Daniel Wolpert On Why You Have A Brain

Daniel Wolpert is absolutely certain about one thing.  “We have a brain for one reason and one reason only, and that’s to produce adaptable and complex movements,” stated Wolpert, Director of the Computational and Biological Learning Lab at the University of Cambridge.  “Movement is the only way you have of affecting the world around you.”  After that assertive opening to his 2011 TED Talk, he reported that, despite this important purpose, we have a long way to go in understanding of how exactly the brain controls our movements.

Daniel Wolpert

The evidence for this is in how well we’ve learned to mimic our movements using computers and robots.  For example, take the game of chess.  Since the late 1990s, computer software has been playing competitive matches and beating human master players by using programmed tactics and sheer computing power to analyze possible moves.  However, Wolpert points out that a five-year-old child can outperform the best robot in actually moving chess pieces around the board.

From a sports context, think of a baseball batter at the plate trying to hit a fastball.  It seems intuitive to watch the ball, time the start of the swing, position the bat at the right height to intercept the ball and send it deep.  So, why is hitting a baseball one of the most difficult tasks in sports?  Why can’t we perform more consistently?

The problem is noise.  Not noise as in the sense of sound but rather the variability of incoming sensory feedback, in other words, what your eyes and ears are telling you.  In baseball, the location and speed of the pitch are never exactly the same, so the brain needs a method to adapt to this uncertainty.  To do this, we need to make inferences or beliefs about the world.


The secret to this calculation, says Wolpert, is Bayesian decision theory, a gift of 18th century English mathematician and minister, Thomas Bayes.  In this framework, a belief is measured between 0, no confidence in the belief at all, and 1, complete trust in the belief.  Two sources of information are compared to find the probability of one result given another.  In the science of movement, these two sources are data, in the form of sensory input, and knowledge, in the form of prior memories learned from your experiences.

Thomas Bayes

So, our brain is constantly doing Bayesian calculations to compute the probability that the pitch that our eyes tell us is a fastball is actually a fastball based on our prior knowledge.  Every hitter knows when this calculation goes wrong when our prior knowledge tells our brain so convincingly that the next pitch will be a fastball, it overrules the real-time sensory input that this is actually a nasty curve ball.  The result is either a frozen set of muscles that get no instructions from a confused brain or a swing that is way too early.

Our actions and movements become a never-ending cycle of predictions.  Based on the visual stimuli of the approaching baseball, we send a command to our muscles to swing at the pitch at a certain time.  We receive instant feedback from our eyes, ears and hands about our success or failure in hitting the ball, then log that experience in our memory.

Wolpert calls this process our “neural simulator” which constantly and subconsciously makes predictions of how our movements will influence our surroundings. “The fundamental idea is you want to plan your movements so as to minimize the negative consequence of the noise,” he explained.

We can get a sense of what its like to break this action-feedback loop.  Imagine a pitcher aiming at the catcher’s mitt, releasing the ball but then never being able to see where the pitch ended up.  The brain would not be able to store that action as a success or failure and the Bayesian algorithm for future predictions would be incomplete.

Try this experiment with a friend.  Pick up a heavy object, like a large book, and hold it underneath with your left hand.  If you now use your right hand to lift the book off of your left hand, you’ll notice that your left hand stays steady.  However, if your friend lifts the book off of your hand, your brain will not be able to predict exactly when that will happen.  Your left hand will rise up just a little after the book is gone, until your brain realizes it no longer needs to compensate for the book’s weight.  When your own movement removed the book, your brain was able to cancel out that action and predict with certainty when to adjust your left hand’s support.

“As we go around, we learn about statistics of the world and lay that down,” said Wolpert.  “But we also learn about how noisy our own sensory apparatus is and then combine those in a real Bayesian way.”

Our movements, especially in sports, are very complex and the brain to body communication pathways are still being discovered.  We’ll rely on self-proclaimed “movement chauvinists” like Daniel Wolpert to continue to map those routes.  In the meantime, you can still brag about the pure genius of your five-year-old hitting a baseball.

