So the rumor that neural networks can replace mental
structure
with statistical learning is not true.
Steven-Pinker-The-Blank-Slate 1
?
?
Connectionism and extreme plasticity are popular among cognitive {75} scientists at the West Pole, who reject a completely blank slate but want to restrict innate organization to simple biases in attention and memory.
Extreme plasticity also appeals to neuroscientists who wish to boost the importance of their field for education and social policy, and to entrepreneurs selling products to speed up infant development, cure learning disabilities, or slow down aging.
Outside the sciences, all three developments have been welcomed by some scholars in the humanities who want to beat back the encroachments of biology.
2 The lean genome, connectionism, and extreme plasticity are the Blank Slate's last stand.
The point of this chapter is that these claims are not vindications of the doctrine of the Blank Slate but products of the Blank Slate. Many people (including a few scientists) have selectively read the evidence, sometimes in bizarre ways, to fit with a prior belief that the mind cannot possibly have any innate structure, or with simplistic notions of how innate structure, if it did exist, would be encoded in the genes and develop in the brain.
I should say at the outset that I find these latest-and-best blank-slate theories highly implausible -- indeed, barely coherent. Nothing comes out of nothing, and the complexity of the brain has to come from somewhere. It cannot come from the environment alone, because the whole point of having a brain is to accomplish certain goals, and the environment has no idea what those goals are. A given environment can accommodate organisms that build dams, migrate by the stars, trill and twitter to impress the females, scent-mark trees, write sonnets, and so on. To one species, a snatch of human speech is a warning to flee; to another, it is an interesting new sound to incorporate into its own vocal repertoire; to a third, it is grist for grammatical analysis. Information in the world doesn't tell you what to do with it.
Also, brain tissue is not some genie that can grant its owner any power that would come in handy. It is a physical mechanism, an arrangement of matter that converts inputs to outputs in particular ways. The idea that a single generic substance can see in depth, control the hands, attract a mate, bring up children, elude predators, outsmart prey, and so on, without some degree of specialization, is not credible. Saying that the brain solves these problems because of its "plasticity" is not much better than saying it solves them by magic. Still, in this chapter I will examine the latest scientific objections to human nature carefully. Each of the discoveries is important on its own terms, even if it does not support the extravagant conclusions that have been drawn. And once the last supports for the Blank Slate have been evaluated, I can properly sum up the scientific case for the alternative.
The human genome is often seen as the essence of our species, so it is not surprising that when its sequence was announced in 2001 commentators rushed to give it the correct interpretation for human affairs. Craig Venter, {76} whose company had competed with a public consortium in the race to sequence the genome, said at a press conference that the smaller-than-expected gene count shows that "we simply do not have enough genes for this idea of biological determinism to be right. The wonderful diversity of the human species is not hard-wired in our genetic code. Our environments are critical. " In the United Kingdom, The Guardian headlined its story, "Revealed: The Secret of Human Behaviour. Environment, Not Genes, Key to Our Acts. "3 An editorial in another British newspaper concluded that "we are more free, it seems, than we had realized. " Moreover, the finding "offers comfort for the left, with its belief in the potential of all, however deprived their background. But it is damning for the right, with its fondness for ruling classes and original sin. "4
All this from the number 34,000! Which leads to the question, What number of genes would have proven that the diversity of our species was wired into our genetic code, or that we are less free than we had realized, or that the political right is right and the left is wrong? 50,000? 150,000? Conversely, if it turned out that we had only 20,000 genes, would that have made us even freer, or the environment even more important, or the political left even more comfortable? The fact is that no one knows what these numbers mean. No one has the slightest idea how many genes it would take to build a system of hard-wired modules, or a general-purpose learning program, or anything in between -- to say nothing of original sin or the superiority of the ruling class. In our current state of ignorance of how the genes build a brain, the number of genes in the human genome is just a number.
If you don't believe this, consider the roundworm Caenorhabditis elegans, which has about 18,000 genes. By the logic of the genome editorialists, it should be twice as free, be twice as diverse, and have twice as much potential as a human being. In fact, it is a microscopic worm composed of 959 cells grown by a rigid genetic program, with a nervous system consisting of exactly 302 neurons in a fixed wiring diagram. As far as behavior is concerned, it eats, mates, approaches and avoids certain smells, and that's about it. This alone should make it obvious that our freedom and diversity of behavior come from having a complex biological makeup, not a simple one.
Now, it is a genuine puzzle why humans, with their hundred trillion cells and hundred billion neurons, need only twice as many genes as a humble little worm. Many biologists believe that the human genes have been undercounted. The number of genes in a genome can only be estimated; right now they cannot literally be totted up. Gene- estimating programs look for sequences in the DNA that are similar to known genes and that are active enough to be caught in the act of building a protein. 5 Genes that are unique to humans or active only in the developing brain of the
? ? ? ? ? ? fetus -- the genes most relevant to human nature -- and other inconspicuous genes could evade the software and get left {77} out of the estimates. Alternative estimates of 57,000,75,000, and even 120,000 human genes are currently being bruited about. 6 Still, even if humans had six times as many genes as a roundworm rather than just twice as many, the puzzle would remain.
Most biologists who are pondering the puzzle don't conclude that humans are less complex than we thought. Instead they conclude that the number of genes in a genome has little to do with the complexity of the organism. 7 A single
gene does not correspond to a single component in such a way that an organism with 20,000 genes has 20,000 components, an organism with 30,000 genes has 30,000 components, and so on. Genes specify proteins, and some of the proteins do become the meat and juices of an organism. But other proteins turn genes on or off, speed up or slow down their activity, or cut and splice other proteins into new combinations. James Watson points out that we should recalibrate our intuitions about what a given number of genes can do: "Imagine watching a play with thirty thousand actors. You'd get pretty conrused. "
Depending on how the genes interact, the assembly process can be much more intricate for one organism than for another with the same number of genes. In a simple organism, many of the genes simply build a protein and dump it into the stew. In a complex organism, one gene may turn on a second one, which speeds up the activity of a third one (but only if a fourth one is active), which then turns off the original gene (but only if a fifth one is inactive), and so on. This defines a kind of recipe that can build a more complex organism out of the same number of genes. The complexity of an organism thus depends not just on its gene count but on the intricacy of the box-and-arrow diagram that captures how each gene impinges on the activity of the other genes. 8 And because adding a gene doesn't just add an ingredient but can multiply the number of ways that the genes can interact with one another, the complexity of organisms depends on the number of possible combinations of active and inactive genes in their genomes. The geneticist Jean-Michel Claverie suggests that it might be estimated by the number two (active versus inactive) raised to the power of the number of genes. By that measure, a human genome is not twice as complex as a roundworm genome but 216,000 (a one followed by 4. 800 zeroes) times as complex. 9
There are two other reasons why the complexity of the genome is not reflected in the number of genes it contains. One is that a given gene can produce not just one protein but several. A gene is typically broken into stretches of DNA that code for fragments of protein (exons) separated by stretches of DNA that don't (introns), a bit like a magazine article interrupted by ads. The segments of a gene can then be spliced together in multiple ways. A gene composed of exons A, B, C, and D might give rise to proteins corresponding to {78} ABC, ABD, ACD, and so on -- as many as ten different proteins per gene. This happens to a greater degree in complex organisms than in simple ones. 10
Second, the 34,000 genes take up only about 3 percent of the human genome. The rest consists of DNA that does not code for protein and that used to be dismissed as "junk. " But as one biologist recently put it, "The term 'junk DNA' is a reflection of our ignorance. "11 The size, placement, and content of the noncoding DNA can have dramatic effects on the way that nearby genes are activated to make proteins. Information in the billions of bases in the non-coding regions of the genome is part of the specification of a human being, above and beyond the information contained in the 34,000 genes.
The human genome, then, is fully capable of building a complex brain, in spite of the bizarre proclamations of how wonderful it is that people are almost as simple as worms. Of course "the wonderful diversity of the human species is not hard-wired in our genetic code," but we didn't need to count genes to figure that out -- we already know it from the fact that a child growing up in Japan speaks Japanese but the same child growing up in England would speak English. It is an example of a syndrome we will meet elsewhere in this book: scientific findings spin-doctored beyond recognition to make a moral point that could have been made more easily on other grounds.
The second scientific defense of the Blank Slate comes from connectionism, the theory that the brain is like the artificial neural networks simulated on computers to learn statistical patterns. 12
Cognitive scientists agree that the elementary processes that make up the instruction set of the brain -- storing and retrieving an association, sequencing elements, focusing attention -- are implemented in the brain as networks of densely interconnected neurons (brain cells). The question is whether a generic kind of network, after being shaped by the environment, can explain all of human psychology, or whether the genome tailors different networks to the demands of particular domains: language, vision, morality, fear, lust, intuitive psychology, and so on. The connectionists, of course, do not believe in a blank slate, but they do believe in the closest mechanistic equivalent, a general-purpose learning device.
