Darwin's Dangerous Idea (75 page)

will account for the structure of the language organ is as pure an anti-Spencerian doctrine as you could find. This explains his misconstrual of my The problem here is how to get the pendulum to stop swinging back and friendly suggestion about passing the buck to biology. I was assuming, as a forth so destructively. Time and again we see the same failure of commu-good Spencerian adaptationist, that "genes are the channel through which the nication. The truly unfortunate communication gap that Simon and Kaplan environment speaks," as Godfrey-Smith puts it, whereas Chomsky prefers to speak of (in their quotation at the head of the previous section) is the think of the genes' getting their message from some intrinsic, ahis-torical, amplified effect of a relatively simple bit of initial misunderstanding. Recall nonenvironmental source of organization—"physics," we may call it the difference between reductionists and greedy reductionists (chapter 3, Spencerians think that even if there are such timeless "laws of form," they section 5): reductionists think everything in nature can all be explained could impose themselves on things only through some selectional process or without skyhooks; greedy reductionists think it can all be explained without other.

cranes. But one theorist's healthy optimism is another theorist's unseemly Evolutionary thinking is just one chapter in the history of Spencerian-greed. One side proposes an oversimple crane, at which the other side versus-anti-Spencerian thinking. Adaptationism is a Spencerian doctrine, and scoffs—"Philistine reductionists!"—declaring, truthfully, that life is much so is Skinner's behaviorism, and so, more generally, is any variety of more complicated than that. "Bunch of crazy skyhook-seekers!" mutters the
empiricism.
Empiricism is the view that we furnish our minds with details first side, in defensive overreaction. That is what they would mutter if they that all come from the outside environment, via experience. Adaptationism is had the term—but, then again, if both sides had the terms, they might be able the view that the selecting environment gradually shapes the genotypes of to see what the issues really were, and avoid the miscommunication organisms, molding them so that the phenotypes they command are some altogether. That is my hope.

near-optimal fit with the encountered world. Behaviorism is the view that What are Chomsky's actual views? If he doesn't think the language organ is what Skinner (1953, especially pp. 129-41) called "the controlling envi-shaped by natural selection, what account does he give of its complexities?

ronment" is what "shapes" the behavior of all organisms. Now we can see The philosopher of biology Peter Godfrey-Smith (1993) has recently focused that Chomsky's famous attack on Skinner was as much an attack on Skinner's on the family of views that maintain, in one way or another, that "there is Spencerian view
that
the environment shaped the organism as it was on the complexity in the organism in virtue of complexity in the environment."

limitations of Skinner's model of
how
this shaping took place.

Since this was one of Herbert Spencer's pet themes, Godfrey-Smith proposes Skinner proclaimed that
one simple iteration
of the fundamental Dar-we call any such view "Spencerian."10 Spencer was a Darwinian—or you winian process—operant conditioning—could account for all mentality, all could say that Charles Darwin was a Spencerian. In any event, the modern learning, not just in pigeons but in human beings. When critics insisted that synthesis is Spencerian to its core, and it is the Spencer-ism of that orthodoxy thinking and learning were much, much more complicated than that, he (and that is most often attacked in one way or another by rebels. Manfred Eigen his followers) smelled skyhooks, and wrote off the critics of behaviorism as and Jacques Monod are both Spencerian, for instance, in their insistence that dualists, mentalists, antiscientinc know-nothings. This was a mis-perception; it is only through environmental selection that molecular function can be the critics—at least the best of them—were simply insisting that the mind specified (chapter 7, section 2; chapter 8, section 3 ), whereas Stuart was composed of a lot more cranes than Skinner imagined.

Kauffman's insistence that order emerges
in spite
of( envi-ronmental) Skinner was a greedy reductionist, trying to explain
all
the design (and selection expresses an anti-Spencerian challenge (chapter 8, section 7). Brian design power) in a single stroke. The proper response to him should have Goodwin's denial (1986) that biology is a
historical
science is another been: "Nice try—but it turns out to be much more complicated than you example of anti-Spencerism, since it is a denial that historical interactions think!" And one should have said it without sarcasm, for Skinner's
was
a nice with earlier environments are the source of the try. It was a great idea, which inspired (or provoked) a half-century of hardheaded experimentation and model-building from which a great deal was learned. Ironically, it was the repeated failures of
another
brand of greedy 10. It is also one of Herbert Simon's pet themes in
Sciences of the Artificial
( 1969 ), so reductionism, dubbed "Good Old-Fashioned AI" or "GOFAI" by Haugeland we might call it Simonian—or Herbertian.

