Book Review: How We Learn to Move
Do we need to "watch film"?
How We Learn to Move is by Rob Gray. The subtitle is "A revolution in the way we coach & practice sports skills". In the book he never gives one overarching name for the revolution, but it involves a few main ideas:
embracing variability (both in how we practice and in how each unique person performs an action)
a holistic philosophy focusing on more "realistic" situations while practicing (playing the sport we're trying to get good at instead of running drills and using cones). "Because there is not a linear flow of separable processes involved, they can’t be broken out into modules, pulled apart to train, and then put back together."
On the whole, my coaching philosophy is pretty close to his. I've written about variation in practice and the problem of transfer in previous articles, namely An introduction to the science of learning, for frisbee throwers and How to practice throwing. At times in this review I'm pretty critical of his ideas, but that's a reflection of small differences in our understanding when we're pretty close to each other in the big picture.
There is no one true throw
Let's start with an area where I'm totally on board with Gray's ideas. There's long been a belief that being good at throwing (which I'll use as an example here, but the arguments apply to other motor skills) is being able to repeat the same action, over and over again. And in a way, that's not too far from the truth. But it's more complex than that.
Let's talk about a simplified example: the backhand throw in frisbee. A real throw obviously involves the movement of many body parts (wrist, elbow, shoulder, truck, leg placement, etc), but for our simple example, let's pretend there's only two variables: the rotation of the shoulder and the rotation of the wrist. The mistaken understanding of expertise is that throwing consistently relies on being able to reliably reproduce the same shoulder rotation and the same wrist rotation. But that's not exactly the case.
For a given throw, there will be a number of different ways we could get the disc to the zone where our target can catch it: lots of wrist rotation but little shoulder rotation; lots of shoulder and less wrist, or anywhere in between. In reality, we don't consistently apply the same exact amount of wrist rotation. Instead, our bodies learn to reproduce the desired outcome while compensating for natural variations.
This is not just a theory, but has been shown in studies. I'm not going to explain all of Gray's ideas in this much detail, but I'm doing it for this idea, mostly because I really loved this graph:
Imagine on the y-axis we have our "shoulder rotation" and on the x-axis our "wrist rotation". The pale red stripe represents the zone where the combination of those factors gives us a successful throw. The black dots ("pre-test") represent a non-expert thrower, and the white dots ("post-test") represent an expert. The expert thrower is able to consistently get their throws much closer to the desired zone that represents a successful outcome.
But they don't do this by completely wiping out their variability, and recreating the same exact throw each time. Instead, their body works together as a whole, with one part of the motion compensating for variation in other parts of the motion. (Their y-axis value does seem to have less deviation, but the x-axis value actually looks like it has more deviation.)
Every throw really is unique, even if there are only tiny differences each time. We should embrace that variability and build it into our practice. I really liked the way he phrases it here:
Weather conditions, different shot directions and opponent’s tendencies provide an ever-changing external context while things like fatigue, growth, and micro traumas in muscles from a workout the day before create a backdrop of variability from the inside.
Repetition without repetition, but don't forget the repetition
Gray's motto for practice is repetition without repetition. Or to expand a little bit:
repetition (of outcome) without repetition (of movement)
At times he seems to be suggesting that we should never "practice", and only "play". But he also spends a lot of time telling us that some of the newer practice methods that he helped invent are very effective (like the study used to create the graph above). So he clearly understands that to be great, you need to practice a whole lot. But I worry that message gets lost in his anti-repetition crusade. He spends an extended paragraph quoting people who believe in repetition (like the John Wooden example above), only to argue against them. Later in the book he does the same thing with the term "automatic":
Becoming automatic through training has been held with the same regard as our old friend repetition and they two are typically thought to go hand in hand. For legendary UCLA basketball coach John Wooden, “The importance of repetition until automaticity cannot be overstated”. Norman Peale emphasized that:” repetition of the same thought of physical action develops into a habit which, repeated frequently enough, becomes an automatic reflex”. For Celtics player Bill Sharman: “It’s a game of habit or repetition. It’s a reflex. The game is so quick you don’t have time to think.”
But he's not arguing that you don't have to practice a lot (assuming that "practice" is done the right way). He's arguing against a very specific definition of the terms "repetition" and "automatic". His own path still involves a lot of hard work, and it still involves experts having a high level of repeatability in their outcomes.
In other words, he's in agreement that the pros are automatic, if automatic means "able to complete 99 out of 100 passes". He's just saying that automatic doesn't mean "able to repeatably bend their elbow at exactly 47 degrees on every single windup". In my mind, these old greats knew what they were doing, they just didn't have the words to describe it accurately.