What is a neural network? Connectionists use the term to refer not to real neural circuitry in the brain but to a kind of computer program based on the metaphor of neurons and neural circuits. In the most common approach, a "neuron" carries information by being more or less active. The activity level indicates the presence or absence (or intensity or degree of confidence) of a simple feature of the world. The feature may be a color, a line with a certain slant, a letter
? ? ? ? ? ? ? ? ? of the alphabet, or a property of an animal such as having four legs.
A network of neurons can represent different concepts, depending on {79} which ones are active. If neurons for "yellow," "flies," and "sings" are active, the network is thinking about a canary; if neurons for "silver," "flies," and "roars" are active, it is thinking about an airplane. An artificial neural network computes in the following manner. Neurons are linked to other neurons by connections that work something like synapses. Each neuron counts up the inputs from other neurons and changes its activity level in response. The network learns by allowing the input to change the strengths of the connections. The strength of a connection determines the likelihood that the input neuron will excite or inhibit the output neuron.
Depending on what the neurons stand for, how they are innately wired, and how the connections change with training, a connectionist network can learn to compute various things. If everything is connected to everything else, a network can soak up the correlations among features in a set of objects. For example, after exposure to descriptions of many birds it can predict that feathered singing things tend to fly or that feathered flying things tend to sing or that singing flying things tend to have feathers. If a network has an input layer connected to an output layer, it can learn associations between ideas, such as that small soft flying things are animals but large metallic flying things are vehicles. If its output layer feeds back to earlier layers, it can crank out ordered sequences, such as the sounds making up a word.
The appeal of neural networks is that they automatically generalize their training to similar new items. If a network has been trained that tigers eat Frosted Flakes, it will tend to generalize that lions eat Frosted Flakes, because eating Frosted Flakes" has been associated not with "tigers" but with simpler features like "roars" and "has whiskers," which make up part of the representation of h'ons, too. The school of connectionism, like the school of associationism championed by Locke, Hume, and Mill, asserts that these generalizations are the crux of intelligence. If so, highly trained but otherwise generic neural networks can explain intelligence.
Computer modelers often set their models on simplified toy problems to ? rove that they can work in principle. The question then becomes whether the models can "scale up" to more realistic problems, or whether, as skeptics say, the modeler "is climbing trees to get to the moon. " Here we have the problem with connectionism. Simple connectionist networks can manage impressive displays of memory and generalization in circumscribed problems like reading a list of words or learning stereotypes of animals. But they are simply too underpowered to duplicate more realistic feats of human intelligence like understanding a sentence or reasoning about living things.
Humans don't just loosely associate things that resemble each other, or things that tend to occur together. They have combinatorial minds that enterrain propositions about what is true of what, and about who did what to {80} whom, when and where and why. And that requires a computational architecture that is more sophisticated than the uniform tangle of neurons used in generic connectionist networks. It requires an architecture equipped with logical apparatus like rules, variables, propositions, goal states, and different kinds of data structures, organized into larger systems. Many cognitive scientists have made this point, including Gary Marcus, Marvin Minsky, Seymour Papert, Jerry Fodor, Zenon Pylyshyn, John Anderson, Tom Bever, and Robert Hadley, and it is acknowledged as well by neural network modelers who are not in the connectionist school, such as John Hummel, Lokendra Shastri, and Paul Smolensky. 13 I have written at length on the limits of connectionism, both in scholarly papers and in popular books; here is a summary of my own case. 14
In a section called "Connectoplasm" in How the Mind Works, I laid out some simple logical relationships that underlie our understanding of a complete thought (such as the meaning of a sentence) but that are difficult to represent in generic networks. 15 One is the distinction between a kind and an individual: between ducks in general and this duck in particular. Both have the same features (swims, quacks, has feathers, and so on), and both are thus represented by the same set of active units in a standard connectionist model. But people know the difference.
A second talent is compositionality: the ability to entertain a new, complex thought that is not just the sum of the simple thoughts composing it but depends on their relationships. The thought that cats chase mice, for example, cannot be captured by activating a unit each for "cats," "mice," and "chase," because that pattern could just as easily stand for mice chasing cats.
A third logical talent is quantification (or the binding of variables): the difference between fooling some of the people all of the time and fooling all of the people some of the time. Without the computational equivalent of x's,y's, parentheses, and statements like "For all x," a model cannot tell the difference.
A fourth is recursion: the ability to embed one thought inside another, so that we can entertain not only the thought that Elvis lives, but the thought that the National Enquirer reported that Elvis lives, that some people believe the National Enquirer report that Elvis lives, that it is amazing that some people believe the National Enquirer report that Elvis lives, and so on. Connectionist networks would superimpose these propositions and thereby confuse their various subjects and predicates.
A final elusive talent is our ability to engage in categorical, as opposed to fuzzy, reasoning: to understand that Bob
? ? ? ? ? Dylan is a grandfather, even though he is not very grandfatherly, or that shrews are not rodents, though they look just like mice. With nothing but a soup of neurons to stand for an object's properties, and no provision for rules, variables, and definitions, the networks fall back on stereotypes and are bamboozled by atypical examples. {81}
In Words and Rules I aimed a microscope on a single phenomenon of language that has served as a test case for the ability of generic associative networks to account for the essence of language: assembling words, or pieces of words, into new combinations. People don't just memorize snatches of language but create new ones. A simple example is the English past tense. Given a neologism like to spam or to snarf, people don't have to run to the dictionary to look up their past-tense forms; they instinctively know that they are spammed and snarfed. The talent for assembling new combinations appears as early as age two, when children overapply the past-tense suffix to irregular verbs, as in We holded the baby rabbits and Horton heared a Who. 16
The obvious way to explain this talent is to appeal to two kinds of computational operations in the mind. Irregular forms like held and heard are stored in and retrieved from memory, just like any other word. Regular forms like walk- walked can be generated by a mental version of the grammatical rule "Add -ed to the verb. " The rule can apply whenever memory fails. It may be used when a word is unfamiliar and no past-tense form had been stored in memory, as in to spam, and it may be used by children when they cannot recall an irregular form like heard and need some way of marking its tense. Combining a suffix with a verb is a small example of an important human talent: combining words and phrases to create new sentences and thereby express new thoughts. It is one of the new ideas of the cognitive revolution introduced in Chapter 3, and one of the logical challenges for connectionism I listed in the preceding discussion.
Connectionists have used the past tense as a proving ground to see if they could duplicate this textbook example of human creativity without using a rule and without dividing the labor between a system for memory and a system for grammatical combination. A series of computer models have tried to generate past-tense forms using simple pattern associator networks. The networks typically connect the sounds in verbs with the sounds in the past-tense form: -am with -ammed, -ing with -ung, and so on. The models can then generate new forms by analogy, just like the generalization from tigers to lions: trained on crammed, a model can guess spammed; trained on folded, it tends to say holded.
But human speakers do far more than associate sounds with sounds, and the models thus fail to do them justice. The failures come from the absence of machinery to handle logical relationships. Most of the models are baffled by new words that sound different from familiar words and hence cannot be generalized by analogy. Given the novel verb to frilg, for example, they come up not with frilged, as people do, but with an odd mishmash like freezled. That is because they lack the device of a variable, like x in algebra or "verb" in grammar, which can apply to any member of a category, regardless of how familiar {82} its properties are. (This is the gadget that allows people to engage in categorical rather than fuzzy reasoning. ) The networks can only associate bits of sound with bits of sound, so when confronted with a new verb that does not sound like anything they were trained on, they assemble a pastiche of the most similar sounds they can find in their network.
The models also cannot properly distinguish among verbs that have the same sounds but different past-tense forms, such as ring the bell-rang the bell and ring the city-ringed the city. That is because the standard models represent
only sound and are blind to the grammatical differences among verbs that call for different conjugations. The key difference here is between simple roots like ring in the sense of "resonate" (past tense rang) and complex verbs derived from nouns like ring in the sense of "form a ring around" (past tense ringed). To register that difference, a language-using system has to be equipped with compositional data structures (such as "a verb made from the noun ring") and not just a beanbag of units.
Yet another problem is that connectionist networks track the statistics of the input closely: how many verbs of each sound pattern they have encountered. That leaves them unable to account for the epiphany in which young children discover the -ed rule and start making errors like holded and heared. Connectionist modelers can induce these errors only by bombarding the network with regular verbs (so as to burn in the -ed) in a way that is unlike anything real children experience. Finally, a mass of evidence from cognitive neuroscience shows that grammatical combination (including regular verbs) and lexical lookup (including irregular verbs) are handled by different systems in the brain rather than by a single associative network.
