Steven Pinker 's Reply to Ahouse & Berwick's Review of How the Mind Works

Both published in The Boston Review.

How the Mind Works is a synthesis of cognitive science and evolutionary biology that aims to explain the human mind with three ideas: (1) Computation: thinking and feeling consist of information-processing in the brain; (2) Specialization: the mind is not a single entity, but a complex system of parts designed to solve different problems; (3) Evolution: as with the organs of the body, our complex mental faculties have biological functions ultimately related to survival and reproduction. The book lays out criteria for attributing an evolutionary function to a trait, and applies them to many hypotheses using data from cognitive science, psychology, anthropology, and biology.

In their review, Jeremy Ahouse and Robert Berwick dismiss it entirely. "Don't believe a word of it," they say about what they take to be its key assumptions. "More is not always more ... it is sometimes disastrously less." Despite their vehemence, it is not easy to see what the substantive disagreement is about. Some scientists disagree with (1) and assert that the mind is a direct product of the biochemistry of the brain, but Berwick cannot be among them: he himself takes a computational approach to language and mind.(Note 1) Other disagree with (2) and assert that there is a generic neural network learning algorithm which the brain uses for every mental process. But Berwick cannot among them, either: he works in a Chomskyan framework in which language is treated as a specialized module, and presumably he does not believe that language is the brain's only specialized module. As for (3), evolution, Berwick cannot be opposed in principal to examining the phylogenetic basis of mental faculties, since he himself has recently done just that for language.(Note 2)

There remains an issue of which mental processes are functional adaptations, as opposed to by-products of adaptations or the result of chance. It is unlikely that Ahouse and Berwick would deny that one can ever say that some part of the mind is an adaptation: the idea that stereo vision, fear, and sexual desire are adaptations to see in depth, avoid danger, and beget children are as indispensible and uncontroversial as the idea that the heart is an adaptation to pump blood or the kidneys are adaptations to filter it. There may be disagreements over whether particular mental faculties (such as probabilistic reasoning, grief, or humor) are adaptations, and if so, what they are for, but such arguments do not appear in Ahouse and Berwick's review.

The review becomes a bit less mysterious when we recognize that their accusations are taken, often wording and all, from their mentor in biology, Richard Lewontin. The review is the latest example of a conventional genre in modern intellectual life: the all-out attack by Lewontin or his collaborators (including Steven Rose, Stephen Jay Gould, and Philip Kitcher) on attempts to connect psychology with standard evolutionary biology. Since 1975, when Lewontin and others published their "Against Sociobiology" manifesto claiming that such attempts were politically reactionary and encouraged eugenics and Nazism, the attacks have recycled the same accusations and tactics: fuzzy scare words ("atomism," "reductionism," "determinism"), misreportings and doctored quotations, shameless straw-manning, empty name-calling ("vulgar," "pop"), political smears, personal innuendo, and most of all, the scientific snow job: seemingly damaging technical findings that general readers are unlikely to know about and hence unlikely to recognize as red herrings.(Note 3)

HTMW tried to pre-empt this kind of attack by patiently raising and refuting all the standard criticisms, both political and scientific. But Ahouse and Berwick seemed determined to discredit the book by any means necessary, and so rather than engaging its arguments, they have redoubled the use of shoddy tactics. As an example of their standards of fairness, here is an excerpt in which they try to show that I am ignorant of basic evolutionary biology:

Pinker asks, "Why don't women give virgin birth?" Certainly the correct
answer will not make particular reference to humans. Mammals, including
humans, just do not have this as a developmental option. Put otherwise,
it would take more than just a shift in selection regimes to make
humans start reproducing asexually. For these reasons it is hollow
bluster to talk about the selective advantage of sex in humans if the
traits we are discussing evolved and became established long before the
human lineage branched.
This guy Pinker sure does sound naive! Unfortunately for the readers of The Boston Review, the "hollow bluster" is a sentence fragment that A&B chopped out of context:
Why is there sex to begin with? ... Biologically speaking, the costs
are damnable indeed, so why do almost all complex organisms reproduce
sexually? Why don't women give virgin birth to daughters who are
clones of themselves instead of wasting half their pregnancies on sons
who lack the machinery to make grandchildren and are nothing but sperm
donors? Why do people and other organisms swap out half their genes
for the genes of another member of the species, generating variety in
their offspring for variety's sake? (p. 461-462).
The discussion that follows also makes it clear that that sex is puzzle for all organisms; humans are presented just as an example.