Join Axon Sports on Twitter and Facebook

Stroke Patients Benefit From New Brain And Motor Skills Research

Bioengineers have taken a small step toward improving physical recovery in stroke patients by showing that a key feature of how limb motion is encoded in the nervous system plays a crucial role in how new motor skills are learned.

Published in a recent issue of Neuron, a Harvard-based study about the neural learning elements responsible for motor learning may help scientists design rehabilitation protocols in which motor adaptation occurs more readily, potentially allowing for a more rapid recovery.

Neuroscientists have long understood that the brain's primary motor cortex and the body's low-level peripheral stretch sensors encode information about the position and velocity of limb motion in a positively-correlated manner rather than as independent variables.

"While this correlation between the brain's encoding of the position and the velocity of motion is well-known, its potential importance and practical use has been unclear until now," says coauthor Maurice A. Smith, Assistant Professor of Bioengineering at the Harvard School of Engineering and Applied Sciences (SEAS) and the Center for Brain Science in the Faculty of Arts and Sciences.

Smith and colleagues showed that the correlated neural tuning to position and velocity is also present in the neural learning elements responsible for motor learning. Moreover, this correlated drive can explain key features of the motor adaptation process.

To study and record motor adaptation, the researchers had subjects grasp a robotic arm. The device was programmed to simulate novel physical dynamics as subjects made reaching motions. In addition, the team used a newly developed measurement technique called an "error-clamp" to tease apart the resulting data.

The method measures motor output during learning, allowing learning-related changes in motor output over the course of a movement to be dissociated from feedback adjustments that correct motor errors that happen simultaneously.

"Conceptually, this error-clamp is analogous to a voltage-clamp, commonly used in electrophysiology to measure how ions move through a neuron's membrane when it fires," explains lead author Gary C. Sing, a graduate student at SEAS. "The general idea is that devising an experimental method to clamp and control the key variable in an experiment can allow for greater insight into the underlying physiology."

Analysis of the data extracted by the error-clamp technique led to the creation of a computational model that identifies a set of vectors that characterize the principal components of motor adaptation in the state space of physical motion. While such analysis is commonplace in systems engineering -- for example, in evaluating how a bridge might react to high winds or earthquakes -- the method has only been recently applied to how motor output evolves.

"We observed that the initial stages of motor learning are often quick but non-specific, whereas later stages of learning are slower and more precise," says Sing. "Further, we saw that some physical patterns of movement are learned more quickly than others."

By understanding what types of motor adaptations are easier to learn, the researchers hope to design rehabilitation activities that will encourage patients to use an affected limb more.

"In stroke rehabilitation, patients who make a greater effort to use their impaired limbs can achieve better outcomes," says Smith. "However, there is often a vicious cycle, as a patient is far less likely to use an impaired limb if his or her other limb is fine. This pattern slows recovery and leads to greater impairment of the affected limb."

Smith and his colleagues are beginning studies with stroke patients to determine whether training them with such optimized patterns will, in fact, improve their rate of motor learning and speed up recovery.
More broadly, untangling the algorithms the brain uses for motor learning could help improve a wide range of neural and muscular rehabilitation programs. The researchers also anticipate that such findings could be one day be adapted for enhancing the brain/machine interfaces increasingly used for those with amputated limbs.

Sources:  Harvard University and "Primitives for Motor Adaptation Reflect Correlated Neural Tuning to Position and Velocity"

The Physiology Of Speed

Usain Bolt, the triple Olympic gold medal sprinter from Jamaica, predicted last week that he could break his own world record of 9.69 seconds in the 100 meter sprint with a time as low as 9.54 seconds.  (8/15 update: he came very close running a 9.58 at the World Championships in Berlin.)

He claimed his coach told him its possible, so he believes him. His coach, Glen Mills, may have just finished reading some new research coming out of Duke University that showed sprinters and swimmers who are taller, heavier but more slender are the ones breaking world records.

