YOU ARE cruising along when you spot something in the road ahead. Will you
swerve around it, hit the brakes, or drive over it? You鈥檇 better make up your
mind鈥攁nd quickly.
We may talk about snap decisions, but when we make judgments like this our
reaction times are rarely up to speed. A signal should pass from eye to brain to
muscles in just 60 milliseconds. But when something appears within view it can
take three times longer to direct our gaze to it. 鈥淭he delay between a visual
stimulus and a reaction is the time it takes the brain to accumulate enough
information to decide on making a movement,鈥� says Roger Carpenter, a
physiologist at the University of Cambridge. 鈥淩eaction times therefore reflect
decision processes rather than mundane mechanisms of signal propagation.鈥�
Carpenter has been studying the eye鈥檚 sluggish reactions for more than two
decades. His simple mathematical analyses of reaction times have helped him
develop a model to explain what is going on during those precious milliseconds
of procrastination as we decide where to look or what to do next. 鈥淚 believe
that my model could describe the mechanism by which all decisions, both
conscious and subconscious, are made,鈥� he says.
Advertisement
So, you spy a movement out of the corner of your eye. You can鈥檛 see it
properly yet, but even so your brain has generated a series of hypotheses as to
what it might be. Some will be plausible, some less so鈥擨s it a bird? Is it
a plane? Or is it Superman? Are you going to look more closely?
Carpenter says that the brain uses a mixture of probability and chance to
decide. Probability reflects our rational side, favouring what is 鈥渕ost likely鈥�.
But chance pushes us towards more random behaviour ensuring that we are never
completely predictable. Carpenter鈥檚 model marries these two extremes in the
excitement of a race between the possible choices. But this is more a race to
climb a hill than a 100-metre sprint. The first to the top is the winner and
will define both the decision and our response.
The contest takes place between the activities of neurons in the brain.
Neurons signal to each other by firing off short electrical pulses. In terms of
the race, a low firing rate is like being at the bottom of the hill. The faster
a neuron fires, the higher it climbs. The summit is a threshold level of
activity sufficient to trigger a response.
But how high is that summit? According to Carpenter, it depends on the
urgency of the situation, and how critical a decision it is. 鈥淚f more accurate
information is needed to make a decision, the finishing post, or threshold, is
set higher. That makes the race longer, and consequently increases our reaction
times even more.鈥�
When the threshold is set, our competing hypotheses鈥攂ird, plane and
Superman鈥攖ake up their starting positions. But this is an unfair
competition, because the starting blocks are set by probability, fixing the race
in favour of the most sensible outcome. So 鈥渂ird鈥� starts the race already well
on the way to the peak, with the neurons already firing quite rapidly.
鈥淪uperman鈥� (extremely unlikely) has a long way to go, while plane would be
somewhere in-between.
How the brain calculates the initial probabilities, and turns them into
physical firing rates is anyone鈥檚 guess, but experience and learning seem to
play a part. We鈥檝e all seen plenty of birds and planes, but not many flying
superheroes, so our expectation of seeing Superman is extremely low. When the
sensory information is inadequate, just a movement out of the corner of the eye,
say, expectation can really colour our judgment.
High hopes
As Carpenter points out, 鈥淚t鈥檚 easy to jump to conclusions, like in
proofreading when you miss glaring errors simply because you didn鈥檛 expect to
see them.鈥� Learning is equally important. If you keep looking round for Superman
every time something moves, your brain quickly realises that your hopes are set
too high, and adjusts your expectations.
Paul Glimcher, a neurophysiologist from New York University, argues that
other abstract parameters must also influence decisions. 鈥淥ur decisions are
affected by how much we value something or what we expect to gain from
particular decisions. For instance, in a foraging animal the estimated value of
a peach varies depending on how hungry the animal is.鈥�
Glimcher has shown recently that some neurons encode 鈥渧alue鈥� (Nature,
vol 400, p 233). He studied the behaviour of a set of neurons in an area of
the monkey brain that controls eye movements, called the parietal cortex.
Glimcher lit up two identical LED displays as targets either side of where the
monkey was looking. If the monkey moved its gaze from the central reference
point to one of the targets, it was rewarded with a mouthful of juice. Glimcher
could see the decision taking place in the firing rate of groups of neurons
linked to movement left or right.
But he also found that when the target on the right yielded a bigger reward
than the one on the left, the initial firing rate of the 鈥渞ight鈥� neurons was
higher. This suggests that the 鈥渧alue鈥� may affect how the brain calculates that
all-important initial likelihood. 鈥淭his has a nice common sense appeal because
what we decide to do doesn鈥檛 just depend on the information we鈥檙e given but on
what we expect to gain at the end,鈥� says Glimcher.
