It's a Jungle in There: How Competition and Cooperation in the Brain Shape the Mind (23 page)

What accounts for the superiority of spaced over massed practice? For one thing, nonstop practice is tiring. Whether it’s your muscles that ache or your head that throbs, after concerted effort on a task, you can feel yourself fading.

Interleaving different kinds of practice—switching between learning to type and learning to fiddle, say—also exposes you to more kinds of information. As a result, spacing leaves you prepared for more kinds of challenges. That’s undoubtedly helpful in the outer jungle, where you never know exactly what challenges you’ll face.

But the most important reason for the superiority of spaced over massed practice is one that fits with the jungle principle. Spaced practice lets some internal agents get strong at the expense of other internal agents, without letting any internal agents get so strong that they “kill off” the others, leaving them unable to deal with the challenges they would handle otherwise. Becoming a complete master of one task rather than a jack-of-all-trades at several can leave you insufficiently prepared for the challenges that may
come along for the tasks that are les practiced. The brain’s internal dynamics favor balance between specialization and diversification.

Another seeming limitation of trial-and-error learning is that it appears to ignore innate dispositions and preferences. The domain where this seeming failure has played out most dramatically is in the debate about whether language learning relies mainly on innate mechanisms or on learning based on trial and error. The latter idea was advanced by the American behaviorist B. F. Skinner.
³²
The former idea was advanced by the American linguist Noam Chomsky.
³³
No issue in cognitive psychology has sparked more heat than the debate between these two thinkers and their disciples.

Chomsky’s view leads to the position that trial-and-error learning cannot explain language learning in children. Consistent with this idea, Roger Brown, a psychologist at Harvard, reported that parents do not reward children for speaking grammatically; instead, they reward children for speaking agreeably.
³⁴
Brown found, in effect, that when a child said, “Wuv you, Mommy,” the mom is more likely to come back with, “Mommy wuv you, too” than she is to scold for the child for speaking unprofessionally. When Sarah, my daughter, declared at the age of 4 that she wanted to go “tennising”—saying this while holding a tennis racquet that was nearly as large as she was—my wife and I didn’t correct her English. Instead, we beamed at her cuteness. Through her language, Sarah showed that she had over-generalized a rule of English. Over-generalization is hard to explain in terms of trial-and-error learning, a point emphasized by Chomsky and his followers.
³⁵
Chomskians argue that over-generalization of grammatical rules reflects an innate affinity for those rules. On this view, very little input is needed to trigger the rules, which, once triggered, obviate trial-and-error learning.
³⁶

Grammatical rules aren’t the only kinds of knowledge that can be acquired with little input, however. Semantic concepts can be acquired this way as well. For example, after just a few exposures to pictures of elephants, children can learn that the word “elephant” refers to a large-nosed pachyderm rather than some other sort of animals. How can such learning occur?

To pursue this question, first consider that if trial-and-error learning helps you learn concepts, then this capacity reduces the need to appeal to less direct explanations, like saying genes alone account for these abilities. You do, of course, have language-related genes in the sense that you, as a human being,
are more likely to develop the capacity for language than to develop the capacity for flying or the capacity for photosynthesis. But saying you have language genes begs the question of what neural mechanisms, allowed by those genes, let you become a language user. Explicating those neural mechanisms remains a supreme challenge of science.
³⁷

Second, it turns out that there is a way that trial and error learning can account for the capacity to learn at surprisingly rapid rates, even with sparse inputs. The method relies on a powerful statistical principle called Bayes’ Rule.

Thomas Bayes was an eighteenth-century Presbyterian minister whose rule relates four quantities: (1) the probability of an event, (2) the probability of the source of the event, (3) the probability of the event given the source, and (4) the probability of the source given the event. The formula is simple, encouraging the belief that students in statistics courses can learn it and, more importantly, that neural networks can embody it.
³⁸

If neural networks embody Bayesian statistics, then learning can be hastened by exploiting Bayes’ Rule. For example, by having an estimate of how likely pachyderms are and how likely the word “elephant” is, a child can use Bayes’ Rule to determine the probability that a new word such as “elephant” refers to pachyderms rather than other beasts. It turns out that Bayesian inference—that is, reliance on Bayes’ Rule to determine which source is most likely—may make it possible to induce linguistic rules to learn new words and concepts and to perform perceptual-motor tasks, even when sensory conditions change dramatically.