At first glance, it may not make sense that bigger athletes would be faster. However, Jordan Charles, a recent engineering grad at Duke, plotted all of the world record holders in the 100 meter sprint and the 100 meter swim since 1900 against their height, weight and a measurement he called "slenderness."

World record sprinters have gained an average of 6.4 inches in height since 1900, while champion swimmers have shot up 4.5 inches, compared to the mere mortal average height gain of 1.9 inches.
During the same time, about 7/10 of a second have been shaved off of the 100-meter sprint while over 14 seconds have come off the 100-meter swim record.

What's going on
Charles applied the "constructal theory" he learned from his mentor Adrian Bejan, a mechanical engineering professor at Duke, that describes how objects move through their environment.

"Anything that moves, or anything that flows, must evolve so that it flows more and more easily," Bejan said. "Nature wants to find a smoother path, to flow more easily, to find a path with less resistance," he said. "The animal design never gets there, but it tries to be the least imperfect that it can be."

Their research is reported in the current online edition of the Journal of Experimental Biology.

For locomotion, a human needs to overcome two forces, gravity and friction. First, an athlete would need to lift his foot off the ground or keep his body at the water line without sinking. Second, air resistance for the sprinter and water resistance for the swimmer will limit speed.

So, the first step is actually weight lifting, which a bigger, stronger athlete will excel at. The second step is to move through the space with the least friction, which emphasizes the new slenderness factor.

By comparing height with a calculated "width" of the athlete, slenderness is a measurement of mass spread out over a long frame. The athlete that can build on more muscle mass over a aerodynamic frame will have the advantage.

The numbers
In swimming, legendary Hawaiian champion Duke Kahanamoku set the world record in 1912 with a time of 61.6 seconds with a calculated slenderness of 7.88. Some 96 years later, Eamon Sullivan lowered the world mark to 47.05 seconds at a slenderness factor of 8.29.

As the athletes’ slenderness factor has risen over the years, the winning times have dropped.  In 1929, Eddie Tolan's world-record 100 meter sprint of 10.4 seconds was achieved with a slenderness factor of 7.61. When Usain Bolt ran 9.69 seconds in the 2008 Olympics, his slenderness was also 8.29 while also being the tallest champion in history at 6-feet 5-inches.

“The trends revealed by our analysis suggest that speed records will continue to be dominated by heavier and taller athletes,” said Charles. “We believe that this is due to the constructal rules of animal locomotion and not the contemporary increase in the average size of humans.”

So, how fast did the original Olympians run? Charles used an anthropology finding for Greek and Roman body mass and plugged it into his formula.

“In antiquity, body weights were roughly 70 percent of what they are today,” Charles said. “Using our theory, a 100-meter dash that is won in 13 seconds would have taken about 14 seconds back then.”
Bolt puts his prediction to the test next month at the track and field world championships in Berlin. One of his main competitors is Asafa Powell, the previous world record holder, who is shorter and has a slenderness factor of 7.85. My money is on the Lightning Bolt.

The Cognitive Benefits Of Being A Sports Fan

When was the last time you listened to a sporting event on the radio? If given a choice between watching the game on a big screen HD or turning on the AM radio, most of us would probably choose the visual sensation of television.

But, for a moment, think about the active attention you need in order to listen to a radio broadcast and interpret the play-by-play announcer's descriptions. As you hear the words, your "mind's eye" paints the picture of the action so you can imagine the scene and situations. Your knowledge of the game, either from playing it or watching it for years helps you understand the narrative, the terms and the game's "lingo".

Now, imagine that you are listening to a broadcast about a sport you know nothing about. Hearing Bob Uecker say, "With two out in the ninth, the bases are loaded and the Brewers' RBI leader has two strikes. The infield is in as the pitcher delivers. Its a hard grounder to third that he takes on the short hop and fires a bullet to first for the final out." If you have no baseball-specific knowledge, those sentences are meaningless.

