. Thus no V = (genetical variants) enter into E~; and no V~ (developmental pathways) enter into E~. This partitioning of variants at one level from environmental events at another does not, however, hold for levels three and four. Some Vv variants become the environmental features, EV~ , that enter into an adaptive match with variants Vc, to form EV~-V~. That is, sociocultural evolution involves the products of intelligence of one individual as part of the environment that has to be adapted to by another intelligent individual. It is this relationship of two, or more, intelligent individuals to one another - - where the intelligence of one becomes t h e <~environmenb> of the other - - that provides levels three and four with such dynamic characteristics when compared to level three acting alone. It is not difficult to maintain the distinction between levels three and four when one is comparing two species, one of which has only evolved level three but not level four, and the other has evolved both. But that distinction is less easy to maintain, and is very poorly understood within a species, like man, that has evolved both. There are no criteria available by which it is possible to know which Vv can become EV~, and those Vv that TON'S (1959)

448

PLOTKIN

cannot become EV~. For example, it is conceivable that the individuals of one species may acquire via social learning information as to the position of a food resource, but cannot, except by individual learning, acquire knowledge as to whether that resource is toxic or not. In such a case, the domain of Vv ~ EV~ is rather small. In man there is a n ability to share a much wider range of knowledge, i.e., the domain of Vv ~ EV~ is very large. This is a significant species characteristic of humans. As the domain of shared variants at levels three and four increases, so must the synergism between them. Increases in one will be reflected in increases in the other, a relationship that must express itself as an increase in neural information processing capacity. Here, perhaps, lies the key to understanding the rapid evolution of brain size and intelligence in the genus H o m o .

References BARRINGTONE.J.W.,1979. Invertebrate Structure and Function. Nelson, London. Br,ADIE M., 1986. Assessing evolutionary epistemology. Biology and Philosophy, 1: 401-459. CALLEBAUTW. & PmXTEN R. (eds.), 1987. Evolutionary Epistemology; a Multiparadigm Program. Reidl, Dordrecht. CAMPBELLD.T., 1974. Evolutionary epistemology. In: P.A. Schilpp (ed.). The Philosophy of Karl Popper. Open Court Publishing, Chicago. DENNETT D.C., 1981. Why the law of ellect will not go away. In: Brainstorms, MIT Press, Cambridge Mass., pp. 71-89. ELDglDGEN. & SALTHES.N., 1985. Hierarchy and evolution. Oxford Surveys in Evolutionary Biology, 1: 184-208. GILLAND.J., PgEMACKD. & WOODRUFFG., 1981. Reasoning in the Chimpanzee: 1. Analogical reasoning. Journal of Experimental Psychology: Animal Behavior Processes, 7: 1-17. JAMES W., 1880. Great men, great thoughts, and the environment. Atlantic Monthly, 46: 441-449. MACPHAILE.M., 1985. Vertebrate intelligence: the null hypothesis. Philosophical Transactions of the Royal Society, London series B, 308, 37-51. PASNAK R., 1979. Acquisition o[ prerequisites to conservation by Macaques. Journal of Experimental Psychology: Animal Behavior Processes, 5: 194-210. PIAGET J., 1936. The Origin of Intelligence in the Child. Reprinted by Penguin, Harmondsworth. PLOTKmH.C., 1982. Evolutionary Epistemology and Evolutionary Theory. In: H.C. Plotkin ed. Learning, Development and Culture: Essays in Evolutionary Epistemology. Wiley, Chichester. PLOTKm H.C., 1987. Evolutionary Epistemology as Science. Biology and Philosophy, 2: 295-313. PLOTKIN H.C. & ODLING-SMEEF.J., 1979. Learning, change and evolution. Advances in the Study of Behaviour, 10: 1-41. PLOTKIn H.C. & ODLING-SMEEF.J., 1981. A multiple level model o[ evolution and its implications [or sociobiology. The Behavioral and Brain Sciences, 4: 225-268. PLOTKINH.C. & ODLmC-SMEEF.J., 1982. Learning in the context o[ a hierarchy of knowledge processes. In: H.C. Plotkin ed. Learning, Development and Culture: Essays in Evolutionary Epistemology. Wiley, Chichester. RIEDLR. & WUKETITSF.M. (eds.), 1987. Die Evolutiondre Enkenntnistheorie. Verlag Paul Parey, Berlin. THOMSONK.S., 1986. The relationship between development and evolution. Oxford Surveys in Evolutionary Biology, 2: 220-233. WADmNGTON C.H., 1959. Evolutionary systems: animal and human. Nature, 183: 634-1638. WILLIAMS G.C., 1966. Adaptation and Natural Selection. Princeton University Press, Princeton. WILLIAMSG.C., 1986. A deqenceof reductionism in evolutionary biology. Oxford Surveys in Evolutionary Biology, 2: 1-27. WILSOn A.C., 1985. The molecular basis of evolution. Scientific American, 253: 148-157. WYLESJ.S., KUNKELJ.G. & WILSOn A.C., 1983. Birds, behavior, and anatomical evolution. Proceedings of the National Academy of Sciences, 80: 4394-4397. Received: April 19, 1988. Accepted: May 28, 1988.