What Darwin Got Wrong

What Darwin Got Wrong


View All Available Formats & Editions
Choose Expedited Shipping at checkout for delivery by Tuesday, December 7


With a New Afterword

What Darwin Got Wrong is a remarkable book, one that dares to challenge the theory of natural selection as an explanation for how evolution works—a devastating critique not in the name of religion but in the name of good science. Combining the results of cutting-edge work in experimental biology with crystal-clear philosophical arguments, Fodor and Piattelli-Palmarini mount a reasoned and convincing assault on the central tenets of Darwin's account of the origin of species. This is a concise argument that will transform the debate about evolution and move us beyond the false dilemma of being either for natural selection or against science.

Related collections and offers

Product Details

ISBN-13: 9780312680664
Publisher: Picador
Publication date: 03/01/2011
Pages: 320
Sales rank: 745,184
Product dimensions: 5.40(w) x 8.20(h) x 1.00(d)

About the Author

Jerry Fodor is a professor of philosophy and cognitive science at Rutgers University.

Massimo Piattelli-Palmarini started his academic career as a biophysicist and molecular biologist and is now a professor of cognitive science at the University of Arizona.

Read an Excerpt







The (neo-)Darwinian theory of evolution (ET) has two distinct but related parts: there's a historical account of the genealogy of species (GS), and there's the theory of natural selection (NS). The main thesis of this book is that NS is irredeemably flawed. However, we have no quarrel to pick with the genealogy of species; it is perfectly possible -in fact, entirely likely-that GS is true even if NS is not. We are thus quite prepared to accept, at least for purposes of the discussion to follow, that most or all species are related by historical descent, perhaps by descent from a common primitive ancestor; and that, as a rule of thumb, the more similar the phenotypes of two species are,1 the less remote is the nearest ancestor that they have in common.2

However, although we take it that GS and NS are independent, we do not suppose that they are unconnected. Think of the GS as a tree (or perhaps a bush) that is composed of nodes and paths; each node represents a species, and each species is an ancestor of whatever nodes trace back to it. The questions now arise: How did the taxonomy of species get to be the way that it is? What determines which nodes there are and which paths there are between them? In particular, by what process does an ancestor species differentiate intoits descendants? These are the questions that Darwin's adaptationism purports to answer. The answer it proposes is that if, in the genealogical tree, node A traces back to node B, then species B arose from species A by a process of natural selection, and the path between the nodes corresponds to the operation of that process.

We will argue that it is pretty clear that this answer is not right; whatever NS is, it cannot be the mechanism that generates the historical taxonomy of species. Jared Diamond in his introduction to Mayr (2001, p. x) remarks that Darwin didn't just present '... a well-thought-out theory of evolution. Most importantly, he also proposed a theory of causation, the theory of natural selection.' Well, if we're right, that's exactly what Darwin did not do; or, if you prefer, Darwin did propose a causal mechanism for the process of speciation, but he got it wrong.

There are certain historical ironies in this because it is the Darwinian genealogy, and not the theory of natural selection, that has been the subject of so much political and theological controversy over the last hundred years or so. To put it crudely, what people who do not like Darwinism have mostly objected to is the implication that there's a baboon in their family tree; more precisely, they do not admit to a (recent) ancestor that they and the baboon have in common. Accordingly, the question doesn't arise for them how the ancestral ape evolved into us on the one hand and baboons on the other. This book is anti-Darwinist, but (to repeat) it is not that kind of anti-Darwinist. It is quite prepared to swallow whole both the baboon and the ancestral ape, but not the thesis that NS is the mechanism of speciation.

The argument for the conclusion that there is something wrong with NS is actually quite straightforward; to some extent, it's even familiar. Not, however, from discussions of Darwinism per se, but from issues that arise in such adjacent fields as the metaphysics of reference, the status of biological teleology and, above all, in the psychology of learning. Bringing out the abstract similarity-indeed, identity-of this prima facie heterogeneous collection is a main goal in what follows. But doing so will require a somewhat idiosyncratic exposition of NS.