The core issue of transfer
Transfer is a measurement of how much a skill drilled in practice will be usable in-game. Gray argues that skills practiced outside of the chaotic, variable context of real games won't transfer. I'd like to argue here for a more nuanced approach.
Gray holds up the way Brazilian kids learn to play soccer as an example of the free-flowing, exploratory new style:
for many young players (including most famously, Pele) a large proportion of practice occurs via a customary pickup game called ‘pelada’. The term, which can be translated from Portuguese as “nude”, captures that fact that it is typically played in a naked environment absent of proper grass fields, nets, line markers or coaches. It usually played outdoors on very irregular surfaces including dirt streets and sandy beaches... there are no structured leagues as players of both sexes and all ages typically play together...If training in Brazil is naked, the emperor in England and many European countries like Germany, Spain, Portugal and the Netherlands is fully clothed!
And he uses UCLA coach John Wooden as an example of the old learning style:
This type of rote repetition of the “correct” technique has long been held to be the key to learning and becoming skillful. Legendary UCLA basketball coach John Wooden once cleverly proclaimed that: “The eight laws of learning are explanation, demonstration, imitation, repetition, repetition, repetition, and repetition”
But John Wooden won 10 national championships, and Brazil hasn't finished better than 4th at the (men's) World Cup in 20+ years! Clearly, the old style works too. Gray seems to imply that "repetitive" practice would result in no transfer, but that clearly can't be the case. At best, it seems that Gray's program will give us slightly better results (while possibly also being more fun).
I have another problem with the way he frames transfer and repetitive practice. In my mind there's this core contradiction that he isn't aware of: If it's OK to vary things from the way they happen in real games (because transfer will happen), then repetitive practice will have some transfer. If it's not OK to vary things from the way they happen in real games, then all of his training ideas wouldn't work either (real baseball players aren't wearing VR glasses or hitting balls over walls).
Gray seems to want to have it both ways—transfer happens (when he wants to argue for variability over repetition in training programs), but transfer doesn't happen (when he wants to argue for realistic games over drills).
A great example of this is his discussion of baseball star Albert Pujols being struck out by softball star Jennie Finch:
Jennie Finch was one of the most successful softball pitchers of all time (2 time All American and Olympic Gold medalist) while Albert Pujols has had a hall of fame career as a baseball batter. Surely, just using a bigger ball, moving the mound a little closer and having the pitcher throw underhand shouldn’t really affect a highly skilled batter like Pujols. But...he did as well as you or I could do. In three pitches he did not once even make contact with the ball never mind getting a hit...
When we break this relationship apart by de-coupling perception from action it fundamentally changes the nature of the task the athlete is performing. This makes it much less likely that there will be transfer of training. For example, Pujols’ many years of practicing hitting a baseball didn’t seem to transfer at all to hitting a softball.
But elsewhere in the book, he argues against the idea that:
the things we vary in practice (speed, distance, spacing of players) should [only] be things that will vary in the game. It makes no sense at all to add something that would never occur (e.g., a larger than normal ball or closing one eye)
So which is it? Do skills transfer across using different sized balls, or not?
Albert Pujols's skills don't immediately transfer to being able to hit a softball, but it's pretty obvious Pujols is much closer to be able to get a hit against Jennie Finch than I am. I don't think it's true that Pujols's skills wouldn't "transfer at all". He just needs to add softball-specific skills to the transferable baseball skills, before that transfer can become apparent.
I do agree with Gray that transfer is a big problem in learning. And honestly, I think he's kind of right that there's a balance—we want to find the right drills that will actually transfer. Let me give a specific example. Here's a paragraph I really liked:
The problem here is that skills like “agility” and “ball handling” are functional and driven by information from the environment. But they are not being practiced that way. As an attacker you go left with the ball because the defender is leaning right or moving towards your right side in attempt to steal the ball from you. You are deciding what to do and acting using information from your opponent’s body posture and movements. These movements have a purpose – getting around a defender, avoiding losing the ball, etc. We don’t cut right or left when running, dribbling, or puck handling just for the fun of it. When we dribble around cones there is no information that we can couple our actions to.
I've noticed this coaching frisbee. Inexperienced players will sometimes "cut" by running in a V-shape, regardless of what their defenders are doing. But cutting isn't fundamentally about making V-shapes, it's about using your defender's momentum against them. We need to teach cutting in a way that, from the beginning, teaches players to "use information from their opponent's body posture and movements".
So we can have transfer between drills and games, but we need to think deeply about whether those drills fundamentally contain the same challenges that a real game do ("start running this way while the defender is still running that way") or not ("take two steps this way, and then run that way"). One of his suggestions, which I really like, is playing tag.