It's not that neural networks are incapable of handling the meanings of sentences or the task of grammatical conjugation. (They had better not be, since the very idea that thinking is a form of neural computation requires that some kind of neural network duplicate whatever the mind can do. ) The problem lies in the credo that one can do everything with a generic model as long as it is sufficiently trained. Many modelers have beefed up, retrofitted, or combined networks into more complicated and powerful systems. They have dedicated hunks of neural hardware to abstract symbols like "verb phrase" and "proposition" and have implemented additional mechanisms (such as synchronized firing patterns) to bind them together in the equivalent of compositional, recursive symbol structures.
? ? ? ? They have installed banks of neurons for words, or for English suffixes, or for key grammatical distinctions. They have built hybrid systems, with one network that retrieves irregular forms from memory and another that combines a verb with a suffix. 17
A system assembled out of beefed-up subnetworks could escape all the criticisms. But then we would no longer be talking about a generic neural network! We would be talking about a complex system innately tailored to {83} compute a task that people are good at. In the children's story called "Stone Soup," a hobo borrows the use of a woman's kitchen ostensibly to make soup from a stone. But he gradually asks for more and more ingredients to balance the flavor until he has prepared a rich and hearty stew at her expense. Connectionist modelers who claim to build intelligence out of generic neural networks without requiring anything innate are engaged in a similar business. The design choices that make a neural network system smart -- what each of the neurons represents, how they are wired together, what kinds of networks are assembled into a bigger system, in which way -- embody the innate organization of the part of the mind being modeled. They are typically hand-picked by the modeler, like an inventor rummaging through a box of transistors and diodes, but in a real brain they would have evolved by natural selection (indeed, in some networks, the architecture of the model does evolve by a simulation of natural selection). 18 The only alternative is that some previous episode of learning left the networks in a state ready for the current learning, but of course the buck has to stop at some innate specification of the first networks that kick off the learning process.
So the rumor that neural networks can replace mental structure with statistical learning is not true. Simple, generic networks are not up to the demands of ordinary human thinking and speaking; complex, specialized networks are a stone soup in which much of the interesting work has been done in setting up the innate wiring of the network. Once this is recognized, neural network modeling becomes an indispensable complement to the theory of a complex human nature rather than a replacement for it. 19 It bridges the gap between the elementary steps of cognition and the physiological activity of the brain and thus serves as an important link in the long chain of explanation between biology and culture. ~
FOR MOST OF its history, neuroscience was faced with an embarrassment: the brain looked as if it were innately specified in every detail. When it comes to the body, we can see many of the effects of a person's life experience: it may be tanned or pale, callused or soft, scrawny or plump or chiseled. But no such marks could be found in the brain. Now, something has to be wrong with this picture. People learn, and learn massively: they learn their language, their culture, their know-how, their database of facts. Also, the hundred trillion connections in the brain cannot possibly be specified individually by a 750-megabyte genome. The brain somehow must change in response to its input; the only question is how.
We are finally beginning to understand how. The study of neural plasticity is hot. Almost every week sees a discovery about how the brain gets wired in the womb and tuned outside it. After all those decades in which no one could find anything that changed in the brain, it is not surprising that the {84} discovery of plasticity has given the nature-nurture pendulum a push. Some people describe plasticity as a harbinger of an expansion of human potential in which the powers of the brain will be harnessed to revolutionize childrearing, education, therapy, and aging. And several manifestos have proclaimed that plasticity proves that the brain cannot have any significant innate organization. 20 In Rethinking Innateness, Jeffrey Elman and a team of West Pole connectionists write that predispositions to think about different things in different ways (language, people, objects, and so on) may be implemented in the brain only as "attention-grabbers" that ensure that the organism will receive "massive experience of certain inputs prior to subsequent learning. "21 In a "constructivist manifesto," the theoretical neuroscientists Stephen Quartz and Terrence Sejnowski write that "although the cortex is not a tabula rasa . . . it is largely equipotential at early stages," and therefore that innatist theories "appear implausible. "22
Neural development and plasticity unquestionably make up one of the great frontiers of human knowledge. How a linear string of DNA can direct the assembly of an intricate three-dimensional organ that lets us think, feel, and learn is a problem to stagger the imagination, to keep neuroscientists engaged for decades, and to belie any suggestion that we are approaching "the end of science. "
And the discoveries themselves are fascinating and provocative. The cerebral cortex (outer gray matter) of the brain has long been known to be divided into areas with different functions. Some represent particular body parts; others represent the visual field or the world of sound; still others concentrate on aspects of language or thinking. We now know that with learning and practice some of their boundaries can move around. (This does not mean that the brain tissue literally grows or shrinks, only that if the cortex is probed with electrodes or monitored with a scanner, the boundary where one ability leaves off and the next one begins can shift. ) Violinists, for example, have an expanded region of cortex representing the fingers of the left hand. 23 If a person or a monkey is trained on a simple task like recognizing shapes or attending to a location in space, neuroscientists can watch as parts of the cortex, or even
? ? ? ? ? ? ? ? ? individual neurons, take on the job. 24
The reallocation of brain tissue to new tasks is especially dramatic when people lose the use of a sense or body part. Congenitally blind people use their visual cortex to read Braille. 25 Congenitally deaf people use part of their auditory cortex to process sign language. 26 Amputees use the part of the cortex formerly serving the missing limb to represent other parts of their bodies. 27 Young children can grow up relatively normal after traumas to the brain that would turn adults into basket cases -- even removal of the entire left hemisphere, which in adults underlies language and logical reasoning. 28 All this {85} suggests that the allocation of brain tissue to perceptual and cognitive processes is not done permanently and on the basis of the exact location of the tissue in the skull, but depends on how the brain itself processes information.
This dynamic allocation of tissue can also be seen as the brain puts itself together in the womb. Unlike a computer that gets assembled in a factory and is turned on for the first time when complete, the brain is active while it is being assembled, and that activity may take part in the assembly process. Experiments on cats and other mammals have shown that if a brain is chemically silenced during fetal development it may end up with significant abnormalities. 29 And patches of cortex develop differently depending on the kind of input they receive. In an experimental tour de force, the neuroscientist Mriganka Sur literally rewired the brains of ferrets so that signals from their eyes fed into the primary auditory cortex, the part of the brain that ordinarily receives signals from the ears. 30 When he then probed the auditory cortex with electrodes, he found that it acted in many ways like the visual cortex. Locations in the visual field were laid out like a map, and individual neurons responded to lines and stripes at a particular orientation and direction of movement, similar to the neurons in an ordinary visual cortex. The ferrets could even use their rewired brains to move toward objects that were detectable by sight alone. The input to the sensory cortex must help to organize it: visual input makes the auditory cortex work something like the visual cortex.
What do these discoveries mean? Do they show that the brain is "able to be shaped, molded, modeled, or sculpted," as the dictionary definition of plastic would suggest? In the rest of this chapter I will show you that the answer is no. 31 Discoveries of how the brain changes with experience do not show that learning is more powerful than we thought, that the brain can be dramatically reshaped by its input, or that the genes do not shape the brain. Indeed, demonstrations of the plasticity of the brain are less radical than they first appear: the supposedly plastic regions of cortex are doing pretty much the same thing they would have been doing if they had never been altered. And the most recent discoveries on brain development have refuted the idea that the brain is largely plastic. Let me go over these points in turn.
The fact that the brain changes when we learn is not, as some have claimed, a radical discovery with profound implications for nature and nurture or human potential. Dmitri Karamazov could have deduced it in his nineteenth- century prison cell as he mulled over the fact that thinking comes from quivering nerve tails rather than an immaterial soul. If thought and action are products of the physical activity of the brain, and if thought and action can be affected by experience, then experience has to leave a trace in the physical structure of the brain. {86}
So there is. no scientific question as to whether experience, learning, and practice affect the brain; they surely do if we are even vaguely on the right track. It is not surprising that people who can play the violin have different brains from those who cannot, or that masters of sign language or of Braille have different brains from people who speak and read. Your brain changes when you are introduced to a new person, when you hear a bit of gossip, when you watch the Oscars, when you polish your golf stroke -- in short, whenever an experience leaves a trace in the mind. The only question is how learning affects the brain. Are memories stored in protein sequences, in new neurons or synapses, or in changes in the strength of existing synapses? When someone learns a new skill, is it stored only in organs dedicated to learning skills (like the cerebellum and the basal ganglia), or does it also adjust the cortex? Does an increase in dexterity depend on using more square centimeters of cortex or on using a greater concentration of synapses in the same number of square centimeters? These are important scientific problems, but they say nothing about whether people can learn, or how much. We already knew trained violinists play better than beginners or we would never have put their heads in the scanner to begin with. Neural plasticity is just another name for learning and development, described at a different level of analysis.
All this should be obvious, but nowadays any banality about learning can be dressed up in neurospeak and treated like a great revelation of science. According to a New York Times headline, "Talk therapy, a psychiatrist maintains, can alter the structure of the patient's brain. "32 I should hope so, or else the psychiatrist would be defrauding her clients. "Environmental manipulation can change the way [a child's] brain develops," the pediatric neurologist Harry Chugani told the Boston Globe. "A child surrounded by aggression, violence, or inadequate stimulation will reflect these connections in the brain and behavior. "33 Well, yes; if the environment affects the child at all, it would do so by changing connections in the brain. A special issue of the journal Educational Technology and Society was intended "to examine the position that learning takes place in the brain of the learner, and that pedagogies and technologies
? ? ? ? ? ? ? ? ? ? ? ? should be designed and evaluated on the basis of the effect they have on student brains. " The guest editor (a
biologist) did not say whether the alternative was that learning takes place in some other organ of the body like the pancreas or that it takes place in an immaterial soul. Even professors of neuroscience sometimes proclaim "discoveries" that would be news only to believers in a ghost in the machine: "Scientists have found that the brain is capable of altering its connections. . . . . . You have the ability to change the synaptic connections within the brain. "34 Good thing, because otherwise we would be permanent amnesiacs.
This neuroscientist is an executive at a company that "uses brain research {87} and technology to develop products intended to enhance human learning and performance," one of many new companies with that aspiration. "The
human being has unlimited creativity if focused and nurtured properly," says a consultant who teaches clients to draw diagrams that "map their neural patterns. " "The older you get, the more connections and associations your brain should be making," said a satisfied customer; "Therefore you should have more information stored in your brain.
You just need to tap into it. "35 Many people have been convinced by the public pronouncements of neuroscience advocates -- on the basis of no evidence whatsoever -- that varying the route you take when driving home can stave off the effects of aging. 36 And then there is the marketing genius who realized that blocks, balls, and other toys "provide visual and tactile stimulation" and "encourage movement and tracking," part of a larger movement of "brain- based" childrearing and education that we will meet again in the chapter on children. 37
These companies tap into people's belief in a ghost in the machine by implying that any form of learning that affects the brain (as opposed, presumably, to the kinds of learning that don't affect the brain) is unexpectedly real or deep or powerful. But this is mistaken. All learning affects the brain. It is undeniably exciting when scientists make a discovery about how learning affects the brain, but that does not make the learning itself any more pervasive or profound. ~
A second misinterpretation of neural plasticity can be traced to the belief that there is nothing in the mind that was not first in the senses. The most highly publicized discoveries about cortical plasticity concern primary sensory cortex, the patches of gray matter that first receive signals from the senses (via the thalamus and other subcortical organs). Writers who use plasticity to prop up the Blank Slate assume that if primary sensory cortex is plastic, the rest of the brain must be even more plastic, because the mind is built out of sensory experience. For example, one neuroscientist was quoted as saying that Sur's rewiring experiments "challenge the recent emphasis on the power of the genes" and "will push people back toward more consideration of environmental factors in creating normal brain organization. "38 But if the brain is a complex organ with many parts, the moral does not follow. Primary sensory cortex is not the bedrock of the mind but a gadget, one of many in the brain, that happens to be specialized for certain kinds of signal processing in the first stages of sensory analysis. Let's suppose that primary sensory cortex really were formless, getting all its structure from the input. Would that mean that the entire brain is formless and gets all of its structure from the input? Not at all. For one thing, even primary sensory cortex is just one part of a huge, intricate system. To put things in perspective, here is a recent diagram of the wiring of the primate visual system:39 {88}
? ? ? ? ? ? ? ? ? Primary visual cortex is the box near the bottom labeled "V1. " It is one of at least fifty distinct brain areas devoted to visual processing, and they are interconnected in precise ways. (Despite the spaghetti-like appearance, not everything is connected to everything else. Only about a third of the logically possible connections between components are actually present in the brain. ) Primary visual cortex, by itself, is not enough to see with. Indeed, it is so deeply buried in the visual system that Francis Crick and the neuroscientist Christof Koch have argued that we are not conscious of anything that goes on in it. 40 What we see -- familiar colored objects arranged in a scene or moving in {89} particular ways -- is a product of the entire contraption. So even if the innards of the V1 box were completely specified by its input, we would have to explain the architecture of the rest of the visual system -- the fifty boxes and their connections. I don't mean to imply that the entire block diagram is genetically specified, but much of it almost certainly is. 41
And of course the visual system itself must be put into perspective, because it is just one part of the brain. The visual system dominates some half-dozen of the more than fifty major areas of the cortex that can be distinguished by their anatomy and connections. Many of the others underlie other functions such as language, reasoning, planning, and
? ? ? social skills. Though no one knows to what extent they are genetically prepared for their computational roles, there are hints that the genetic influence is substantial. 42 The divisions are established in the womb, even if the cortex is cut
off from sensory input during development. As development proceeds, different sets of genes are activated in different regions. The brain has a well-stocked toolbox of mechanisms to interconnect neurons, including molecules that attract or repel axons (the output fibers of neurons) to guide them to their targets, and molecules that glue them in place or ward them away. The number, size, and connectivity of cortical areas differ among species of mammals, and they differ between humans and other primates. This diversity is caused by genetic changes in the course of evolution that are beginning to be understood. 43 Geneticists recently discovered, for example, that different sets of genes are activated in the developing brain of humans and the developing brains of chimpanzees. 44
The possibility that cortical areas are specialized for different tasks has been obscured by the fact that different parts of the cortex look similar under a microscope. But because the brain is an information-processing system, that means little. The microscopic pits on a CD look the same regardless of what is recorded on it, and the strings of characters in different books look the same to someone who cannot read them. In an information-carrying medium, the content lies in combinatorial patterns among the elements -- in the case of the brain, the details of the microcircuitry -- and not in their physical appearance.
And the cortex itself is not the entire brain. Tucked beneath the cortex are other brain organs that drive important parts of human nature. They include the hippocampus, which consolidates memory and supports mental maps, the amygdala, which colors experience with certain emotions, and the hypothalamus, which originates sexual desire and other appetites. Many neuroscientists, even when they are impressed by the plasticity of the cortex, acknowledge that subcortical structures are far less plastic. 45 This is not a minor cavil about anatomy. Some commentators have singled out evolutionary psychology as a casualty of neural plasticity, saying that the changeability of the cortex proves that the brain cannot support evolutionary specializations. 46 But most proposals in evolutionary psychology are about drives like fear, sex, love, and {90} aggression, which reside largely in subcortical circuitry. More generally, on anyone's theory an innately shaped human ability would have to be implemented in a network of cortical and subcortical areas, not in a single patch of sensory cortex. ~
Another basic point about the brain has been lost in the recent enthusiasm for plasticity. A discovery that neural activity is crucial for brain development does not show either that learning is crucial in shaping the brain or that genes fail to shape the brain.
The study of neural development is often framed in terms of nature and nurture, but it is more fruitful to think of it as a problem in developmental biology -- how a ball of identical cells differentiates into a functioning organ. Doing so stands the conventional assumptions of associationism on their head. Primary sensory cortex, rather than being the firmest part of the brain on top of which successive stories can only be even more plastic, may be the part of the brain that is most dependent on the input for proper development.
In assembling a brain, a complete genetic blueprint is out of the question for two reasons. One is that a gene cannot anticipate every detail of the environment, including the environment consisting of the other genes in the genome. It has to specify an adaptive developmental program that ensures that the organism as a whole functions properly across variations in nutrition, other genes, growth rates over the lifespan, random perturbations, and the physical and social environment. And that requires feedback from the way the rest of the organism is developing.
Take the development of the body. The genes that build a femur cannot specify the exact shape of the ball on top, because the ball has to articulate with the socket in the pelvis, which is shaped by other genes, nutrition, age, and chance. So the ball and the socket adjust their shapes as they rotate against each other while the baby kicks in the womb. (We know this because experimental animals that are paralyzed while they develop end up with grossly deformed joints. ) Similarly, the genes shaping the lens of the growing eye cannot know how far back the retina is going to be or vice versa. So the brain of the baby is equipped with a feedback loop that uses signals about the sharpness of the image on the retina to slow down or speed up the physical growth of the eyeball. These are good examples of "plasticity," but the metaphor of plastic material is misleading. The mechanisms are not designed to allow variable environments to shape variable organs. They do the opposite: they ensure that despite variable environments, a constant organ develops, one that is capable of doing its job.
Like the body, the brain must use feedback circuits to shape itself into a working system. This is especially true in the sensory areas, which have to cope with growing sense organs. For that reason alone we would expect the activity
{91} of the brain to play a role in its own development, even if its end state, like those of the femur and the eyeball, is in some sense genetically specified. How this happens is still largely a mystery, but we know that patterns of neural stimulation can trigger the expression of a gene and that one gene can trigger many others. 47 Since every brain cell contains a complete genetic program, the machinery exists, in principle, for neural activity to trigger the development
? ? ? ? ? ? ? ? of an innately organized neural circuitry in any of several different regions. If so, brain activity would not be sculpting the brain; it would merely be telling the genome where in the brain a certain neural circuit should go.
So even an extreme innatist need not believe that the brain differentiates itself by the equivalent of GPS coordinates in the skull, following rules like "If you are between the left temple and the left ear, become a language circuit" (or a fear circuit, or a circuit for recognizing faces). A developmental program may be triggered in a part of the developing brain by some combination of the source of the stimulation, the firing pattern, the chemical environment, and other signals.
The point of this chapter is that these claims are not vindications of the doctrine of the Blank Slate but products of the Blank Slate. Many people (including a few scientists) have selectively read the evidence, sometimes in bizarre ways, to fit with a prior belief that the mind cannot possibly have any innate structure, or with simplistic notions of how innate structure, if it did exist, would be encoded in the genes and develop in the brain.
I should say at the outset that I find these latest-and-best blank-slate theories highly implausible -- indeed, barely coherent. Nothing comes out of nothing, and the complexity of the brain has to come from somewhere. It cannot come from the environment alone, because the whole point of having a brain is to accomplish certain goals, and the environment has no idea what those goals are. A given environment can accommodate organisms that build dams, migrate by the stars, trill and twitter to impress the females, scent-mark trees, write sonnets, and so on. To one species, a snatch of human speech is a warning to flee; to another, it is an interesting new sound to incorporate into its own vocal repertoire; to a third, it is grist for grammatical analysis. Information in the world doesn't tell you what to do with it.
Also, brain tissue is not some genie that can grant its owner any power that would come in handy. It is a physical mechanism, an arrangement of matter that converts inputs to outputs in particular ways. The idea that a single generic substance can see in depth, control the hands, attract a mate, bring up children, elude predators, outsmart prey, and so on, without some degree of specialization, is not credible. Saying that the brain solves these problems because of its "plasticity" is not much better than saying it solves them by magic. Still, in this chapter I will examine the latest scientific objections to human nature carefully. Each of the discoveries is important on its own terms, even if it does not support the extravagant conclusions that have been drawn. And once the last supports for the Blank Slate have been evaluated, I can properly sum up the scientific case for the alternative.
The human genome is often seen as the essence of our species, so it is not surprising that when its sequence was announced in 2001 commentators rushed to give it the correct interpretation for human affairs. Craig Venter, {76} whose company had competed with a public consortium in the race to sequence the genome, said at a press conference that the smaller-than-expected gene count shows that "we simply do not have enough genes for this idea of biological determinism to be right. The wonderful diversity of the human species is not hard-wired in our genetic code. Our environments are critical. " In the United Kingdom, The Guardian headlined its story, "Revealed: The Secret of Human Behaviour. Environment, Not Genes, Key to Our Acts. "3 An editorial in another British newspaper concluded that "we are more free, it seems, than we had realized. " Moreover, the finding "offers comfort for the left, with its belief in the potential of all, however deprived their background. But it is damning for the right, with its fondness for ruling classes and original sin. "4
All this from the number 34,000! Which leads to the question, What number of genes would have proven that the diversity of our species was wired into our genetic code, or that we are less free than we had realized, or that the political right is right and the left is wrong? 50,000? 150,000? Conversely, if it turned out that we had only 20,000 genes, would that have made us even freer, or the environment even more important, or the political left even more comfortable? The fact is that no one knows what these numbers mean. No one has the slightest idea how many genes it would take to build a system of hard-wired modules, or a general-purpose learning program, or anything in between -- to say nothing of original sin or the superiority of the ruling class. In our current state of ignorance of how the genes build a brain, the number of genes in the human genome is just a number.
If you don't believe this, consider the roundworm Caenorhabditis elegans, which has about 18,000 genes. By the logic of the genome editorialists, it should be twice as free, be twice as diverse, and have twice as much potential as a human being. In fact, it is a microscopic worm composed of 959 cells grown by a rigid genetic program, with a nervous system consisting of exactly 302 neurons in a fixed wiring diagram. As far as behavior is concerned, it eats, mates, approaches and avoids certain smells, and that's about it. This alone should make it obvious that our freedom and diversity of behavior come from having a complex biological makeup, not a simple one.
Now, it is a genuine puzzle why humans, with their hundred trillion cells and hundred billion neurons, need only twice as many genes as a humble little worm. Many biologists believe that the human genes have been undercounted. The number of genes in a genome can only be estimated; right now they cannot literally be totted up. Gene- estimating programs look for sequences in the DNA that are similar to known genes and that are active enough to be caught in the act of building a protein. 5 Genes that are unique to humans or active only in the developing brain of the
? ? ? ? ? ? fetus -- the genes most relevant to human nature -- and other inconspicuous genes could evade the software and get left {77} out of the estimates. Alternative estimates of 57,000,75,000, and even 120,000 human genes are currently being bruited about. 6 Still, even if humans had six times as many genes as a roundworm rather than just twice as many, the puzzle would remain.
Most biologists who are pondering the puzzle don't conclude that humans are less complex than we thought. Instead they conclude that the number of genes in a genome has little to do with the complexity of the organism. 7 A single
gene does not correspond to a single component in such a way that an organism with 20,000 genes has 20,000 components, an organism with 30,000 genes has 30,000 components, and so on. Genes specify proteins, and some of the proteins do become the meat and juices of an organism. But other proteins turn genes on or off, speed up or slow down their activity, or cut and splice other proteins into new combinations. James Watson points out that we should recalibrate our intuitions about what a given number of genes can do: "Imagine watching a play with thirty thousand actors. You'd get pretty conrused. "
Depending on how the genes interact, the assembly process can be much more intricate for one organism than for another with the same number of genes. In a simple organism, many of the genes simply build a protein and dump it into the stew. In a complex organism, one gene may turn on a second one, which speeds up the activity of a third one (but only if a fourth one is active), which then turns off the original gene (but only if a fifth one is inactive), and so on. This defines a kind of recipe that can build a more complex organism out of the same number of genes. The complexity of an organism thus depends not just on its gene count but on the intricacy of the box-and-arrow diagram that captures how each gene impinges on the activity of the other genes. 8 And because adding a gene doesn't just add an ingredient but can multiply the number of ways that the genes can interact with one another, the complexity of organisms depends on the number of possible combinations of active and inactive genes in their genomes. The geneticist Jean-Michel Claverie suggests that it might be estimated by the number two (active versus inactive) raised to the power of the number of genes. By that measure, a human genome is not twice as complex as a roundworm genome but 216,000 (a one followed by 4. 800 zeroes) times as complex. 9
There are two other reasons why the complexity of the genome is not reflected in the number of genes it contains. One is that a given gene can produce not just one protein but several. A gene is typically broken into stretches of DNA that code for fragments of protein (exons) separated by stretches of DNA that don't (introns), a bit like a magazine article interrupted by ads. The segments of a gene can then be spliced together in multiple ways. A gene composed of exons A, B, C, and D might give rise to proteins corresponding to {78} ABC, ABD, ACD, and so on -- as many as ten different proteins per gene. This happens to a greater degree in complex organisms than in simple ones. 10
Second, the 34,000 genes take up only about 3 percent of the human genome. The rest consists of DNA that does not code for protein and that used to be dismissed as "junk. " But as one biologist recently put it, "The term 'junk DNA' is a reflection of our ignorance. "11 The size, placement, and content of the noncoding DNA can have dramatic effects on the way that nearby genes are activated to make proteins. Information in the billions of bases in the non-coding regions of the genome is part of the specification of a human being, above and beyond the information contained in the 34,000 genes.
The human genome, then, is fully capable of building a complex brain, in spite of the bizarre proclamations of how wonderful it is that people are almost as simple as worms. Of course "the wonderful diversity of the human species is not hard-wired in our genetic code," but we didn't need to count genes to figure that out -- we already know it from the fact that a child growing up in Japan speaks Japanese but the same child growing up in England would speak English. It is an example of a syndrome we will meet elsewhere in this book: scientific findings spin-doctored beyond recognition to make a moral point that could have been made more easily on other grounds.
The second scientific defense of the Blank Slate comes from connectionism, the theory that the brain is like the artificial neural networks simulated on computers to learn statistical patterns. 12
Cognitive scientists agree that the elementary processes that make up the instruction set of the brain -- storing and retrieving an association, sequencing elements, focusing attention -- are implemented in the brain as networks of densely interconnected neurons (brain cells). The question is whether a generic kind of network, after being shaped by the environment, can explain all of human psychology, or whether the genome tailors different networks to the demands of particular domains: language, vision, morality, fear, lust, intuitive psychology, and so on. The connectionists, of course, do not believe in a blank slate, but they do believe in the closest mechanistic equivalent, a general-purpose learning device.
What is a neural network? Connectionists use the term to refer not to real neural circuitry in the brain but to a kind of computer program based on the metaphor of neurons and neural circuits. In the most common approach, a "neuron" carries information by being more or less active. The activity level indicates the presence or absence (or intensity or degree of confidence) of a simple feature of the world. The feature may be a color, a line with a certain slant, a letter
? ? ? ? ? ? ? ? ? of the alphabet, or a property of an animal such as having four legs.
A network of neurons can represent different concepts, depending on {79} which ones are active. If neurons for "yellow," "flies," and "sings" are active, the network is thinking about a canary; if neurons for "silver," "flies," and "roars" are active, it is thinking about an airplane. An artificial neural network computes in the following manner. Neurons are linked to other neurons by connections that work something like synapses. Each neuron counts up the inputs from other neurons and changes its activity level in response. The network learns by allowing the input to change the strengths of the connections. The strength of a connection determines the likelihood that the input neuron will excite or inhibit the output neuron.
Depending on what the neurons stand for, how they are innately wired, and how the connections change with training, a connectionist network can learn to compute various things. If everything is connected to everything else, a network can soak up the correlations among features in a set of objects. For example, after exposure to descriptions of many birds it can predict that feathered singing things tend to fly or that feathered flying things tend to sing or that singing flying things tend to have feathers. If a network has an input layer connected to an output layer, it can learn associations between ideas, such as that small soft flying things are animals but large metallic flying things are vehicles. If its output layer feeds back to earlier layers, it can crank out ordered sequences, such as the sounds making up a word.
The appeal of neural networks is that they automatically generalize their training to similar new items. If a network has been trained that tigers eat Frosted Flakes, it will tend to generalize that lions eat Frosted Flakes, because eating Frosted Flakes" has been associated not with "tigers" but with simpler features like "roars" and "has whiskers," which make up part of the representation of h'ons, too. The school of connectionism, like the school of associationism championed by Locke, Hume, and Mill, asserts that these generalizations are the crux of intelligence. If so, highly trained but otherwise generic neural networks can explain intelligence.
Computer modelers often set their models on simplified toy problems to ? rove that they can work in principle. The question then becomes whether the models can "scale up" to more realistic problems, or whether, as skeptics say, the modeler "is climbing trees to get to the moon. " Here we have the problem with connectionism. Simple connectionist networks can manage impressive displays of memory and generalization in circumscribed problems like reading a list of words or learning stereotypes of animals. But they are simply too underpowered to duplicate more realistic feats of human intelligence like understanding a sentence or reasoning about living things.
Humans don't just loosely associate things that resemble each other, or things that tend to occur together. They have combinatorial minds that enterrain propositions about what is true of what, and about who did what to {80} whom, when and where and why. And that requires a computational architecture that is more sophisticated than the uniform tangle of neurons used in generic connectionist networks. It requires an architecture equipped with logical apparatus like rules, variables, propositions, goal states, and different kinds of data structures, organized into larger systems. Many cognitive scientists have made this point, including Gary Marcus, Marvin Minsky, Seymour Papert, Jerry Fodor, Zenon Pylyshyn, John Anderson, Tom Bever, and Robert Hadley, and it is acknowledged as well by neural network modelers who are not in the connectionist school, such as John Hummel, Lokendra Shastri, and Paul Smolensky. 13 I have written at length on the limits of connectionism, both in scholarly papers and in popular books; here is a summary of my own case. 14
In a section called "Connectoplasm" in How the Mind Works, I laid out some simple logical relationships that underlie our understanding of a complete thought (such as the meaning of a sentence) but that are difficult to represent in generic networks. 15 One is the distinction between a kind and an individual: between ducks in general and this duck in particular. Both have the same features (swims, quacks, has feathers, and so on), and both are thus represented by the same set of active units in a standard connectionist model. But people know the difference.
A second talent is compositionality: the ability to entertain a new, complex thought that is not just the sum of the simple thoughts composing it but depends on their relationships. The thought that cats chase mice, for example, cannot be captured by activating a unit each for "cats," "mice," and "chase," because that pattern could just as easily stand for mice chasing cats.
A third logical talent is quantification (or the binding of variables): the difference between fooling some of the people all of the time and fooling all of the people some of the time. Without the computational equivalent of x's,y's, parentheses, and statements like "For all x," a model cannot tell the difference.
A fourth is recursion: the ability to embed one thought inside another, so that we can entertain not only the thought that Elvis lives, but the thought that the National Enquirer reported that Elvis lives, that some people believe the National Enquirer report that Elvis lives, that it is amazing that some people believe the National Enquirer report that Elvis lives, and so on. Connectionist networks would superimpose these propositions and thereby confuse their various subjects and predicates.
A final elusive talent is our ability to engage in categorical, as opposed to fuzzy, reasoning: to understand that Bob
? ? ? ? ? Dylan is a grandfather, even though he is not very grandfatherly, or that shrews are not rodents, though they look just like mice. With nothing but a soup of neurons to stand for an object's properties, and no provision for rules, variables, and definitions, the networks fall back on stereotypes and are bamboozled by atypical examples. {81}
In Words and Rules I aimed a microscope on a single phenomenon of language that has served as a test case for the ability of generic associative networks to account for the essence of language: assembling words, or pieces of words, into new combinations. People don't just memorize snatches of language but create new ones. A simple example is the English past tense. Given a neologism like to spam or to snarf, people don't have to run to the dictionary to look up their past-tense forms; they instinctively know that they are spammed and snarfed. The talent for assembling new combinations appears as early as age two, when children overapply the past-tense suffix to irregular verbs, as in We holded the baby rabbits and Horton heared a Who. 16
The obvious way to explain this talent is to appeal to two kinds of computational operations in the mind. Irregular forms like held and heard are stored in and retrieved from memory, just like any other word. Regular forms like walk- walked can be generated by a mental version of the grammatical rule "Add -ed to the verb. " The rule can apply whenever memory fails. It may be used when a word is unfamiliar and no past-tense form had been stored in memory, as in to spam, and it may be used by children when they cannot recall an irregular form like heard and need some way of marking its tense. Combining a suffix with a verb is a small example of an important human talent: combining words and phrases to create new sentences and thereby express new thoughts. It is one of the new ideas of the cognitive revolution introduced in Chapter 3, and one of the logical challenges for connectionism I listed in the preceding discussion.
Connectionists have used the past tense as a proving ground to see if they could duplicate this textbook example of human creativity without using a rule and without dividing the labor between a system for memory and a system for grammatical combination. A series of computer models have tried to generate past-tense forms using simple pattern associator networks. The networks typically connect the sounds in verbs with the sounds in the past-tense form: -am with -ammed, -ing with -ung, and so on. The models can then generate new forms by analogy, just like the generalization from tigers to lions: trained on crammed, a model can guess spammed; trained on folded, it tends to say holded.
But human speakers do far more than associate sounds with sounds, and the models thus fail to do them justice. The failures come from the absence of machinery to handle logical relationships. Most of the models are baffled by new words that sound different from familiar words and hence cannot be generalized by analogy. Given the novel verb to frilg, for example, they come up not with frilged, as people do, but with an odd mishmash like freezled. That is because they lack the device of a variable, like x in algebra or "verb" in grammar, which can apply to any member of a category, regardless of how familiar {82} its properties are. (This is the gadget that allows people to engage in categorical rather than fuzzy reasoning. ) The networks can only associate bits of sound with bits of sound, so when confronted with a new verb that does not sound like anything they were trained on, they assemble a pastiche of the most similar sounds they can find in their network.
The models also cannot properly distinguish among verbs that have the same sounds but different past-tense forms, such as ring the bell-rang the bell and ring the city-ringed the city. That is because the standard models represent
only sound and are blind to the grammatical differences among verbs that call for different conjugations. The key difference here is between simple roots like ring in the sense of "resonate" (past tense rang) and complex verbs derived from nouns like ring in the sense of "form a ring around" (past tense ringed). To register that difference, a language-using system has to be equipped with compositional data structures (such as "a verb made from the noun ring") and not just a beanbag of units.
Yet another problem is that connectionist networks track the statistics of the input closely: how many verbs of each sound pattern they have encountered. That leaves them unable to account for the epiphany in which young children discover the -ed rule and start making errors like holded and heared. Connectionist modelers can induce these errors only by bombarding the network with regular verbs (so as to burn in the -ed) in a way that is unlike anything real children experience. Finally, a mass of evidence from cognitive neuroscience shows that grammatical combination (including regular verbs) and lexical lookup (including irregular verbs) are handled by different systems in the brain rather than by a single associative network.
It's not that neural networks are incapable of handling the meanings of sentences or the task of grammatical conjugation. (They had better not be, since the very idea that thinking is a form of neural computation requires that some kind of neural network duplicate whatever the mind can do. ) The problem lies in the credo that one can do everything with a generic model as long as it is sufficiently trained. Many modelers have beefed up, retrofitted, or combined networks into more complicated and powerful systems. They have dedicated hunks of neural hardware to abstract symbols like "verb phrase" and "proposition" and have implemented additional mechanisms (such as synchronized firing patterns) to bind them together in the equivalent of compositional, recursive symbol structures.
? ? ? ? They have installed banks of neurons for words, or for English suffixes, or for key grammatical distinctions. They have built hybrid systems, with one network that retrieves irregular forms from memory and another that combines a verb with a suffix. 17
A system assembled out of beefed-up subnetworks could escape all the criticisms. But then we would no longer be talking about a generic neural network! We would be talking about a complex system innately tailored to {83} compute a task that people are good at. In the children's story called "Stone Soup," a hobo borrows the use of a woman's kitchen ostensibly to make soup from a stone. But he gradually asks for more and more ingredients to balance the flavor until he has prepared a rich and hearty stew at her expense. Connectionist modelers who claim to build intelligence out of generic neural networks without requiring anything innate are engaged in a similar business. The design choices that make a neural network system smart -- what each of the neurons represents, how they are wired together, what kinds of networks are assembled into a bigger system, in which way -- embody the innate organization of the part of the mind being modeled. They are typically hand-picked by the modeler, like an inventor rummaging through a box of transistors and diodes, but in a real brain they would have evolved by natural selection (indeed, in some networks, the architecture of the model does evolve by a simulation of natural selection). 18 The only alternative is that some previous episode of learning left the networks in a state ready for the current learning, but of course the buck has to stop at some innate specification of the first networks that kick off the learning process.
So the rumor that neural networks can replace mental structure with statistical learning is not true. Simple, generic networks are not up to the demands of ordinary human thinking and speaking; complex, specialized networks are a stone soup in which much of the interesting work has been done in setting up the innate wiring of the network. Once this is recognized, neural network modeling becomes an indispensable complement to the theory of a complex human nature rather than a replacement for it. 19 It bridges the gap between the elementary steps of cognition and the physiological activity of the brain and thus serves as an important link in the long chain of explanation between biology and culture. ~
FOR MOST OF its history, neuroscience was faced with an embarrassment: the brain looked as if it were innately specified in every detail. When it comes to the body, we can see many of the effects of a person's life experience: it may be tanned or pale, callused or soft, scrawny or plump or chiseled. But no such marks could be found in the brain. Now, something has to be wrong with this picture. People learn, and learn massively: they learn their language, their culture, their know-how, their database of facts. Also, the hundred trillion connections in the brain cannot possibly be specified individually by a 750-megabyte genome. The brain somehow must change in response to its input; the only question is how.
We are finally beginning to understand how. The study of neural plasticity is hot. Almost every week sees a discovery about how the brain gets wired in the womb and tuned outside it. After all those decades in which no one could find anything that changed in the brain, it is not surprising that the {84} discovery of plasticity has given the nature-nurture pendulum a push. Some people describe plasticity as a harbinger of an expansion of human potential in which the powers of the brain will be harnessed to revolutionize childrearing, education, therapy, and aging. And several manifestos have proclaimed that plasticity proves that the brain cannot have any significant innate organization. 20 In Rethinking Innateness, Jeffrey Elman and a team of West Pole connectionists write that predispositions to think about different things in different ways (language, people, objects, and so on) may be implemented in the brain only as "attention-grabbers" that ensure that the organism will receive "massive experience of certain inputs prior to subsequent learning. "21 In a "constructivist manifesto," the theoretical neuroscientists Stephen Quartz and Terrence Sejnowski write that "although the cortex is not a tabula rasa . . . it is largely equipotential at early stages," and therefore that innatist theories "appear implausible. "22
Neural development and plasticity unquestionably make up one of the great frontiers of human knowledge. How a linear string of DNA can direct the assembly of an intricate three-dimensional organ that lets us think, feel, and learn is a problem to stagger the imagination, to keep neuroscientists engaged for decades, and to belie any suggestion that we are approaching "the end of science. "
And the discoveries themselves are fascinating and provocative. The cerebral cortex (outer gray matter) of the brain has long been known to be divided into areas with different functions. Some represent particular body parts; others represent the visual field or the world of sound; still others concentrate on aspects of language or thinking. We now know that with learning and practice some of their boundaries can move around. (This does not mean that the brain tissue literally grows or shrinks, only that if the cortex is probed with electrodes or monitored with a scanner, the boundary where one ability leaves off and the next one begins can shift. ) Violinists, for example, have an expanded region of cortex representing the fingers of the left hand. 23 If a person or a monkey is trained on a simple task like recognizing shapes or attending to a location in space, neuroscientists can watch as parts of the cortex, or even
? ? ? ? ? ? ? ? ? individual neurons, take on the job. 24
The reallocation of brain tissue to new tasks is especially dramatic when people lose the use of a sense or body part. Congenitally blind people use their visual cortex to read Braille. 25 Congenitally deaf people use part of their auditory cortex to process sign language. 26 Amputees use the part of the cortex formerly serving the missing limb to represent other parts of their bodies. 27 Young children can grow up relatively normal after traumas to the brain that would turn adults into basket cases -- even removal of the entire left hemisphere, which in adults underlies language and logical reasoning. 28 All this {85} suggests that the allocation of brain tissue to perceptual and cognitive processes is not done permanently and on the basis of the exact location of the tissue in the skull, but depends on how the brain itself processes information.
This dynamic allocation of tissue can also be seen as the brain puts itself together in the womb. Unlike a computer that gets assembled in a factory and is turned on for the first time when complete, the brain is active while it is being assembled, and that activity may take part in the assembly process. Experiments on cats and other mammals have shown that if a brain is chemically silenced during fetal development it may end up with significant abnormalities. 29 And patches of cortex develop differently depending on the kind of input they receive. In an experimental tour de force, the neuroscientist Mriganka Sur literally rewired the brains of ferrets so that signals from their eyes fed into the primary auditory cortex, the part of the brain that ordinarily receives signals from the ears. 30 When he then probed the auditory cortex with electrodes, he found that it acted in many ways like the visual cortex. Locations in the visual field were laid out like a map, and individual neurons responded to lines and stripes at a particular orientation and direction of movement, similar to the neurons in an ordinary visual cortex. The ferrets could even use their rewired brains to move toward objects that were detectable by sight alone. The input to the sensory cortex must help to organize it: visual input makes the auditory cortex work something like the visual cortex.
What do these discoveries mean? Do they show that the brain is "able to be shaped, molded, modeled, or sculpted," as the dictionary definition of plastic would suggest? In the rest of this chapter I will show you that the answer is no. 31 Discoveries of how the brain changes with experience do not show that learning is more powerful than we thought, that the brain can be dramatically reshaped by its input, or that the genes do not shape the brain. Indeed, demonstrations of the plasticity of the brain are less radical than they first appear: the supposedly plastic regions of cortex are doing pretty much the same thing they would have been doing if they had never been altered. And the most recent discoveries on brain development have refuted the idea that the brain is largely plastic. Let me go over these points in turn.
The fact that the brain changes when we learn is not, as some have claimed, a radical discovery with profound implications for nature and nurture or human potential. Dmitri Karamazov could have deduced it in his nineteenth- century prison cell as he mulled over the fact that thinking comes from quivering nerve tails rather than an immaterial soul. If thought and action are products of the physical activity of the brain, and if thought and action can be affected by experience, then experience has to leave a trace in the physical structure of the brain. {86}
So there is. no scientific question as to whether experience, learning, and practice affect the brain; they surely do if we are even vaguely on the right track. It is not surprising that people who can play the violin have different brains from those who cannot, or that masters of sign language or of Braille have different brains from people who speak and read. Your brain changes when you are introduced to a new person, when you hear a bit of gossip, when you watch the Oscars, when you polish your golf stroke -- in short, whenever an experience leaves a trace in the mind. The only question is how learning affects the brain. Are memories stored in protein sequences, in new neurons or synapses, or in changes in the strength of existing synapses? When someone learns a new skill, is it stored only in organs dedicated to learning skills (like the cerebellum and the basal ganglia), or does it also adjust the cortex? Does an increase in dexterity depend on using more square centimeters of cortex or on using a greater concentration of synapses in the same number of square centimeters? These are important scientific problems, but they say nothing about whether people can learn, or how much. We already knew trained violinists play better than beginners or we would never have put their heads in the scanner to begin with. Neural plasticity is just another name for learning and development, described at a different level of analysis.
All this should be obvious, but nowadays any banality about learning can be dressed up in neurospeak and treated like a great revelation of science. According to a New York Times headline, "Talk therapy, a psychiatrist maintains, can alter the structure of the patient's brain. "32 I should hope so, or else the psychiatrist would be defrauding her clients. "Environmental manipulation can change the way [a child's] brain develops," the pediatric neurologist Harry Chugani told the Boston Globe. "A child surrounded by aggression, violence, or inadequate stimulation will reflect these connections in the brain and behavior. "33 Well, yes; if the environment affects the child at all, it would do so by changing connections in the brain. A special issue of the journal Educational Technology and Society was intended "to examine the position that learning takes place in the brain of the learner, and that pedagogies and technologies
? ? ? ? ? ? ? ? ? ? ? ? should be designed and evaluated on the basis of the effect they have on student brains. " The guest editor (a
biologist) did not say whether the alternative was that learning takes place in some other organ of the body like the pancreas or that it takes place in an immaterial soul. Even professors of neuroscience sometimes proclaim "discoveries" that would be news only to believers in a ghost in the machine: "Scientists have found that the brain is capable of altering its connections. . . . . . You have the ability to change the synaptic connections within the brain. "34 Good thing, because otherwise we would be permanent amnesiacs.
This neuroscientist is an executive at a company that "uses brain research {87} and technology to develop products intended to enhance human learning and performance," one of many new companies with that aspiration. "The
human being has unlimited creativity if focused and nurtured properly," says a consultant who teaches clients to draw diagrams that "map their neural patterns. " "The older you get, the more connections and associations your brain should be making," said a satisfied customer; "Therefore you should have more information stored in your brain.
You just need to tap into it. "35 Many people have been convinced by the public pronouncements of neuroscience advocates -- on the basis of no evidence whatsoever -- that varying the route you take when driving home can stave off the effects of aging. 36 And then there is the marketing genius who realized that blocks, balls, and other toys "provide visual and tactile stimulation" and "encourage movement and tracking," part of a larger movement of "brain- based" childrearing and education that we will meet again in the chapter on children. 37
These companies tap into people's belief in a ghost in the machine by implying that any form of learning that affects the brain (as opposed, presumably, to the kinds of learning that don't affect the brain) is unexpectedly real or deep or powerful. But this is mistaken. All learning affects the brain. It is undeniably exciting when scientists make a discovery about how learning affects the brain, but that does not make the learning itself any more pervasive or profound. ~
A second misinterpretation of neural plasticity can be traced to the belief that there is nothing in the mind that was not first in the senses. The most highly publicized discoveries about cortical plasticity concern primary sensory cortex, the patches of gray matter that first receive signals from the senses (via the thalamus and other subcortical organs). Writers who use plasticity to prop up the Blank Slate assume that if primary sensory cortex is plastic, the rest of the brain must be even more plastic, because the mind is built out of sensory experience. For example, one neuroscientist was quoted as saying that Sur's rewiring experiments "challenge the recent emphasis on the power of the genes" and "will push people back toward more consideration of environmental factors in creating normal brain organization. "38 But if the brain is a complex organ with many parts, the moral does not follow. Primary sensory cortex is not the bedrock of the mind but a gadget, one of many in the brain, that happens to be specialized for certain kinds of signal processing in the first stages of sensory analysis. Let's suppose that primary sensory cortex really were formless, getting all its structure from the input. Would that mean that the entire brain is formless and gets all of its structure from the input? Not at all. For one thing, even primary sensory cortex is just one part of a huge, intricate system. To put things in perspective, here is a recent diagram of the wiring of the primate visual system:39 {88}
? ? ? ? ? ? ? ? ? Primary visual cortex is the box near the bottom labeled "V1. " It is one of at least fifty distinct brain areas devoted to visual processing, and they are interconnected in precise ways. (Despite the spaghetti-like appearance, not everything is connected to everything else. Only about a third of the logically possible connections between components are actually present in the brain. ) Primary visual cortex, by itself, is not enough to see with. Indeed, it is so deeply buried in the visual system that Francis Crick and the neuroscientist Christof Koch have argued that we are not conscious of anything that goes on in it. 40 What we see -- familiar colored objects arranged in a scene or moving in {89} particular ways -- is a product of the entire contraption. So even if the innards of the V1 box were completely specified by its input, we would have to explain the architecture of the rest of the visual system -- the fifty boxes and their connections. I don't mean to imply that the entire block diagram is genetically specified, but much of it almost certainly is. 41
And of course the visual system itself must be put into perspective, because it is just one part of the brain. The visual system dominates some half-dozen of the more than fifty major areas of the cortex that can be distinguished by their anatomy and connections. Many of the others underlie other functions such as language, reasoning, planning, and
? ? ? social skills. Though no one knows to what extent they are genetically prepared for their computational roles, there are hints that the genetic influence is substantial. 42 The divisions are established in the womb, even if the cortex is cut
off from sensory input during development. As development proceeds, different sets of genes are activated in different regions. The brain has a well-stocked toolbox of mechanisms to interconnect neurons, including molecules that attract or repel axons (the output fibers of neurons) to guide them to their targets, and molecules that glue them in place or ward them away. The number, size, and connectivity of cortical areas differ among species of mammals, and they differ between humans and other primates. This diversity is caused by genetic changes in the course of evolution that are beginning to be understood. 43 Geneticists recently discovered, for example, that different sets of genes are activated in the developing brain of humans and the developing brains of chimpanzees. 44
The possibility that cortical areas are specialized for different tasks has been obscured by the fact that different parts of the cortex look similar under a microscope. But because the brain is an information-processing system, that means little. The microscopic pits on a CD look the same regardless of what is recorded on it, and the strings of characters in different books look the same to someone who cannot read them. In an information-carrying medium, the content lies in combinatorial patterns among the elements -- in the case of the brain, the details of the microcircuitry -- and not in their physical appearance.
And the cortex itself is not the entire brain. Tucked beneath the cortex are other brain organs that drive important parts of human nature. They include the hippocampus, which consolidates memory and supports mental maps, the amygdala, which colors experience with certain emotions, and the hypothalamus, which originates sexual desire and other appetites. Many neuroscientists, even when they are impressed by the plasticity of the cortex, acknowledge that subcortical structures are far less plastic. 45 This is not a minor cavil about anatomy. Some commentators have singled out evolutionary psychology as a casualty of neural plasticity, saying that the changeability of the cortex proves that the brain cannot support evolutionary specializations. 46 But most proposals in evolutionary psychology are about drives like fear, sex, love, and {90} aggression, which reside largely in subcortical circuitry. More generally, on anyone's theory an innately shaped human ability would have to be implemented in a network of cortical and subcortical areas, not in a single patch of sensory cortex. ~
Another basic point about the brain has been lost in the recent enthusiasm for plasticity. A discovery that neural activity is crucial for brain development does not show either that learning is crucial in shaping the brain or that genes fail to shape the brain.
The study of neural development is often framed in terms of nature and nurture, but it is more fruitful to think of it as a problem in developmental biology -- how a ball of identical cells differentiates into a functioning organ. Doing so stands the conventional assumptions of associationism on their head. Primary sensory cortex, rather than being the firmest part of the brain on top of which successive stories can only be even more plastic, may be the part of the brain that is most dependent on the input for proper development.
In assembling a brain, a complete genetic blueprint is out of the question for two reasons. One is that a gene cannot anticipate every detail of the environment, including the environment consisting of the other genes in the genome. It has to specify an adaptive developmental program that ensures that the organism as a whole functions properly across variations in nutrition, other genes, growth rates over the lifespan, random perturbations, and the physical and social environment. And that requires feedback from the way the rest of the organism is developing.
Take the development of the body. The genes that build a femur cannot specify the exact shape of the ball on top, because the ball has to articulate with the socket in the pelvis, which is shaped by other genes, nutrition, age, and chance. So the ball and the socket adjust their shapes as they rotate against each other while the baby kicks in the womb. (We know this because experimental animals that are paralyzed while they develop end up with grossly deformed joints. ) Similarly, the genes shaping the lens of the growing eye cannot know how far back the retina is going to be or vice versa. So the brain of the baby is equipped with a feedback loop that uses signals about the sharpness of the image on the retina to slow down or speed up the physical growth of the eyeball. These are good examples of "plasticity," but the metaphor of plastic material is misleading. The mechanisms are not designed to allow variable environments to shape variable organs. They do the opposite: they ensure that despite variable environments, a constant organ develops, one that is capable of doing its job.
Like the body, the brain must use feedback circuits to shape itself into a working system. This is especially true in the sensory areas, which have to cope with growing sense organs. For that reason alone we would expect the activity
{91} of the brain to play a role in its own development, even if its end state, like those of the femur and the eyeball, is in some sense genetically specified. How this happens is still largely a mystery, but we know that patterns of neural stimulation can trigger the expression of a gene and that one gene can trigger many others. 47 Since every brain cell contains a complete genetic program, the machinery exists, in principle, for neural activity to trigger the development
? ? ? ? ? ? ? ? of an innately organized neural circuitry in any of several different regions. If so, brain activity would not be sculpting the brain; it would merely be telling the genome where in the brain a certain neural circuit should go.
So even an extreme innatist need not believe that the brain differentiates itself by the equivalent of GPS coordinates in the skull, following rules like "If you are between the left temple and the left ear, become a language circuit" (or a fear circuit, or a circuit for recognizing faces). A developmental program may be triggered in a part of the developing brain by some combination of the source of the stimulation, the firing pattern, the chemical environment, and other signals.