In the first place, we propose to introduce NS in a way that distinguishes between: (1) the theory considered simply as a 'black box' (that is, simply as a function that maps certain sorts of inputs onto certain sorts of outputs); and (2) the account that the theory gives of the mechanisms that compute that function and of the constraints under which the computations operate. This is, as we say, a somewhat eccentric way of cutting up the pie; but it will pay its way later on when we try to make clear what we take to be the trouble with NS.

In the second place, we want to develop our exposition of Darwin's account of evolution in parallel with an exposition of B. F. Skinner's theory of learning by operant conditioning (OT). Some of the similarities between the two have been widely noted, not least by Skinner himself.3 But we think, even so, that the strength of the analogy between NS and OT has been seriously underestimated, and that its implications have generally been misunderstood. In fact, the two theories are virtually identical: they propose essentially the same mechanisms to compute essentially similar functions under essentially identical constraints. This raises a question about which prior discussions of NS have been, it seems to us, remarkably reticent: it is pretty generally agreed, these days, that the Skinnerian account of learning is dead beyond resuscitation. So, if it is true that Skinner's theory and Darwin's are variations on the same theme, why aren't the objections that are routinely raised against the former likewise raised against the latter? If nobody believes Skinner any more, why does everybody still believe Darwin? We're going to argue that the position that retains the second but not the first is not stable.

Natural selection considered as a black box

As just remarked, one way to think about NS is as an account of the process that connects ancestral species with their descendants. Another (compatible) way is to think of it is as explaining how the phenotypic properties of populations change over time in response to ecological variables.4 By and large, contemporary discussions of evolution tend to stress the second construal; indeed, it's sometimes saidthat this sort of 'population thinking' was Darwin's most important contribution to biology.5

Whether or not that is so, 'population thinking' is convenient for our present purposes; it allows us to construe an evolutionary theory abstractly, as a black box in which the input specifies the distribution of phenotypes at a certain time (the GN [Generation N] distribution, together with the relevant aspects of its ecology), and in which the output specifies the distribution of phenotypes in the next generation (GN+1). This provides a perspective from which the analogies between NS and OT become visible, since OT is also plausibly viewed as a black box that maps a distribution of traits in a population at a time (a creature's behavioural repertoire at that time), together with a specification of relevant environmental variables (viz. the creature's history of reinforcement), onto a succeeding distribution of traits (viz. the creature's behavioural repertoire consequent to training). We therefore propose, in what follows, to indulge in a little 'population thinking' about both NS and OT.

Operant conditioning theory considered as a black box

If we are to think of the Skinnerian theory of learning in this way, we will first have to decide what is to count as a 'psychological trait'. Fortunately, Skinner has an explicit view about this, which, although by no means tenable, will serve quite nicely for the purposes of exposition.

Let's stipulate that a creature's 'psychological profile' at a certain time is the set of psychological traits of the creature at that time.6 For Skinner, a psychological trait is paradigmatically a stimulus-response (S-R) association; that is, it is a disposition to perform a token of a certain type of behaviour 'in the presence of' a token of a certain type of environmental event.7 Skinner takes S-R associations to be typically probabilistic, so a creature's psychological profile at a certain time is a distribution of probabilities over a bundle of S-R associations. Correspondingly, OT is a theory about how the distribution of probabilities in a population of S-R connections varies over time as afunction of specified environmental variables (including, notably, 'histories of reinforcement'). The picture is, in effect, that the totality of a creature's dispositions to produce responses to stimuli constitutes its psychological profile. These dispositions compete for strength, and environmental variables determine which dispositions win the competitions; they do so in accordance with the laws of conditioning that OT proposes to specify, and of which the so-called 'law of effect' is the paradigm.