This is not the first time that Berwick has been caught doctoring a quotation in these pages; he also did it in his dismissive review of Richard Dawkins' fine book Climbing Mount Improbable.(Note 4) Nor is it the last time. As we shall see, A&B repeatedly fabricate the content of HTMW and misinform their readers about the technical results they marshal to criticize it.

Ahouse & Berwick's Distortions

Distortion 1: I uncritically believe that all traits are adaptations. Let's compare the accusations with what I wrote:

A&B: "In evolutionary terms, every trait we examine is an admixture
of physical constraints, natural selection and chance with history.
These features are constitutive, not optional. Pinker presents a
mutually exclusive conception pitting these factors against each
other."
HTMW: "Natural selection should not be pitted against developmental,
genetic, or phylogenetic constraints, as if the more important one of
them is, the less important the others are" (p. 169). ... "Organisms
can be understood only as interactions among adaptations, by-products
of adaptations, and noise" (p. 174).
A&B: "All traits do not lead to a selective advantage, the origin of
traits and their maintenance do not demand the same selectionist
account. Pinker never embraces this distinction as a vivid part of his
adaptationism."
HTMW on all traits leading to selective advantage: "The mind is an
adaptation designed by natural selection, but that does not mean that
everything we think, feel, and do is biologically adaptive (p. 23)."
"The major faculties of the mind ... show the handiwork of selection.
That does not mean that every aspect of the mind is adaptive" (p. 174).
"Some readers may be surprised to learn that after seven chapters of
reverse-engineering the major parts of the mind, I will conclude by
arguing that some of the activities we consider most profound are
nonadaptive by-products" (p. 525).
HTMW on origin versus maintenance: "Many organs that we see today
have maintained their original function. .... Others changed their
function. ... [Sometimes] before an organ was selected to assume its
current form, it was adapted for something else. (p. 170)"

Distortion 2: I am an atomist and a genetic reductionist, reducing every behavior to a simple trait, and then, in a straight line, to a single gene. This is exactly backwards. Behavior reduced to a single trait?

[The generator of behavior is] the package of information-processing
and goal-pursuing mechanisms called the mind. ...Any particular deed
done today is the effect of dozens of causes. Behavior is the outcome
of an internal struggle among many mental modules, and it is played out
on the chessboard of opportunities and constraints defined by other
people's behavior. (p. 42)
The on-board computers of social organisms, especially of humans,
should run sophisticated programs that assess the opportunities and
risks at hand and compete or cooperate accordingly. (p. 429)
Behavioral traits reduced to single genes?
The organization of our mental modules comes from our genetic program,
but that does not mean that there is a gene for every trait (p. 23).
An emphasis on innate design should not, by the way, be confused with
the search for "a gene for" this or that mental organ. ... Complex
mental organs, like complex physical organs, surely are built by
complex genetic recipes, with many genes cooperating in as yet
unfathomable ways (p. 34-35).
Neither adultery nor any other behavior can be in our genes (p. 42).

A&B's specific charges are equally backwards. They charge, "A behavior like `incest avoidance' gets boiled down to a simple heritable trait." I wrote the opposite: "Incest avoidance showcases the complicated software engineering behind our emotions for other people" (p. 456). A particularly blatant caricature comes in an attempt to educate us on birth order:

Birth order is a clear example of environment, not inheritance, as a
prime mover (later-born children could not be expected on average to
receive more gullibility genes from their parents). With the wind of
adaptationism filling his sails, Pinker must claim that the whole
behavioral repertoire response [sic] of first-borns or the inverse
behavior of later-borns is all selected for.
Must claim? Here is what I do claim:
[Children] should calculate how to make the best of the hand that
nature dealt them and of the dynamics of the poker game they were born
into. The historian Frank Sulloway has argued that the elusive
nongenetic component of personality is a set of strategies to compete
with siblings for parental investment, and that is why children in the
same family are so different. Each child develops in a different family
ecology and forms a different plan for getting out of childhood alive.
(453)
All children being equipped with strategies for competing with siblings is very different from parents' directing gullibility genes to later-borns.

