快猫短视频

DR STRANGELOVE, the mad scientist played by Peter Sellers in the 1964 film,
had a problem with his right arm鈥攊t seemed to have a mind of its own.
Sometimes it would give Nazi salutes, and occasionally it tried to strangle him.
Newborn babies face a similar challenge: hand them a rattle and they may
accidentally clout themselves on the head. We adults reckon that they flail
about like upturned beetles because they cannot yet control their limbs. But how
in the world do any of us manage to do that?

Keeping track of all our muscles and joints as we move is no mean feat, and
no one really knows how we coordinate all this information to perform quick and
smooth movements. But neuroscientists are close to an answer. The secret, they
say, lies in learning to make accurate predictions about our bodily movements.
And the brain鈥檚 chief oracle lurks somewhere in the microcircuitry of the
cerebellum, at the back of the skull. This mysterious brain region occupies a
tenth of the skull space, yet is packed with half the central nervous system鈥檚
neurons. Adults with a damaged cerebellum, through accident, stroke or tumour,
act like a perpetual drunk. They slur their speech, stagger and struggle to do
something as simple as fitting a key in a lock.

Platoons of muscles are recruited to execute even a simple action, and
researchers are now beginning to realise that coordination breaks down in
patients with damaged cerebellums because suddenly they can no longer predict
what their body should do. The cerebellum is emerging as our virtual reality
centre, creating simulations of our movements to help control the real thing. It
seems to specialise in generating dynamic, ever-changing internal models of
every part of our body鈥攙irtual limbs, and more.

It is this remarkable ability that enables us to pick up a milk carton with
just the right amount of muscular force. 鈥淭he cerebellum generates a prediction
of what is going to happen given my motor command,鈥� says Daniel Wolpert, a
specialist in motor control at the Institute of Neurology in London. It has to
predict the consequences of activating certain muscles, and which would be best
to achieve a particular end. When reality does not fit its prediction鈥攊f
you thought the carton was full and it turns out to be nearly empty鈥攖he
milk goes flying.

Coordinating even a simple movement like this is a real challenge because
your brain has to send what it judges to be appropriate instructions to your arm
muscles long before it gets any sensory feedback from the hand doing the
lifting. 鈥淭he sensory information the brain receives is always a little bit out
of date,鈥� says neurophysiologist Chris Miall of the University of Oxford. 鈥淎t
any one time, the brain doesn鈥檛 know where the hand is.鈥� If we relied
exclusively on sensory feedback to guide our actions, we would have to move much
more slowly.

Five years ago, Miall and Wolpert noticed a similarity between the challenge
the brain faces as we cycle, ski or drive and one faced by engineers designing
steel foundries. The engineers鈥� problem was how to regulate the thickness of a
plate of molten steel being squeezed by giant rollers, when there wasn鈥檛 time to
measure the thickness and adjust the roller pressure accordingly. What they
needed was a form of controller that could cope with long and unavoidable
feedback delays.

One solution is the Smith Predictor, named after the American engineer Otto
Smith who first suggested it in the late 1950s. The key is to predict how the
object you want to control will behave when treated in a particular way. In
other words, you build up an internal model that follows the same control
commands as the system itself. In the steel plant, the commands regulate the
pressure of the rollers and the model holds information about temperature,
speed, size and so on, and predicts what thickness it expects the sheet to be,
according to past experience. The model can be updated and improved when the
feedback eventually arrives to tell it what actually happened. Miall and Wolpert
suggested that we might have a model of our own body inside the cerebellum,
working just like the Smith Predictor to help control our movements.

The exact nature of the representations in the cerebellum is hotly debated,
but one possibility is that there are two linked models
(see Diagram).

The
first鈥攖he predictor or 鈥渇orward model鈥濃�攊s a virtual-reality
simulator that uses its knowledge of the present position of the body, the state
of the motor system and a copy of any active motor commands to predict the
body鈥檚 future state. The forward model is supplemented by a controller or
鈥渋nverse model鈥�, which is a neural representation of the motor commands that are
required to achieve a particular end. The researchers believe that a huge
variety of physical actions are possible because we gradually learn a whole
series of these linked models, or 鈥渕odules鈥�, to cope with diverse physical
challenges鈥攅verything from walking on ice and swimming to using a pen and
eating with chopsticks.