We've been describing OT as a kind of 'population thinking' in order to emphasize its similarity to evolutionary theory (ET): both are about how traits in a population change over time in response to environmental variables (ecological variables; see footnote 5). That is, we suppose, a mildly interesting way of looking at things, but if it were all that the ET/OT analogy amounted to, it would warrant only cursory attention. In fact, however, there is quite a lot more to be said. Both theories postulate certain strong constraints (we'll call them 'proprietary' constraints)8 on how the empirical facts about population-to-population mappings are to be explained; and in both cases, the choice among candidate theories relies heavily on the imposition of these constraints.9 Some proprietary constraints derive from (what purport to be) general methodological considerations;10 but many of them are contingent and substantive. They derive from assumptions about the nature of evolution on the one hand and of learning on the other. The substance of the analogy between Darwin's version of evolutionary theory and Skinner's version of learning theory consists, in part, in the fact that the proprietary constraints that they endorse are virtually identical.

Proprietary constraints (1): iterativity

OT and NS are both formulated so as to apply 'iteratively' in their respective domains. That's to say that psychological profiles are themselves susceptible to further conditioning, and evolved phenotypes are themselves susceptible to further evolution. Iterativity is required in order that OT and ET should acknowledge the open-endednessof their respective domains: ET implies no bounds on the varieties of phenotypes that may be subject to evolution, and OT implies no bounds on the variety of behavioural profiles that may be modified by learning. The effect of this is to permit both theories to begin their explanations in medias res. ET presupposes some presumably very simple unevolved self-replicators with phenotypic traits to which the laws of evolution apply in the first instance; OT presupposes some presumably very simple repertoire of S-R associations to which the putative laws of conditioning apply in the first instance. In both cases, there are serious questions as to exactly what such 'starting assumptions' a theorist ought to endorse. In OT, the usual view is that an organism at birth (or perhaps in utero) is a random source of behaviours. That is, prior to operant learning, any stimulus may evoke any response, although the initial probability that a given stimulus will evoke a given response is generally very small. In ET, a lot depends on what kind of self-replicator evolutionary processes are supposed to have first applied to. Whatever it was, if it was ipso facto subject to evolution, it must have been a generator of heritable phenotypes, some of which were more fit than others in the environmental conditions that obtained.

Proprietary constraints (2): environmentalism

What phenotypes there can be is presumably determined by (among other things) what genotypes there can be; and what is genotypically possible is constrained by what is possible at 'lower' levels of organization: physiological, genetic, biochemical or whatever. Likewise for the effects of physiological (and particularly neurological) variables on psychological phenomena. It is, however, characteristic of both ET and OT largely to abstract from the effects of such endogenous variables, claiming that the phenomena of evolution on the one hand and of psychology on the other are very largely the effects of environmental causes.

A striking consequence of this assumption is that, to a first approximation, the laws of psychology and of evolution are both supposedto hold very broadly across the phylogenetic continuum, abstracting both from differences among individuals and from differences among species. (In the darkest days of conditioning theory, one psychologist claimed that, if we had a really adequate theory of learning, we could use it to teach English to worms. Happily, however, he later recovered.) Likewise, it is characteristic of evolutionary biologists to claim that the same laws of selection that shape the phenotypes of relatively simple creatures such as protozoa also shape the phenotypes of very complex creatures such as primates. It's clearly an empirical issue whether, or to what extent, such environmentalist claims are true in either case. It turned out that OT greatly underestimated the role of endogenous structures in psychological explanation; much of the 'cognitive science' approach to psychology has been an attempt to develop alternatives to OT's radical environmentalism. In Part one we will consider a number of recent findings in biology that suggest that analogous revisions may be required in the case of ET.11

Proprietary constraints (3): gradualism

ET purports to specify causal laws that govern transitions from the census of phenotypes in an ancestral population to the census of phenotypes in its successor generation. Likewise, OT purports to specify causal laws that connect a creature's psychological profile at a given time with its succeeding psychological profile. In principle, it is perfectly possible that such laws might tolerate radical discontinuities between successive stages; gradualism amounts to the empirical claim that, as a matter of fact, they do not. This implies, in the case of ET, that even speciation is a process in which phenotypes alter gradually, in response to selection pressure.12 'Saltations' (large jumps from a phenotype to its immediate successor) perhaps occur from time to time; but they are held to be sufficiently infrequent that theories of evolution can generally ignore them.13 In OT, gradualism implies that learning curves are generally smooth functions of histories of reinforcement. Learning consists of a gradual increment of the strength of S-R associations and not, for example, in suddeninsights into the character of environmental contingencies. Strictly speaking, according to OT, there is no such thing as problem solving; there is only the gradual accommodation of a creature's behaviours and behavioural dispositions to regularities in its environment.