Oddly enough, A&B write as if they themselves believe that genetic atomism and determinism are required by the theory of evolution:

If genes are to serve as accurate bookkeeping chits for maximizing
fitness, all the way to love's blushes, then the dotted lines from
genes to behaviors must run straight and true. Any deviation, any
non-determinism or interaction between and among the stepping-stones,
and our explanatory hold slips [emphasis added] (p. 38).
Must? It is a commonplace of evolutionary biology that genes are selected because of their probabilistic effects: a gene "for" a trait is a gene that, in comparison with rival alleles, and averaged over environments and over the other genes it appears with, increases the probability that the trait will appear. As A&B acknowledge later in the review, "Biologists have long said a lot about evolution without any detailed knowledge of the steps from genotype to phenotype. All you need are statistical correlations to move from gene to phenotype without a full causal story." Indeed they have. But A&B have just refuted their own claim that "the dotted lines from gene to behavior must run straight and true"!

Distortion 3: HTMW us just story-telling, and presents no basis for evaluating its hypotheses about the biological function of mental faculties:

Pinker knows that we can always replace one adaptive story by another
or add more useful modules as required, but offers us no way to
reasonably refrain from doing do.
In fact I "know" no such thing, do offer us a way, and do refrain from doing so myself.

First, I present standard criteria in biology for evaluating hypotheses about adaptive function. Obviously such criteria must exist. As Ernst Mayr, one of the century's seminal evolutionary biologists, has noted, "The adaptationist question, `What is the function of a given structure or organ?' has been for centuries the basis of every advance in physiology. If it had not been for the adaptationist program, we probably would still not yet know the functions of thymus, spleen, pituitary, and pineal. Harvey's question `Why are there valves in the veins?' was a major stepping stone in his discovery of the circulation of blood."

I detail these criteria right from the first chapter (p. 36-40; also 155-174, 524-526). A good adaptationist explanation specifies a goal relevant to survival or reproduction, the causal structure of the organism's environment, and the engineering designs suited to attain that goal in that environment. It then requires empirical data showing that the trait in question uncannily meets the engineering specs, showing signs of complexity, effectiveness, and specialization in solving the assigned problem, especially in comparison with alternative designs that are biologically possible for that kind of organism.

These criteria are applied throughout. Frequent comparisons with other primates and mammals show that alternatives to what we find in humans, such as lacking language, living in solitude, or mating indiscriminately, are developmentally possible (thus A&B's logical possibility in which genetic variation for such traits is "meager" can be dismissed). And HTMW cites close to 300 empirical reports or literature reviews in evaluating specific hypotheses.

A&B conceal this from the reader. For example, they write, "Are wicked step-parents and parent-child conflict the natural outcome of genetic-payoff investment calculations? HTMW relies on readers' willingness to supply examples from their own lives ..." No, HTMW relies on cross-cultural and ethnographic surveys, studies of of stepparents in the United States, and national statistics on child abuse and filicide (p. 433-434).

Here is a particularly unfortunate example:

A&B: "There are many stories we tell ourselves. HTMW presents one
that allows Pinker to ... rationalize his particular view of
relationships. Read as autobiography this may provide some insight, but
as storytelling there are certainly more interesting organizing myths.
A&B do not explain my personal views of relationships, my autobiographical history, or how they claim to know them; their innuendo is another attempt to hide the empirical content of HTMW. The "story" in question is the theory of sexual selection and parental investment, and it is a clear counterexample to the claim that adaptationist hypotheses are sterile and unfalsifiable post hoc stories. Darwin first noticed an asymmetry in mating in much of the animal kingdom: males compete, females choose. In the 1960s and 1970s George Williams, John Maynard Smith, and Robert Trivers provided an elegant explanation in terms of parental investment: whichever sex invests more in offspring becomes a limiting resource for the other, and so the less-investing sex competes, the greater-investing sex chooses. In 1979 Donald Symons amassed a vast set of data on human sexuality that supported this theory, refuted competing adaptationist hypotheses, and made many new predictions about promiscuity, jealousy, mate selection, and physical attractiveness. In the 1980s and 1990s these predictions were tested in laboratory experiments, sociological data, in vivo field studies, surveys of the ethnographic literature, and cross-cultural surveys involving tens of thousands of people in thirty-seven cultures -- and largely confirmed (p. 460-493). A&B make no mention of these empirical tests, preferring snide, ad hominem remarks.