Whatever we鈥檙e doing, the brain is bombarded with sensory feedback from our
limbs, which is vital for keeping the movement control systems in check. But
there鈥檚 an added complication. The brain needs to be able to recognise which
sensations are caused by our own movement, and which might be caused by
something in the environment. 鈥淲e can ignore our own touch, but we had better
pay attention to someone else鈥檚 touch,鈥� explains Sarah Blakemore from the
Institute of Neurology. After all, it could be a venomous spider crawling up our
leg.

The cerebellum鈥檚 powers of prediction seem vital not only for guiding
movement, but also for making sure that self-generated sensations are perceived
as less intense than the same externally produced stimuli. No wonder, then, that
the cerebellum is remarkably large in electric fish, echolocating bats and
whales鈥攁nimals that analyse the echoes of electrical discharges or
auditory signals that they themselves have generated. To gain information about
the world, they must be able to predict with extraordinary precision which
echoes result from their own movements and which reflect important, unexpected
objects in the environment.

Our ability to predict the consequences of our actions explains why we can鈥檛
tickle ourselves, argues Blakemore. In her tests, volunteers were stroked on the
palms of their hands by a robotic arm clutching a piece of foam rubber. The
sensation failed to prove ticklish when the subject directly controlled the
robotic arm, but did feel funny if the robot did the stroking. But when a time
lag of just 200 milliseconds was introduced into the self-stimulating task, this
sensation starts to feel ticklish.

Brain scans of the volunteers suggested an explanation. When the robot
generated the stroking motion, 鈥減leasure鈥� centres in the somatosensory
cortex鈥攖he part of the brain that registers skin sensations鈥攚ere
active, while the cerebellum was quiet. When the sensation was self-generated,
these regions were largely silent, but this time the cerebellum was active. 鈥淭he
cerebellum seems to send a signal that is used to cancel the sensory response to
self-generated stimulation,鈥� Blakemore says. 鈥淭hat area is telling the rest of
the brain that it鈥檚 you who is responsible for the sensation on your hand.鈥�

Recent work by Blakemore, Wolpert and Chris Frith at the Institute of
Neurology suggests that in schizophrenia something may be going awry with the
brain鈥檚 powers of prediction. Typical symptoms of this mental illness include
hearing voices and feelings of being controlled by outside forces. Both
delusions, say the researchers, could be the result of patients somehow losing
the ability to predict the sensory consequences of their own actions, so that
their own movements are perceived as external events. The voices a patient hears
may stem from their own words silently mouthed, but unrecognised as
self-generated.

Remarkably ticklish

Blakemore and her colleagues speculated that if schizophrenics have
difficulty predicting the outcome of their actions, they might be able to tickle
themselves. Remarkably, preliminary results support that conclusion.
Schizophrenic patients reported identical ticklish sensations regardless of
whether or not they were controlling the robotic arm.

How any of this is achieved in terms of neuronal wiring is still far from
certain, even though we know more about the microscopic structure of the
cerebellum than any other part of the brain. But clues come from the
cerebellum鈥檚 strong interconnections with other parts of the brain. One vital
input, says Richard Apps, a neurophysiologist at the University of Bristol, is
from part of the brainstem at the top of the spinal cord called the inferior
olive, a relay station for a collection of pathways that convey sensory
information from the limbs. These pathways make contact with individual neurons
called Purkinje cells in the cerebellum, via connections known as climbing
fibres. Each Purkinje cell looks rather like a flattened oak, entwined in the
ivy of a single climbing fibre. Researchers have long known that damage to the
inferior olive results in movement deficits that closely resemble the
consequences of damage to the cerebellum itself, so these communication channels
must convey vital information. 鈥淵et we still don鈥檛 know precisely what these
pathways are signalling to the cerebellum,鈥� says Apps.