So all the starting positions have been assessed, and the contestants take
their marks鈥攍et the race begin! The neuronal activities begin to rise at a
steady rate towards the threshold
(see Diagram). Carpenter thinks that
this rise is probably caused by positive feedback loops in the neuronal
circuitry, where the output of a neuron feeds back to its own input. If the
output increases, so does the input鈥攚hich in turn pushes the output up
another notch.
Jeff Schall from Vanderbilt University in Nashville, Tennessee, has spied the
hill-racing neurons in action (Science, vol 274, p 427). When a visual
target is displayed, he finds that the firing rates of the same neurons that
Glimcher studied increase in a linear way until a threshold is reached. This
triggers an eye movement, then the firing rates fall again. The contestants may
all race at a constant speed, but they don鈥檛 all race at the same
pace鈥攐therwise the hypothesis that starts closest to the threshold will
always win. This is where chance finally gets to play its part鈥攖he rate at
which the neuronal firing rises to the threshold varies randomly within a
certain range.
鈥淭his gives an incredible variation to reaction times,鈥� explains Carpenter.
鈥淚n two consecutive and identical trials, the reaction time could vary by as
much as 200 milliseconds. This isn鈥檛 just natural statistical variation or
background noise in our experiments. There鈥檚 a specific mechanism that
deliberately makes you make irrational decisions occasionally.鈥� So the least
likely hypothesis may occasionally overtake the others and reach the summit
first: you really may decide it is Superman.
There are good biological reasons for this random and irrational behaviour, says Carpenter
(快猫短视频, 22 August 1998, p 32). In nature,
unpredictability can mean survival. A shrimp can escape a hungry fish by
randomly changing its escape patterns, so the fish can鈥檛 anticipate its
movements. 鈥淚f we responded in the same way to the most probable stimulus, life
would be extremely boring鈥�, says Carpenter. 鈥淪o at the cost of making the
occasional stupid decision, our brain allows a certain degree of random
behaviour. I believe this is central to what we call creativity.鈥�
It may even be the root of what we call free will. If we always responded in
exactly the same way to the same stimuli, then we would be nothing more than
robots. This random rise, central to Carpenter鈥檚 model, does at least provide a
semblance of both spontaneity and choice. But it鈥檚 a long stretch of the
imagination to get from simple subconscious decisions about eye movement to free
will. Just how far can Carpenter鈥檚 model be applied to higher, more conscious,
levels of decision making?
Which way?
Mike Shadlen, from the University of Washington in Seattle, is another
researcher who has measured the activity of neurons in the parietal cortex. But
he stresses the differences in his experiments. 鈥淲e show monkeys a screen of
moving dots. The dots move about randomly, but there is a net movement in one
direction or another. The monkey has to decide which way the dots are going and
then look to the left or right to indicate her judgment.鈥�
Here the eye movement is more than homing in on a target. It communicates a
judgment, a decision reached after acute observation and reflection. Yet Shadlen
observes the same neuronal behaviour鈥攁 rise of the firing rate to a
threshold that triggers the eye movement. You can imagine that the neurons
controlling left movements race against those controlling rightward glances.
Shadlen says that this adds a completely new element to the idea of decision
making. He says that information from the senses directly influences the
structures in the brain related to planning the motor response. In other words,
your brain is less interested in whether it鈥檚 a bird, or a plane or even
Superman than in whether to duck, do nothing, or grab a camera. Carpenter鈥檚
鈥渞ace鈥� takes place in the cells that control responses, not in some special
region of the brain assigned to decision making. 鈥淭hat鈥檚 a radical departure
from a commonly held view that the brain has some internal control centre that鈥檚
very smart and knows the answer,鈥� says Shadlen.
Carpenter argues that this also makes sense because the number of possible
responses is quite small. It is much easier to hook up a limited number of
response neurons to all the necessary sensory inputs. 鈥淪o, for example, if you
tell somebody to press a button when they see a black cross and pull a lever for
white circle, then you don鈥檛 have to have neurons encoding black crosses and
white circles linked to all manner of response neurons. Instead, you simply link
the visual neurons to ones for pressing a button and pulling a lever and let
them race each other.鈥�
Historical accuracy
Can this explain more complex behaviour than where to look or whether to
press a button? Carpenter is confident that all decisions probably follow his
model. He says history, at least, is on his side. 鈥淚鈥檝e gone back and looked at
all the historical data on reaction times going back a hundred years or more.
People have been measuring reaction times quite accurately for a long period.
The extraordinary thing is that the data fit the model extremely well.鈥�
Carpenter has carried out the same mathematical analyses on historical
experiments as he did for his own work. Whether the reactions are for eye, arm
or leg movements, or more complex judgments, he finds that the reaction times
all have the same delays, errors and variations鈥攃haracteristics that can
be explained in terms of his model. 鈥淚t looks like this could underlie all
decision making,鈥� he concludes.
So, you鈥檝e successfully avoided the dog sitting in the road. Now you need to
overtake the bus. Is there room to do it? Your mind is racing again鈥