Finally, HTMW refutes A&B's cliche that "in adaptationist history/fictions that Pinker fancies there is no end to plausible story telling ... we can always replace one adaptive story by another or add more modules as required, but [Pinker] offers us no way to reasonably refrain from doing so." A&B do not mention that a major conclusion of HTMW is that many of the most momentous human activities do not meet the criteria for adaptations, including written language, dreams, science, mathematics, music, art, religion, philosophy, and most narrative (p. 174, 302-306, 340-342, and all of Chapter 8).(Note 5)

Technical Issues

1. Vision. In the chapter on vision, I suggest that the visual system uses a local, topographic, viewer-centered representation of visible surfaces and their depths, which David Marr called a "2 1/2-D sketch" and which contemporary researchers have modified into what they call a "visual surface representation" (p. 256-261). A&B claim that "practically no one" believes in it anymore. That is false. Recent overviews by leading perception psychologists such as Berkeley's Stephen Palmer and Harvard's Ken Nakayama rely on a modified 2 1/2-D sketch; HTMW's discussion was based on a 1995 synthesis by Nakayama whose title tells a story: "Visual surface representation: A critical link between lower-level and higher-level vision." A&B cite "scientists at MIT" as disproving the visual-surface representation; when I showed two of them A&B's claim, the responses were: "To say `no one believes in it' sounds overly strong. My sense is that surfaces of this sort remain quite popular, both in computer vision and in human vision," and "The claim you cite below is very misleading and exaggerated!"(Note 6)

2. Behavioral Genetics. I report a near-consensus among human behavioral geneticists: that "much of the variation in personality -- about fifty percent -- has genetic causes." A&B assert that this naively confuses causation with correlation. I agree that A&B's definition of heritability -- "the tendency of offspring to resemble their parents" -- does not show that a trait is genetically influenced. But A&B have caricatured an entire field. The claim that variation in adult personality has substantial genetic causes is based on far more than that: that identical twins are more similar than fraternal twins, that biological siblings are more similar than adopted siblings, that these correlations are much the same whether the siblings are reared together or apart, that they remain even after one statistically controls for every possible contaminating factor ever proposed (including shared placentas), that these findings replicate across labs, decades, and countries, and that in several recent studies the variation has been tied to identifiable genes for neurotransmitters or their receptors or transpoters.(Note 7)

Perhaps fearful that they cannot make the charge stick, A&B turn to other tactics. First, guilt-by-association: they triumphantly refute a silly nine-year-old quotation about IQ from a newspaper reporter. Second, misreporting: they write that I "evidently" rely on Thomas Bouchard's studies of "25" monozygotic twins. In fact Bouchard and his collaborators have studied over 1200 monozygotic twins (112 of them raised apart), and I cite reviews of converging studies of thousands of twins and adoptees conducted by independent teams in half a dozen countries. And finally, the political card: "Pinker's assertion is simply the authority of modern science pressed into the service of speculative fictions -- truly biology as ideology." The latter is another trope from Lewontin, summarizing his theory that the current attention to DNA by molecular biologists is an attempt to root social problems in individual failings and thus preserve the status quo.(Note 8)

3. Population genetics. A&B claim that HTMW falters by not presenting calculations from population genetics, the mathematical modeling of genetic change in evolution:

HTMW contains nothing -- literally not one thing -- resembling either
evolutionary modeling, explicit fitness calculations, or the basics of
population or behavioral genetics.