In neither learning nor evolution is the claim for gradualism self-evidently true. Apparent discontinuities in the fossil record were a cause of considerable worry to Darwin himself, and there continues to be a tug-of-war about how they ought to be interpreted: evolutionary biologists may see fortuitous geological artefacts where palaeontologists see bona fide evidence that evolution sometimes proceeds in jumps (Eldredge, 1996). Likewise, a still robust tradition in developmental psychology postulates a more-or-less fixed sequence of cognitive 'stages', each with its distinctive modes of conceptualization and correspondingly distinctive capacities for problem solving. Piagetian psychology is the paradigm; for decades Piaget and Skinner seemed to be exclusive and exhaustive approaches to the psychology of learning.14

It is thus possible to wonder why gradualism has seemed, and continues to seem, so attractive to both evolutionary theorists and learning theorists. Some of the answer will become apparent when we, as it were, open the two black boxes and consider how ET and OT go about computing their respective outputs. Suffice to say, in the meantime, that the case for evolutionary gradualism was strengthened by the 'modern synthesis' of evolutionary biology with genetics. To a first approximation, the current view is that alterations of phenotypes typically express corresponding alterations of genotypes, alterations of genotypes are typically the consequence of genetic mutation, and macromutations generally decrease fitness. If all that is true, and if evolution is a process in which fitness generally increases over time, it follows that saltations cannot play a major role in evolutionary processes.

The allegiance to gradualism in the psychology of learning is perhaps less easily explained; at a minimum, there would appear to be abundant anecdotal evidence for discontinuities in cognitive processes that mediate learning, problem solving and the like ('and then it suddenly occurred to me ...', 'and then we realized ...' and so forth).But it's important to bear in mind that OT is a direct descendant of the associationism of the British empiricists. In particular, OT inherited the empiricist's assumption that learning consists mostly of habit formation; and, practically by definition, habits are traits that are acquired gradually as a consequence of practice. In this respect, the differences between Skinner and (e.g.) Hume turn mostly on issues about behaviourism, not on their theories of learning per se. Both think that learning is primarily associative and that association is primarily the formation of habits.

Proprietary constraints (4): monotonicity

If the ecology remains constant, selection increases fitness more or less monotonically;15 likewise for the effects of operant condition in increasing the efficiency of psychological profiles.16 The motivation for these constraints is relatively transparent: ET and OT are one-factor theories of their respective domains. According to the former, selection is overwhelmingly decisive in shaping the evolution of phenotypes; 17 according to the latter, reward is overwhelmingly decisive in determining the constitution of psychological profiles. Because, by assumption, there are no variables that interact significantly with either selection or reinforcement, the monotonicity of each is assured: if selection for a phenotypic trait increases fitness on one occasion, then it ought also to increase fitness on the next; if a certain reinforcing stimulus increases the strength of a certain response habit, the next reinforcement should do so too.18

These are, of course, very strong claims. In real life (that is, absent radical idealization) practically nothing is a monotonic function of practically anything else. So perhaps it's unsurprising that there are counter-intuitive consequences, both for theories of evolution, according to which the effect of selection on fitness is monotonic, and for theories of learning, which claim monotonicity for the effects of reinforcement on habit strength. What is interesting for our present purposes, however, is that the prima facie anomalies are very similar in the two cases. Thus, for example, OT has notorious problems withexplaining how 'local maximums' of efficiency are ever avoided in the course of learning, and ET has exactly the same problems with explaining how they are ever avoided in the course of selection. According to OT, creatures should persist in a relatively stupid habit so long as it elicits significant reinforcement; and that will be so even though there are, just down the road, alternative behavioural options that would increase the likelihood of reinforcement. Likewise, according to ET, if evolution finds a phenotypic trait that increases fitness, then selection will continue to favour that trait so long as the ecology isn't altered. This is so even if the phenotype that evolution has settled on is less good than alternative solutions would be. It is thus often said that evolution, as ET understands it, 'satisfices' but does not optimize: given enough time and a constant ecology, natural selection is guaranteed to converge on some fit phenotype or other; but if it happens to converge on the best of the possible adaptations, that's merely fortuitous. Exactly likewise in the case of the selection of S-R pairs by reinforcers. Neither ET nor OT provides a way of taking one step backwards in order to then take two steps forward.19

Plainly, however, the claim that evolution is a (mere) satisficer is prima facie a good deal more plausible than the corresponding claim about learning. Intuitively (though not, of course, according to OT), Scrooge can think to himself: 'I would be even richer if we didn't heat the office' and thence turn down the thermostat. But evolution can't think to itself 'frogs would catch still more flies if they had longer tongues' and thence lengthen the frog's tongue in order that they should do so.

Proprietary constraints (5): locality

The problems about local maximums exhibit one of a number of respects in which selection, as ET understands it, and learning, as OT understands it, are both 'local' processes: their operation is insensitive to the outcomes of merely hypothetical contingencies. What happened can affect learning or evolution; what might have happened but didn't ipso facto can't.

Likewise (and for much the same reasons) natural selection and reinforcement learning are insensitive to future outcomes (that the river will dry up next week does not affect any creature's fitness now; that the schedule of reinforcement will be changed on Tuesday does not affect the strength of any habits on Monday). Similarly for past events (unless they leave present traces); similarly for events that are merely probable (or merely improbable); similarly for events that happen too far away to affect the causal interactions that a creature is involved in; similarly for events from which the creature is mechanically isolated (there's an ocean between it and what would otherwise be its predators, and neither can fly or swim); similarly, indeed, for events from which a creature is causally isolated in any way at all. The general principle is straightforward: according to ET, nothing can affect selection except actual causal transactions between a creature and its actual ecology. According to OT, nothing can affect learning except actual reinforcements of a creature's actual behaviours.

We make a similar point to one we made above: although ET and OT have acknowledged much the same proprietary constraints, there is no principled reason why both of them should do so: it seems perfectly possible, for example, that selection should be locally caused even if learning is not. After all, learning, but not evolving, typically goes on in creatures that have minds, and minds are notoriously the kind of thing that may register the effects of events that are in the past (but are remembered) and of events that are in the future (but are anticipated) and events that are merely possible (but are contemplated) and so forth. We think, in fact, that whereas selection processes are ipso facto local, psychological processes are quite typically not. If we are right to think that, then the similarity of standard versions of ET and OT is a reason for believing that at least one of them is false.

Proprietary constraints (6): mindlessness

There is at least one way (or perhaps we should say, there is at least one sense) in which a creature can be affected by an event from whichit is causally isolated: namely, it can be affected by the event as mentally represented. Thus: we consider cheating on our income tax; we find that we are very strongly tempted. 'But,' we think to ourselves, 'if we cheat, they are likely to catch us; and if they catch us, they are likely to put us in jail; and if they were to put us in jail, our cats would miss us'. So we don't cheat (anyhow, we don't cheat much). What's striking about this scenario is that what we do, or refrain from doing, is the effect of how we think about things, not of how the things we think about actually are. We don't cheat but we consider doing so; we don't go to jail, although the possibility that we might conditions our behaviour.

Now, causal interactions with events that are (merely) mentally represented would, of course, violate the locality constraint; it is presumably common ground that nothing counts as local unless it exists in the actual world. But mental representations themselves can act as causes, as when we cheat, or don't, because we've thought through the likely consequences. Darwin, however, held that the scientific story about how phenotypes evolve could dispense with appeals to mental causes. Indeed, one might plausibly claim that getting mental causes out of the story about how phenotypes evolve was his primary ambition.

There is, according to Darwin, no point at which an acceptable evolutionary explanation could take the form: such and such a creature has such and such a trait because God (or Mother Nature, or selfish genes or the Tooth Fairy) wished (intended, hoped, decided, preferred, etc.) that it should. This is so not only because there isn't any Tooth Fairy (and mere fictions do not cause things), but also because natural selection does not involve agency. That is, of course, a crucial respect in which the way natural selection is unlike artificial selection. If there are rust-resistant plants, that's because somebody decided to breed for them. But nobody decided to breed for the rust; not even God.20 Mental causation (in particular, what philosophers call 'intentional causation')21 literally does not come into natural selection; Skinner himself rightly emphasized that Darwin was committed to this;22 not, however, because Darwin was a behaviourist (hewasn't) but because Darwin didn't believe in the Tooth Fairy (or, quite likely, in God either). There is, as we will presently find reason to emphasize, an occasional tendency among neo-Darwinians to flout this principle. That is entirely deplorable, and has caused endless confusion both in the journals and in the press.

We assume that Darwin was right that natural selection is not a kind of mental causation. It is not, however, at all obvious that the psychology of learning (or the psychology of anything else) can operate under a corresponding mindlessness constraint. There probably isn't a God or a Tooth Fairy; but there are minds, and they do have causal powers, and it is not implausible that one of their functions is to represent how things might have been, or might be, or are in some other part of the forest, or would have been but that ... and so forth. Commonsense psychology embraces causation by mental representations as a matter of course. By contrast, it's of the essence of Skinner's behaviourism, hence of OT, to deny that there is any such thing. In this respect, OT really was (and really was intended to be) a radical departure from commonsense ways of thinking about the mental. The analogy between OT and ET is exact in this respect: both prescind from the postulation of mental causes. The difference is that Darwin was right: evolution really is mindless. But Skinner was wrong: learning is not.

Thinking inside the boxes

So much for some of the similarities between ET and OT that emerge when they are viewed from the 'outside'-that's to say, from the perspective of what they propose to do rather than that of the mechanisms by which they propose to do it. Both are functions from states of populations to their successor states; and there are a number of substantive and methodological constraints that both endorse and are, to varying degrees, contentious. We now wish to change the point of view and consider the mechanisms that are proposed to implement these functions.

Two things strike the eye when the boxes are opened: first, theextremely exiguous character of the resources on which ET and OT rely to account for the rich and complex domains of data to which they are respectively responsible; and second, the all-but-identity of the causal mechanisms that the two theories postulate. We want to have a look at both of these. In a nutshell, ET and OT both offer 'generate and test' theories of the data that they seek to explain: each consists of a random generator of traits and a filter over the traits that are so generated. And that is all.23

We will look first at OT. It is convenient to do so because Darwin is, in certain crucial respects, less explicit about what mechanisms he thinks mediate the evolution of new phenotypes than Skinner is about what mechanisms he thinks mediate the fixation of new psychological profiles.

The mechanism of learning according to operant conditioning theory

It is crucial for Skinner that, in its initial state (which is to say, in abstraction from effects of prior learning), an organism is 'a random generator of operants.'24 That is (prior learning and unconditioned reflexes aside), the psychological profiles on which OT operates are an unsystematic collections of S-R dispositions, each with an associative strength at or near zero. As previously remarked, OT undertakes to explain how reinforcement alters the strength of such dispositions in the direction of generally increasing efficiency.25

If, in the first instance, creatures generate S-R associations at random, then some 'shaping' mechanism must determine that the relative associative strength of some such pairs increases over time and that of others declines. It is characteristic of OT (as opposed, for example, to other varieties of associationism) to claim that shaping mechanisms are sensitive solely to exogenous variables; which is to say that association is sensitive solely to the effects of reinforcements on habit strength26 in accordance with such laws of operant conditioning as, for example, the 'law of effect' (the strength of an association increases with the frequency with which it is 'followedby' reinforcement). In brief: S-R associations that are generated at random are then 'filtered' by a mechanism that implements the laws of association.