(1985), that really convinced psychologists that the mind was 396 LOSING OUR MINDS TO DARWIN

Nice Tries
397

indeed a phenomenon of surpassing architectural complexity—much too made possible such pioneers as Donald Hebb ( 1949) and Frank Rosenblatt complicated for behaviorism to describe. The founding insight of GOFAI was
(
1962 ), whose "Perceptrons" were, as Minsky and Papert soon pointed out Turing's recognition that a computer could be
indefinitely compli cated
but
(
1969), a nice try, but much too simple. Now, several decades later, another that all computers could be made from simple parts. Whereas Skinner's wave of more complicated but still usefully simple nice tries, flying the flag simple parts had been randomly mated stimulus-response pairings that could of Connectionism, are exploring portions of Design Space left unexamined then be subjected, over and over again, to the selection pressure of by their intellectual ancestors.12

reinforcement from the environment, Turing's simple parts were internal data-The human mind is an amazing crane, and there is a lot of design work that structures—different "machine states" that could be composed to respond has to have been done to build it, and to keep it working and up-to-date now.

differentially to indefinitely many different inputs, creating input output That is Darwin's "Spencerian" message. One way or another, the history of behavior of any imaginable sophistication. Which of these internal states environmental encounters over the eons (and during the last ten minutes) has were innately specified and which were to be revised by experience was shaped the mind you have right now. Some of the work must have been done something left to be investigated. Like Charles Babbage ( see note 13 of by natural selection, and the rest by one or another internal generate-and-test chapter 8), Turing saw that the behavior of an entity need not be any
simple
process of the sort we looked at earlier in the chapter. None of it is magic; function of its
own
history of stimulation, since it could have accrued huge none of it involves an internal skyhook. Whatever models we propose of amounts of design over the eons, which would permit it to use its internal these cranes will surely be too simple in one regard or another, but we are complexity to mediate its responses. That abstract opening was eventually closing in, trying out the simple ideas first. Chomsky has been one of the filled by GOFAI-modelers with contrivances of dazzling complexity that
still
leading critics of these nice tries, dismissing everyone from B. F. Skinner, fell comically short of producing human-style cognition.

through such GOFAI mavens and mavericks as Herbert Simon and Roger Today the reigning orthodoxy in cognitive science is that yesterday's Schank, to all the Connectionists, and he has always been right that their simple models of perception, learning, memory, language production, and ideas have been too simple by far, but he has also exhibited a hostility to the language understanding are orders of magnitude too simple, but those simple
tactic
of trying for simple models that has unduly raised the temperature of models were often nice tries, without which we would still be wondering how the debates. Suppose, for the sake of argument, we grant that Chomsky could simple it might, after all, turn out to be. It makes sense to err on the side of see better than anyone else that the mind, and the language organ which plays greedy reductionism, to try for the simple model before wallowing around in such a central role in its superiority over animal minds, are structures of a complexities. Mendel's simple genetics was a nice try, and so was the rather systematicity and complexity that beggar all models to date. All the more more complex "bean-bag genetics" it became in the hands of population reason, one would think, to search for an evolutionary explanation of these geneticists, even though it has often relied on such retrospectively outrageous brilliant devices. But although Chomsky uncovered for us the abstract oversimplifications that Francis Crick was tempted to kick it out of science.

structure of language, the crane that is most responsible for lifting all the Graham Cairns-Smith's clay crystals are a nice try, and Art Samuel's other cranes of culture into place, he has vigorously discouraged us from checkers-player was a nice try—much too simple, as we learned, but on the treating it as a crane. No wonder yearners for skyhooks have often taken him right track.

as their authority.