Regarding perceptual-motor tasks (those being more closely aligned with the kinds of tasks that have mainly been discussed in this chapter), suppose Sarah, now grown up, is on a tennis court. She’s out there enjoying the game, but a fog suddenly descends on the court and the scene is less visible than it was before. It turns out that Sarah and her tennis partner can continue to play despite the reduced visibility. They can do this thanks to their ability to exploit Bayesian statistics, as shown in an experiment that used a computerized analogue of a tennis game played in conditions of high or low visibility. Without Bayesian inference, the participants in the study might have been unable to adapt as well as they did, but with Bayesian inference—with the ability to use Bayes’ Rule implicitly—they could easily adjust to conditions analogous to the sudden settling of a fog.
³⁹

Given the power of Bayesian inference, interest in the topic has swelled.
⁴⁰
Scientists who endorse the view that people and animals behave as if they rely on Bayesian inference believe that people and animals don’t
explicitly
calculate probabilities using Bayes’ formula. Instead, they believe
that the nervous system acts
as if
it is making those calculations, doing the statistics implicitly.
⁴¹

How does this conclusion relate to the jungle principle? It indicates that neural networks with known properties can cope with sparse data. No hidden rule system is needed to explain how, with relatively little input, the nervous system can act as if it is endowed with a privileged way of learning in special domains. Trial-and-error learning can occur, but it can occur more efficiently than might be imagined were it not for the capacity to exploit Bayesian inference.

On the list of possible problems with trial-and-error learning, the final one is that a lot of learning seems to occur without any trial and error at all. Sometimes observation alone leads to learning.

The most influential demonstration of observation-based learning came from Albert Bandura at Stanford University. In Bandura’s experiment, children watched a video of another child striking a Bobo doll—a big rubber clown, filled with air and weighted on its bottom to keep it from toppling over when it was bopped. Children watched a child on a video playing with the Bobo doll. In one condition, the observed kid hit the doll a lot. Later, those who observed that behavior did likewise. By contrast, child observers who saw the model
play
with the Bobo doll were less likely to hit it. Instead, they played with it, too.
⁴²

Bandura was interested in observation because the prevailing view of learning prior to his study was the one espoused by B. F. Skinner—that learners can learn only by engaging in behaviors that lead to success or failure. Merely imagining behaviors or watching others perform the behaviors shouldn’t work, Skinner’s theory implied. Bandura showed otherwise. Via his Bobo-doll experiment, he showed that modeling is an effective way to learn.

Learning through observation doesn’t occur only by watching Bobos get bopped, however. It can also be observed in other contexts. Students of musical instruments watch and listen to their teachers play their trumpets, trombones, or tambourines to get a sense of how they themselves should play. Dancers watch choreographers demonstrate the steps and gestures the dancers should perform. My daughter, Nora (Sarah’s non-identical twin), learned to walk by watching Sarah stand and fall, over and over again. When Nora finally became convinced that walking was possible, she stood up and did so. So while her sister used trial and error to learn to locomote, Nora used modeling. Trial-and-error learning isn’t necessary, therefore, at least for walking in children. Modeling is another option.
⁴³

Is modeling a blight on Darwinian learning? If modeling works well, should Darwinian learning be dumped? Not at all! Learning how to act by watching others is an outcome you would expect if you thought the brain mechanisms involved in acting cooperate with the brain mechanisms involved in perceiving. There is abundant evidence that such cooperation exists. At the same time, equally strong evidence exists for
competition
between mechanisms for acting and for perceiving. The competition, like the cooperation, is so tight that the linkages bespeak nurturing niches for perception-action relations.

Take the case of blink suppression. When you blink, which you do every 2 seconds or so, your eyelid covers your cornea for about.2 seconds. This means that for much of your waking life, your eyes are covered and you literally see nothing. Yet you are completely unaware of this momentary darkening. The reason is that when your brain issues a command to close your eyes, your blink command center also sends signals to your visual center to disregard the darkness.

How can you be sure this claim is correct? You can rely on one of the most imaginative studies in experimental psychology.
⁴⁴
The researchers who carried out the experiment asked volunteers to allow an optic fiber to be placed in their mouths. The end of the optic fiber lightly touched the roof of the volunteers’ mouths. The researchers delivered flashes of light through the optic fibers. In case you didn’t know—and why would you?—it turns out that you can see red when light is directed to the roof of your mouth even if your eyes are closed. You can confirm this with a flashlight. Light manages to make its way to your retina through the blood-infused passages within your head.

In the study of blinking, light flashes were briefly presented through the roof of participants’ mouths at random times. Meanwhile, the blinks of the participants were recorded. This made it possible to determine after the fact when the flashes were delivered relative to the participants’ blinks. The experimenters found that sensitivity to light was dramatically reduced during the blinks. For a light to be detected around the time of a blink, it had to be brighter than at other times. This outcome suggests that the brain centers for blinking inhibit the brain centers for seeing. The inhibition allows the brain to distinguish between darkening of the world due to blinking and darkening of the world due to external dimming.

Other examples of action-related perceptual suppression have also been found. Another is saccadic suppression. Just as you’re less able to detect flashes
when you blink, you’re less able to detect flashes or locate pinpoints of light when you shoot your eyes from one place to another. Such eye jumps are known as saccades. The perceptual attenuation that accompanies these ocular hops is called
saccadic suppression
.

You can demonstrate saccadic suppression by looking into a mirror and trying to see your own eyes move. You will fail. You can’t see your own eye movements. Yet a friend watching you carry out the saccades can easily see your eyes dart from place to place. Therefore, your eyes don’t move too quickly to be seen while moving. Instead, it’s that you, the agent of your own saccades, can’t see the saccades you generate. If you’re unsure whether you can see eyes moving quickly, you and your friend can change places. If you watch your friend attempt to see his or her own eyes moving in the mirror, you’ll be able to see his or her eyes flitting about, though he or she won’t be able to see those same saccades. So you can see rapid eye movements. You just can’t see your own eyes move when you perform saccades.

What accounts for this curious outcome? The answer is that the swoosh of the visual scene across your retina is suppressed when you move your eyes. “Pay no attention to that swoosh,” your eye-movement command center tells your visual center, much as the Wizard of Oz tells Dorothy and her accomplices not to pay attention to the man behind the curtain. The competition implicit in this communication has a cooperative effect on a larger visual scale. By not seeing visual swooshes as you perform saccades, you experience the world as a coherent whole, not a jittery jumble.
⁴⁵

Another example of inner competition related to perception and action concerns tickling. No matter how much you might like to be tickled, you can’t tickle yourself. The reason is that when you try to tickle yourself, your brain centers that issue tickle commands tell other brain centers sensing touch to disregard the sensations coming along for the ride. The sense of self-touch isn’t completely eliminated, of course, but the unexpectedness of the stimulation diminishes, so the giddiness you’d otherwise enjoy is diminished.
⁴⁶

Acting doesn’t only
suppress
perception; it can also facilitate it. This possibility was noticed in the nineteenth century in Germany by Hermann Lötze and in America by William James.
⁴⁷
James wrote about imagining himself getting out of bed and then carrying out the just-imagined action. In much the same vein, he suggested that the mental image of his own writing triggered that
writing, that the mental image of his own speaking triggered that speaking, and so on. The notion that the idea of what you want to achieve triggers the action for achieving it came to be known as the
ideomotor
theory of action. It’s a theory that has attracted “Lötze” interest lately.
⁴⁸