But he has recently obtained striking evidence that these pathways are not
always open for communication, and this provides clues to their function. He has
shown that during active movement, the flow of sensory information into the
cerebellum via the climbing fibres is sometimes shut off, or 鈥済ated out鈥�. Could
this gating explain why sensations such as our own touch aren鈥檛 ticklish?

The timing of this gating is certainly consistent with a role in detecting
mismatches between the predicted consequences of movement and the reality. When
the foot touches the ground during a step, Apps found that sensory touch
information coming from the leg via the climbing fibres is gated out. Yet the
same sensory channels are open, and even enhanced, when the leg swings forward,
when something untoward might trip you up or alter your step. In other words,
self-generated sensory information is suppressed while unexpected sensory
information is heightened. 鈥淥ne possibility,鈥� says Apps, 鈥渋s that the gating may
reflect the output of an internal model which anticipates and cancels out the
sensory effects of movement.鈥� Gating could serve to prevent self-generated
鈥渋rrelevant鈥� sensory inputs from being relayed to the cerebellum, while signals
that might need to be acted upon are selected for transmission.

Progress is also being made in understanding the connections that flow
downwards from the cerebellum to influence the motor circuits, and upwards to
the many different areas of the cerebral cortex which help the cerebellum to
construct our movement models.

Disturbing symptoms linked to damage to one cortical region, the superior
parietal lobe, have led Wolpert to suggest that, normally, this region acts as a
store of the representation of the self and its position in the world generated
by the cerebellum. Last year, a patient seen by Wolpert and his colleagues was
found to have developed a cyst in her superior parietal lobe. One consequence is
that 鈥渟he perceives her right arm and leg to drift and then fade unless she is
able to see them,鈥� Wolpert explains. 鈥淔or example, while lying in bed, she might
realise that she has `lost鈥� her right arm. Only when she looks at her arm does
she know where it is. Similarly, she can be sitting on a bus and find that
another passenger has tripped over her right foot, which is situated, without
her knowledge, in the middle of the aisle.鈥� Wolpert interprets her plight as
further evidence that the brain stores a regularly updated virtual image of the
body that gives an estimate of its current state. 鈥淲hatever is storing the model
between updates is damaged in this patient,鈥� says Wolpert.

But even with a healthy brain, keeping our bodies under control can be a
struggle. Why, for instance, can鈥檛 we throw a dart the same way twice in a row?
Something stops us from repeating a movement perfectly. Wolpert suspects that we
do the best we can given unavoidable randomness or 鈥渘oise鈥� in the motor system.
Some people may be much better at darts than others because they have less noise
in their systems. 鈥淚f we could measure the randomness or noise properties in
motor firing units鈥攊n the muscles and their neural circuitry鈥攚e
might find marked differences in intrinsic noise levels,鈥� says Wolpert. We might
also be able to identify potential darts champions.

Although the noise levels are beyond our control, most aspects of hand-eye
proficiency are almost certainly largely learnt, as are all our motor skills.
鈥淭he cerebellum is not hard-wired when we are born, but has to learn through
experience to build up new models,鈥� Miall explains. The baby thrashes around,
sees what happens as its arm flies around, and begins to devise a set of
internal models.

He suspects that it鈥檚 easier to learn a complex new motor skill such as
riding a bicycle when we are young because there are lots of unallocated modules
in the brain. Early learning also gives enduring skills: as modules become
specialised they are protected from being overwritten, Wolpert says, especially
if they have a distinctive function. 鈥淎lmost nothing is like bicycling or
skiing, and these skills once learnt are protected from damage.鈥� Other skills,
such as squash and tennis, may share many modules and even interfere with one
another, whereas squash and bowling modules would not.

Our burgeoning understanding of human movement may have far-reaching
implications for everything from athletics training to primary school education.
And the growing realisation that active prediction is at the core of every
sensory experience we have might alter the way we think about perception. 鈥淵ou
could argue that there is no such thing as an independent sensory system,鈥� Miall
says, because what we perceive is always influenced by our own plans and
movements. It seems action is at the very heart of being alive.

Virtual limb: a controller and predictor in the cerebellum plan what moves we
should make given our current position and our goal. Feedback arrives later to
keep the models in check