A word about 'random' generators: as gradualism is in force, the successor of a psychological profile cannot differ arbitrarily from its immediate ancestor. Reinforcement cannot, in one step, replace a low-strength S-R habit by a high-strength habit that connects some quite different stimulus to some quite different response. Perhaps reinforcing a random bar press in the presence of a light will produce a more urgent bar press next time the light goes on. But it won't produce a high-strength association between, as it might be, the light and an ear twitch; or between the bar press and the sound of a piano.27 Reinforcement can lead to the association of 'new' kinds of responses (and/or of responses to new kinds of stimuli) but only via intermediate psychological profiles. The glaring analogy is to 'no saltation' theories of evolution (including ET), according to which the radical discontinuities between a creature's phenotype and the phenotype of its relatively remote ancestors must be mediated by the evolution of intermediate phenotypic forms. Accordingly, just as much of the serious scientific debate about OT has turned on whether learning curves are generally smooth enough to sustain its predictions, so much of the serious scientific debate about ET has turned on the extent to which the palaeontological record sustains the existence of intermediate forms in evolution.

The mechanism of selection according to evolutionary theory

According to the versions of Darwinism that have been standard since the 'new synthesis' of evolutionary theory with genetics, the overall picture is as follows.

1. Phenotypic variation 'expresses' genotypic variation.28

2. Genotypic variation from one generation to the next is the effect of random mutation.

3. Macromutations are generally lethal.

4. The phenotypic expression of viable mutations is generally random variation around population means.

In short, what ET says about the role of random genetic variation in the genesis of new species is exactly what OT says about the role of random operants in the formation of new behaviour profiles. The only relevant difference between the two is that random genetic variations can be heritable but random variations in the strength of operants cannot.

If the distribution of traits in a population is produced by filtering the output of a random generator, what is the filter? It's here that Skinner's story about the effect of conditioning in filtering randomly generated psychological profiles is more explicit than Darwin's story about the effect of selection in filtering randomly generated phenotypes. We will argue that, in fact, ET can offer no remotely plausible account of how filtering by natural selection might work. So here, finally, the analogy between OT and ET breaks down.

The putative laws of association provide Skinner with an account of how exogenous variables (in particular, schedules of reinforcement) filter populations of psychological profiles; they explain why the effect of such variables is that the relative strength of some habits increases over time and the relative strength of others does not.

So that answers the rhetorical question that is the title of this chapter. What kind of theory is Darwin's theory of natural selection? The same kind as Skinner's theory of operant conditioning. With, however, the following caveat: all that's wrong with Skinner's story about the filtering of psychological profiles is that it is a variety of associationism, and quite generally, associationism is not true. But Darwinism has (we'll claim) no analogous story about the evolutionary filtering of randomly generated phenotypes. In consequence, whereas Skinner's theory of conditioning is false, Darwin's theory of selection is empty.

So, anyhow, we will argue in Part two. 29

Copyright © 2010 by Jerry Fodor and Massimo Piattelli-Palmarini

Table of Contents

Acknowledgements xiv

Terms of engagement xv

1 What kind of theory is the theory of natural selection? 1

Part 1 The Biological Argument 17

2 Internal constraints: what the new biology tells us 19

3 Whole genomes, networks, modules and other complexities 40

4 Many constraints, many environments 57

5 The return of the laws of form 72

Part 2 The Conceptual Situation 93

6 Many are called but few are chosen: the problem of 'selection-for' 95

7 No exit? Some responses to the problem of 'selection-for' 117

8 Did the dodo lose its ecological niche? Or was it the other way around? 139

9 Summary and postlude 153

10 Afterword and reply to the critics 165

Appendix 191

Notes 207

References 251

Index 277

Customer Reviews