This is false. HTMW was written for a wide audience and does not present equations or proofs, but I explain many results from the explicit modeling of evolution. These include the fundamental theorem of population genetics (p. 163), models of the rate of evolution (p. 163-164, 205), reconstruction of phylogeny through DNA comparisons (p. 202-205), Hamilton's models of kin selection (p. 398-401, 429-430), Trivers' formulation of reciprocal altruism (pp. 402-403, 502-504), Trivers' calculations on parental investment (p. 440-442, 463-464), models of the genetic costs of incest (p. 456-458), Grafen's modeling of sexual selection (p. 500-501), Maynard Smith & Price's game-theoretic analysis of conflict (p. 494-495), the game-theoretic analyses of bargains, threats, and promises by Schelling, Frank, and Hirshleifer (p. 409-412), Rogers' calculations of age- and sex-specific discounting of the future (p. 498), Tooby & Cosmides' calculations on coalitional aggression (p. 514), and computer simulations of the evolution of neural networks (pp. 177-179), eyes (p. 164), and strategies of cooperation (p. 503-504). Data on fitness are cited in discussions of children's age-specific value to parents (p. 452), incest (p. 456), male parental investment (p. 469), dominance (p. 495), and aggression (p. 510), among others. Behavioral genetics is discussed in pages 20-21, 448-449, and elsewhere.

It is true that I do not back up every hypothesis with mathematical models of changing gene frequencies, but then neither do biologists when they study the evolutionary function of the stomach or the eye. Pointedly, neither does Berwick himself in his own foray into evolutionary psychology, in which he confidently states that Chomsky's latest grammatical theory supports a saltationist account of the evolution of language.(Note 9)

A&B claim to know of specific results from population genetics that undermine HTMW. First they say that population genetics casts strong doubt on whether adaptations can easily evolve: if genes combine in complex ways to determine phenotypes, then the fittest phenotype may not evolve. Of course, fit phenotypes do evolve. Living things have improbable, intricately engineered parts: eyes and wings and lungs and immune systems and leaves and DNA repair and the other complex machinery that lets plants and animals stay alive. So the fact that some combinations of genes don't evolve into adapted phenotypes simply shows that organisms are not restricted to those combinations of genes over the evolutionary long run. A&B concede the point late in their review:

In recent conversation, James Crow, our foremost population geneticist,
has insisted to us that if there were not some trait independence,
evolution would grind to a halt, because any change would change all
the traits in an organism and so nothing of lasting substance could be
built. There is surely something to this.
There surely is, but they did not have to go to "our foremost population geneticist" to learn it. They could have asked me -- or any biologist who treats population genetics as a tool to be applied with care and common sense, rather than as a weapon with which to beat opponents.

A&B's specific charge is that I am unaware of a seldom-cited twenty-year-old paper reporting a mathematical model in which "the entire kin-selection- maximize fitness [sic] edifice collapses." Since kin selection -- the evolution of traits that benefit relatives because of their shared genes -- is a cornerstone of the modern evolutionary analysis of social relations, this would appear to be a major flaw. But A&B's reporting is highly distorted.(Note 10)

First of all, A&B are incorrect in saying that "Hamilton's original calculation was done only for haploid-diploid organisms where sisters have only half the usual number of chromosomes; this makes the bookkeeping easy, allowing us to conflate genes and genotypes." Or as Berwick put it in his other review of HTMW in the Los Angeles Times, the "cost-benefit arithmetic doesn't balance the genetic books properly, unless you assume people are either ants or bees." In fact Hamilton published his results in two classic papers: the first analyzed the standard diploid case, the second extended it to the haplodiploid social insects.

Second, the paper by Feldman and Cavalli-Sforza does not say that in their interpretation, the entire edifice of kin selection "collapses." It says only that Hamilton's original results are "model-dependent" -- a very different claim. Apparently A&B are unaware that in the ensuing two decades many mathematical biologists have developed plausible models in which kin-directed altruism can be proven to evolve, and without the bizarre parameter values that Feldman & Cavalli-Sforza deduced.(Note 11)

There is a more general error in A&B's referring to population genetics as "the auto mechanics of evolution" and their implication that those who do not use it at every step are courting elementary errors. Models in population genetics are, by necessity, grossly simplified thought experiments about evolution over the short term. A typical model might posit two genes (compared to our 75,000), each coming in two or three alleles (compared to the astronomical numbers possible over the long term), with their interactions fixed (though the interactions themselves may change over the long term), and model the entire environment and entire phenotype as a single coefficient. As in economics, sophisticated mathematical models may be restricted to artificial toy worlds and have little connection to reality. One reformed population geneticist, a collaborator of Feldman's, has recently written:

It seems, thus, that the repeated failure to apply empirically the
results of population genetic models to non-trivial evolutionary
problems (see e.g. Lewontin, 1974) stems from an attempt to predict
one process (say, long-term evolution) on the basis of a model,
corresponding to another process (say, short-term evolution). This
long-persisting attempt was based on the postulate, tacitly accepted by
most in the field (the author of this note included), that at least
qualitatively, the behaviour of the long-term process can be fully
understood by extrapolation of the analytically well-defined short-term
process. We see that this postulate is mathematically wrong.(Note 12)

No doubt there is much to criticize in HTMW. But bystanders have a right to an accurate characterization of the book's content and of the relevant technical results. Readers of their review and of similar attacks on evolutionary psychology should ask: if the approach were really that empty, would these critics need to resort to so many misreportings, caricatures, and other questionable tactics?

NOTES

(1)R. C. Berwick, 1985. The acquisition of syntactic knowledge. MIT Press.

(2)Robert Berwick, 1997, Syntax facit saltum: Computation and the genotype and phenotype of language, Journal of Neurolinguistics, 10, 231-249.

(3)See HTMW 44-58, 165-174, and my letter to The New York Review of Books, October 9, 1997; also Richard Dawkins' The extended phenotype, Oxford University Press, 1982, Daniel Dennett's Darwin's dangerous idea, Simon & Schuster, 1995, and Robert Trivers' chapter in In E. White (Ed.) (1981) Sociobiology and human politics, D. C. Heath.

(4)Dan Dennett, Letter, January 1997.

(5)An irony in A&B's accusation about positing modules is that, Berwick, when arguing in the area of research in which we both work, is quite happy to "add more modules as required" in support of his critique of -- surprise! -- me:

We can show that the Pinker and Maratsos accounts are misguided. We can
do a better job of explaining things via the interaction of two
components of language, the morphological and the syntactic. By carving
things this way, we shall arrive at an explanation where each "module"
is simpler than one in which the explanatory burden is shouldered by
the syntactic component alone. (R. Berwick & A. Weinberg, 1984, The
grammatical basis of linguistic performance
, p. 214. Cambridge, MA:
MIT Press.)

(6)A&B also misreport the Nature letter by Sinha & Poggio. (1) It was Clinton's facial features, not his "head and shoulders," that were grafted onto Gore. (2) Sinha & Poggio's conclusions were about head shape, not "global pose and stance." (3) Sinha & Poggio say nothing about a 2 1/2-D representation, because their demonstration has nothing to do with it.

(7)Plomin, R. 1989. Environment and genes: Determinants of behavior. American Psychologist, 44, 105--111; Genes and behavior: A special report, Science, 17, 1994, 1685-1739; Special issue of Current Directions in Psychological Science on Behavioral Genetics, October, 1997; Dean Hamer, 1998, Living with our genes, Doubleday. The Devlin et al. Nature letter that A&B refer to is highly tendentious and is answered by Thomas Bouchard in a forthcoming issue.

(8)Richard Lewontin, 1991, Biology as ideology: The doctrine of DNA. HarperCollins.

(9)Robert Berwick, 1997, Syntax facit saltum: Computation and the genotype and phenotype of language, Journal of Neurolinguistics, 10, 231-249.

(10)A&B unsystematically insert Trivers' very different theory of reciprocal altruism into their critique, but kin selection is the theory relevant to Hamilton and to Feldman & Cavalli-Sforza's principal model and to A&B's mention of step-parents that kicks off their discussion.

(11)See, e.g., Charlesworth, B. and Charnov, E. L., 1981, Kin selection in age-structured populations. Journal of Theoretical Biology, 88, 103-119; Toro, M., Abugov, R., Charlesworth, B. and Michod, R. E., 1982, Exact versus heuristic models of kin selection, Journal of Theoretical Biology, 97, 699-713; Taylor, P. D. and Frank, S. A., 1996, How to make a kin selection model, Journal of Theoretical Biology, 180, 27-37; Taylor, P. D., 1996, Inclusive fitness arguments in genetic models of behaviour, Journal of Mathematical Biology, 34, 654-674; Grafen, A., 1985, A geometric view of relatedness, Oxford Surveys in Evolutionary Biology, 2 28-89.

(12)Eshel, I., 1996, On the changing concept of evolutionary population stability as a reflection of a changing point of view in the quantitative theory of evolution. Journal of Mathematical Biology, 34, 485-510.