In the earliest days of the computer, Warren McCulloch and W. H. Pitts He is not the only candidate, however. John Searle is another favorite (1943 ) proposed a magnificently simple "logical neuron" from which "neural champion of skyhook-seekers, and he is certainly no Chomskian. We saw in nets" might be woven, and for a while it looked as if perhaps they had broken the back of the brain problem. Certainly, before they made their modest proposal, neurologists were desperately confused about how to think of the (1964 ) inspired me when I was a graduate student, and whose later work (e.g., 1989 ) has brain's activity. One has only to go back and read their brave flounderings, in persistently carved out new territories, still underappreciated by many in the trenches, the more speculative books of the 1930s and 1940s, to see what a tremendous and on the outskirts, in my opinion.

lift neuroscience got from McCulloch and Pitts.l' They 12. Other nice tries have been the neuroscientists' many models of learning as "Darwin-Jan" evolution in the nervous system, going back to the early work of Ross Ashby ( I960) and J. Z. Young (1965), and continuing today in the work of such people as Arbib, Grossberg (1976), Changeux and Danchin (1976), and Calvin (1987)—and Edelman 11. One of Warren McCulloch's students, himself a major contributor to these early (1987), whose work would be a nicer try if he didn't present it as if it were such a developments, is Michael Arbib, whose crystal-clear early discussion of these issues saltation in the wilderness.

398 LOSING OUR MINDS TO DARWIN

Nice Tries
399

chapter 8 (section 4) that Searle has defended a version of John Locke's That is certainly biting the bullet, and biting the same bullet as Chomsky, Mind-first vision, under the banner of Original Intentionality. According to but in a somewhat different spot: Yes of course, the LAD evolved, and so did Searle, automata (computers or robots) don't have real intentionality; at best consciousness (Searle 1992, pp. 88ff.), but Chomsky is right that there is no they have mere
as if
intentionality. Moreover, original or real intentionality hope of a reverse-engineering account of either of them. Chomsky is wrong, cannot be composed of, derived from—or, presumably, descended from—

however, to grant even the coherence of an automaton-level description of mere
as if
intentionality. This creates a problem for Searle, because, whereas the process, for that opens the door to "strong Artificial Intelligence."

Artificial Intelligence says you are
composed of
automata, Darwinism says If Chomsky's position on the slippery slope is hard to maintain, Searle's you are
descended from
automata. It is hard to deny the former if you admit has even more awkward consequences.13 He grants, as we can see in the the latter; how could anything born of automata ever be anything but a much, passage quoted above, that there is a "functional" story to be told about how much fancier automaton? Do we somehow reach escape velocity and leave the brain does its work in language acquisition. There is also, he grants, a our automaton heritage behind? Is there some threshold that marks the onset

"functional" story to be told about how parts of the brain arrive at depth or of real intentionality? Chomsky's original hierarchy of ever fancier automata distance judgments in vision.
"But there is no mental content whatever at this
permitted him to draw the line, showing that
the minimal complexity
of an
functional level"
(Searle 1992, p. 234, Searle's emphasis). He then puts to automaton capable of generating the sentences of a human language puts it in himself the following quite reasonable retort from the cognitive scientists: a special class—still a class of automata, but at least an advanced class. This

"the distinction [between "function" talk and "mental content" talk] does not was not quite enough for Chomsky. As we have just seen, he planted his feet really make much difference to cognitive science. We continue to say what and said, in effect: "Yes, language makes the difference—but don't try to we have always said and do what we have always done, we simply substitute explain how the language organ got designed. It's a hopeful monster, a gift, the word 'functional' for the word 'mental' in these cases." (This is in fact nothing that could ever be explained." An awkward position to maintain: the what Chomsky has often said, in reply to such criticisms. See, for instance, brain is an automaton, but not one we can reverse-engineer. Is this perhaps a 1980.) To answer this retort, Searle (1992, p. 238) is obliged to take a step tactical mistake? According to Searle, Chomsky took one step too many backwards himself: not only is there no information-processing level of before planting his feet. He should have denied that the language organ had a explanation for the brain, he says; there is also really no "functional level" of structure that could even be described in automaton terms at all. By lapsing explanation in biology: