It's a Jungle in There: How Competition and Cooperation in the Brain Shape the Mind (15 page)

There are other tasks, however, where it’s clearer that competition plays a role in shaping RTs. One is the task of pressing a left key if a left visual stimulus appears or pressing a right key if a right visual stimulus appears. Choice RTs are short in this circumstance. By contrast, if the mapping is reversed, so the left key must be pressed when a right stimulus appears, or the right key must be pressed when a left stimulus appears, the choice RTs are longer.

The ability to respond more quickly to same-side stimuli reflects a principle called
stimulus-response compatibility
. According to this principle, responses are quick and accurate if the stimuli calling for them arrive via “natural” or “compatible” stimulus-response mappings, but are less so stimulus-response mappings are “less natural” or “less compatible.”
¹⁵

What do the terms
natural
and
compatible
mean? Natural or compatible stimulus-response mappings are ones that have strong connections. Less natural or less compatible stimulus-response mappings are ones that have weaker connections. Accordingly, when responses that enjoy strong stimulus-response (S-R) associations must be made, those responses whisk through by virtue of their strong S-R bonds. Conversely, when responses that do not enjoy strong S-R associations are the ones that must be made, the
other
responses must somehow be made and the powerful S-R bonds must be inhibited. The left response “wants to go” when the left stimulus appears, so if the right response is necessary, the left response must be held in check. Likewise, if the right response “wants to go” when the left response is necessary, the right response (the response on the right) must be restrained.
¹⁶

So great is the need for inhibition of prepotent responses that the inhibition may be necessary even when the relevant stimulus-response mappings are
more indirect. Suppose you’re instructed to press a left key if you see the letter
X
or press a right key if you see the letter
Y
. Where the
X
or
Y
appears is irrelevant to the task description. Nevertheless, if the
X
is on the left, the left-key response is quick, but if the
X
is on the right, the left-key response is slow, and vice versa for the right-key response to the
Y
. This phenomenon is known as the Simon effect, in honor of the psychologist who discovered it.
¹⁷

What mechanism underlies the Simon effect? Why should the side of the stimulus affect choice RT when only the identity of the stimulus officially matters? Inner competition among response tendencies provides the answer once again. A left-side stimulus activates a left-side response independent of what the stimulus happens to be. Similarly, a right-side stimulus activates a right-side response independent of what stimulus happens to appear there.
¹⁸
A response is potentiated by the stimulus appearing on the responses’ associated side. For the other response to occur, extra time is needed to resolve the conflict.
¹⁹

Another RT task that reflects inner competition is a third one introduced by Donders on top of his simple RT and choice RT tasks. This third task lies somewhere between the other two.

Recall that in Donders’ simple RT task there is just one possible response, whereas in Donders’ choice RT task there is more than one possible response. The third, intermediate task that Donders also developed is the go/no-go task. Here participants are supposed to respond when one signal appears, but they are
not
supposed to respond when another signal appears. This is a familiar enough requirement. If you’re driving and the traffic light is green, you’re supposed to go. If the traffic light is red, you’re
not
supposed to go. The red light means, literally, STOP!

Stopping is something you need to do even in simple RT situations. It’s important not to jump the gun before the “go” signal comes on. Yet that demand is often violated. Swimmers poised to lunge into pools for their racing laps or sprinters leaning on their starting blocks sometimes start prematurely, reflecting a failure to fight the urge to lunge forward when the starting shot is fired.

Going only when you’re supposed to takes self-control or, more specifically, inhibition. When the need to inhibit is small, RTs can be zero or even negative. RTs are zero when responses coincide perfectly with the stimuli that are supposed to elicit them. RTs are negative when responses come too early, leading rather than following the stimuli.

In Donders’ go/no-go task, the need to refrain from responding is an explicit response alternative, not an implicit one. A useful feature of the go/no-go task is that within it, an experimenter can vary the probability that a response is called for. If a response has a high probability, it tends to have short RTs, but if a response has a low probability, it tends to have long RTs.

A natural way to interpret this finding is to say that a rarely-required response is greatly inhibited, whereas an often-required response is less greatly inhibited. Varying the degree of inhibition of a response is a way to match its likelihood. Go/no-go RTs depend on the probability of a response being called for or not.

Is this probability effect actually due to graded inhibition? An alternative view is that on each trial the participant simply predicts that the response will be required or not. If the prediction is correct, the RT can be quick and the accuracy high. If the prediction is incorrect, the RT will be longer and the accuracy lower.

Invoking an all-or-none prediction model has a problem, however. It suggests that RTs for a given response should either be very short (when the response that is predicted is actually required) or very long (when a response that’s not predicted is required). But RT distributions don’t usually have two clear humps.
²⁰

The other problem with the all-or-none prediction model is that it begs the question of what prediction is. Predictions aren’t always made; sometimes one is in a don’t-know state of mind. When one
is
in a predictive state of mind, it’s very difficult, in my opinion, to understand what the neural representation of that state of mind could be. How exactly can the brain represent what might happen with some probability? I don’t know, and I don’t think anyone else does either.

An alternative, simpler, explanation is that activations of S-R alternatives grow with their frequency, and the inhibitory effects of those S-R alternatives on other S-R alternatives also grow with their frequency. Activation and inhibition, varying in degree, tell the whole story.

Inhibiting responses is clearly required when you get ready to perform a response and then have to stop it. This situation has been studied in RT tasks where participants get a ready signal followed either by a go signal or by a don’t-go signal. If there’s a good chance a don’t-go signal will come shortly after a ready signal appears, participants become adept at stopping the
response. In fact, if the probability of a don’t-go signal is
very
high, participants don’t even bother to prepare the response, as shown by the relative lack of electrical activity in the muscles and in the brain compared to cases where the probability of the don’t-go signal is more moderate.

For less probable don’t-go signals, the dynamics are more interesting. If the don’t-go signal comes on soon after the go signal, the participant can withhold the response. However, if the don’t-go signal comes on late after the go signal but before the response has been made, the don’t-go signal has no impact. The response occurs anyway, suggesting that the internal signals coalescing in the brain to permit a response reach a point of no return. Like a competitive swimmer who has leapt from the launch platform before the starting gun goes off, as soon as he or she is airborne there’s no turning back.
²¹

How long is the delay between a ready signal and a don’t-go signal such that people can’t suppress their responses? A particular time value is less important than the fact that the time is within the range of normal RTs. This is relevant to this discussion because if you think the tactics people use in the stopping procedure are very different from the tactics they use in normal RT tasks, you might expect the temporal point-of-no-return to be outside the normal RT range. The fact that it’s within the normal range is consistent with the hypothesis that the inner workings of the nervous system are essentially the same when people are on the lookout for stimuli telling them which response to make or which response
not
to make. Inhibiting responses, by this line of argument, isn’t something that occurs only in stop-signal studies. Rather, it happens all the time.

Just as it is instructive to compare RTs for different kinds of tasks—and we will consider still more of them momentarily—it is also instructive to consider the lengths of RTs themselves. A few paragraphs back, I mentioned that RTs can be zero or even negative when responses coincide with or lead their associated stimuli. RTs can also be positive when people react to stimuli after the stimuli are presented. Little attention has been paid to understanding why positive RTs have the values they do. Why are simple RTs about a quarter of a second, and why are choice RTs about a half to a full second? These questions are interesting because RTs are longer than might be expected. The reason, I’ll argue, is that it’s a jungle in there.
²²

The speed of nerve conduction is about 32 meters per second, or about 72 miles per hour.
²³
Given this rate, you can ask how far nerve signals would
travel in .25 seconds (a respectable simple RT time) or in .5 seconds (a respectable choice RT). Because distance equals rate times time, the neural travel distance for a simple RT would be .25 seconds times 32 meters per second, or 8 meters. The neural travel distance for a choice RT would be .50 seconds times 32 meters per second, or 16 m.

How do 8 meters and 32 meters compare to the length of the human body? An average-height American man is 1.73 meters (5 feet, 8 inches) tall, so 8 meters is 4.62 times the length of his body, and 32 meters is 18.49 times the length of his body. If you thought RTs reflect the distance traveled by nerve signals, you’d have to say that in simple RT tasks nerve signals cover nearly 5 times the length of the body, and in choice RT tasks nerve signals cover nearly 19 times the length of the body.

What’s the use of these numbers? If you suppose a typical neuron, including its dendrites and axons, has a length of 1 millimeter, then 8,000 neurons are needed to cover the 8 meters for a simple RT, and 32,000 neurons are needed to cover the 32 meters for a choice RT. You’ve then got an awful of neurons lined up, doing what exactly? That’s the puzzle. Too many neurons seem to be involved.

Why do there seem to be too many neurons? Think of a neuron as a switch. The nervous system might prepare for a simple RT task by keeping one switch open until a stimulus is detected, at which time the switch could close and the response could be triggered. Alternatively, the nervous system might prepare for a
choice
RT task by keeping
two
switches open, one per S-R choice. The added time for a choice RT compared to a simple RT could then be ascribed to the need for two switches rather than one. The problem with this scenario is that the time needed for an individual neuron to switch on is so short that the time is virtually unmeasurable.

A different way to address the question is to say that most of the time contributing to RTs is possibly non-neural. It could be, for example, that most of the RT is spent getting the muscles to physically generate the response. You could allow that more time is spent getting the muscles to move in choice RTs than in simple RTs, and you could assign the difference between choice RTs and simple RTs to muscle activation rather than to neural signaling. For example, you could say that in choice RTs, participants often need to relax some of their muscles that they initially tensed in order to generate the required response. All or most of the RT might then be taken up with relaxing and activating muscles rather than with sending signals down the neural transom. But it turns out that muscle activation times are very short relative to total RTs. Neurophysiologists have measured the times between neural signals in
the motor cortex and subsequent button presses. The times are about 50–70 ms, and that’s true regardless of whether the data come from simple RTs or choice RTs.
²⁴
Thus, muscle activation can’t account for the lengths of RTs.

Another possibility is that there’s untold complexity in stimulus processing. To reach the point of finally throwing a neural switch one way or the other based on the detection of a stimulus in a simple RT task, it might be that a very long line of neurons is needed just to reach that point.

This possibility, like the possibility that muscle activation accounts for RT length, can be set aside. When neurophysiologists record from sensory regions of the brain, they pick up reliable differences in the neurons that fire depending on which stimulus is presented, and they pick up those differences in extraordinarily short times after the stimuli come on. The times are so short—on the order of a few milliseconds—that it’s hard to believe the largest portion of RTs reflects sensory processing. Only a small portion of RTs actually reflect the differential registration of one stimulus among the alternative possible stimuli, at least if the stimuli are easily discriminated, which they have been in the tasks I’ve been describing—for example, when there’s a light several centimeters to the left or right of a computer screen’s midline.

So what’s the best hypothesis about the length of the RT vis-à-vis the number of neurons underlying them? On the way to answering this question, let me say that you don’t need to equate RTs with strings of neurons. It’s useful to do so for the sake of discussion, but the main point of the argument I’ve been making has actually been to string you along, so to speak, to show that it’s not especially helpful to equate RTs with neural strings. Linking RTs to some number of neurons working in a linear chain is not a profitable line of argument, not least because if any neuron in the chain malfunctions, the chain breaks. Nature seldom puts all her eggs in one basket; instead, she favors redundancy. If you allow for more than one linear string of neurons between a stimulus and a response—a possible way of avoiding putting all the eggs in one basket—it’s not clear what each neuron in the chain actually adds. In fact, adding more neurons to a single chain might just add statistical noise. As in the game “whispering down the lane,” the more communicators there are, the more degraded the signal.