However, for those of us that have grown up with baseball, that description makes perfect sense and our mind's eye helped us picture the scene. That last sentence about the "hard grounder" and the thrown "bullet" may have even triggered some unconscious physical movements by you as your brain interpreted those action phrases. That sensorimotor reaction is at the base of what is called "embodied cognition".

Sian Beilock, associate professor of psychology and leader of the Human Performance Lab at the University of Chicago, defined the term this way: "In contrast to traditional views of the mind as an abstract information processor, recent work suggests that our representations of objects and events are grounded in action. That is, our knowledge is embodied, in the sense that it consists of sensorimotor information about potential interactions that objects or events may allow."

She cites a more complete definition of the concept in Six Views of Embodied Cognition by Margaret Wilson. Another terrific overview of the concept is provided by science writer Drake Bennet of the Boston Globe in his article, "Don't Just Stand There, Think".

In a recent study, "Sports Experience Changes the Neural Processing of Action Language", Dr. Beilock's team continued their research into the link between our learned motor skills and our language comprehension about those motor skills. Since embodied cognition connects the body with our cognition, the sports domain provides a logical domain to study it.
Their initial look at this concept was in a 2006 study where the team designed an experiment to compare the knowledge representation skill of experienced hockey players and novices. Each group first read sentences describing both hockey-related action and common, "every-day" action, (i.e. "the referee saw the hockey helmet on the bench" vs. "the child saw the balloon in the air"). They were then shown pictures of the object mentioned in the sentences and were asked if the picture matched the action in the sentence they read.

Both groups, the athletes and the novices, responded equally in terms of accuracy and response time to the everyday sentences and pictures, but the athletes responded significantly faster to the hockey-specific sentences and pictures. The conclusion is that those with the sensorimotor experience of sport give them an advantage of processing time over those that have not had that same experience.

This may seem pretty obvious that people who have played hockey will respond faster to sentence/picture relationships about hockey than non-hockey players. But the 2006 study set the groundwork for Beilock's team to take the next step with the question, "is there any evidence that the athletes are using different parts of their brain when processing these match or no match decisions?" The link between our physical skill memory and our language comprehension would be at the base of the embodied cognition theory. 

So, in the latest research, the HPL team kept the same basic experimental design, but now wanted to watch the participants' brain activity using fMRI scanning. This time, there were three groups, hockey players, avid fans of hockey and novices who had no playing or viewing experience with hockey at all. First, all groups passively listened to sentences about hockey actions and also sentences about everyday actions while being monitored by fMRI.  Second, outside of the fMRI scanner, they again listened to hockey-related and everyday-related action sentences and then were shown pictures of hockey or every day action and asked if there was a match or mis-match between the sentence and the picture.

This comprehension test showed similar results as in 2006, but now the team could try to match the relative skill in comprehension to the neural activity shown in the fMRI scans when listening. Both the players and the fans showed increased activity in the left dorsal premotor cortex, a region thought to support the selection of well-learned action plans and procedures. 

You might be surprised that the fans' brains showed activity in the same regions as the athletes. We saw this effect in a previous post, "Does Practice Make Perfect", where those that practiced a new dance routine and those that only watched it showed similar brain area activity. On the other side, the total novices showed activity in the bilateral primary sensory-motor cortex, an area typically known for carrying out step by step instructions for new or novel tasks. 

When playing or watching, we are actually calling on additional neural networks in our brains to help our normal language comprehension abilities. In other words, the memories of learned actions are linked and assist other cognitive tasks. That sounds pretty much like the definition of embodied cognition and Dr. Beilock's research has helped that theory take another step forward. Beilock added, "Experience playing and watching sports has enduring effects on language understanding by changing the neural networks that support comprehension to incorporate areas active in performing sports skills."

So, take pride in your own brain the next time you hear, "Kobe dribbles the ball to the top of the key, crosses over, drives the lane, and finger rolls over Duncan for two." If you can picture that play in your mind, your left dorsal premotor cortex just kicked into gear!