A final comment on transfer: Gray brings up the concept of freezing in learning a new skill. In short, this is when athletes simplify a new movement by not using their full body. In frisbee, we can think of beginners learning to throw a forehand by flicking their wrist but not moving (i.e. freezing) their elbow. Gray says this is a natural growth stage:
So, in sum, if the athlete you are working with self-organizes into a freezing solution it can be good in the short term but, at some point, we need to vary the constraints in practice to encourage them to “free” themselves and find a better one.
But this itself is an example of unrealistic practice effectively transferring to more realistic situations. The beginner frisbee player who freezes their elbow while learning to flick their wrist isn't wasting all of that practice time—the lessons they learn about flicking their wrist stay with them in the future when they start moving their elbow in a more realistic way. Transfer happens even though the motion isn't fully "realistic".
What is the timescale for transfer?
Let's go back to that Albert Pujols story for a moment. Gray gave Albert Pujols three swings of a baseball bat before he was ready to claim that there was no transfer between baseball and softball skills.
This is part of a general pattern. Gray biases the science in favor of his favorite conclusions. Albert Pujols gets three swings. But studies researching the self-organizing methods he prefers get a much longer leash:
In a 2017 study, 10-11 year-old soccer players completed 22 weeks of practice involving small-sided games. They were not given any traditional, technical instruction but instead the goal was for them to learn skills like dribbling and passing “in the game”... While there were not improvements from 0-11 weeks, there were significant improvements in both decision making in execution of the skills by the end of training. This seems to be consistent with what we see when we use other types of manipulations that encourage self-organization in practice like increasing variability. [I added the emphasis]
Here’s a third example: I've messed around a bit with little "brain training" computer games. Do they actually help make me a better athlete in any way? I'm not sure. Gray, of course, is not a fan of these apps that abstract the skill of 'noticing things' too far outside of the sport-specific context. Here's his argument:
For example, in 2019 Formenti and colleagues studied the effects of training on a system called Neurotracker for improving volleyball performance...What was found in this study? After training, the volleyball players were much better at doing the Neurotracker task but there was absolutely no improvement in setting, passing or serving accuracy on the court as compared to a control group.
I looked up the study and wasn't at all surprised to find that it lasted...just six weeks. As I mentioned in a recent article before I read How We Learn to Move, I'm a believer that transfer (and skill acquisition generally) can take 18 weeks or more. There are skills that can take years to fully integrate. Sometimes we need to build up one sport-specific building block before a skill that we already have from another sport can transfer successfully. Admitting the timescales of interest are this long definitely makes the science a lot harder. But doing hard science is a lot better than not learning the truth of how the world works because any effect that takes longer than 6 weeks to show up is assumed to not exist.
Gray admits that this book is "a movie trailer. Its purpose is to introduce you to some of the main characters and ideas [of his movement]". But in this day and age, anyone writing about scientific topics should have fully integrated lessons from the replication crisis. It should be possible to write a simple, readable introduction to your movement without twisting the science.
Practice expands what we can do
Gray admits one benefit to simplified practice: expanding our capabilities. In other words, there are more options we have access to if we're more athletic:
The other main way constraints can be manipulated to create action opportunities for a performer is by changing their individual constraints through appropriate strength and conditioning training...goalkeepers with quicker movement times consistently waited a bit longer before starting their dive and subsequently got more information about where the shot was going.
I think Gray is arguing against himself again here—because there is no clear bright line between "strength and conditioning training" and "unrealistic practice". For example, in ultimate frisbee, a faster spinning disc will have a more stable flight path. Developing the movements and muscles we use to make the disc spin really fast could be called "strength training", but because it stabilizes the flight path it also makes our throws more accurate. So is simple throwing practice useless (because it doesn't represent real game situations), or is it effective (because we are less constrained by accuracy and distance when we have the ability to make the disc spin faster)?
Again, I feel like Gray misses an opportunity to believe in a spectrum—in a holistic philosophy, there should be no hard separation between body and mind. If simplified practice can help reduce our physical constraints, why wouldn't it also be able to reduce our mental constraints? Practicing skills out-of-context gives us just that little bit more mental strength we need to believe we'll be able to perform them in-context.
How much variation is right?
One nuance that I appreciated is Gray's comment on finding the right level of variation for different levels of athlete. He comments that "finding the right amount of noise seems to be one of the key skills a coach needs to have", and later offers this advice:
We want the amount [of variability] we add to be inversely related to the amount of inherent variability the athlete brings to practice themselves.
In other words, newer athletes who are less in control over their sport can be safely presented with simpler practice situations, while experienced athletes who deftly control their in-game environment should be given lots of challenging complexity in practice. New players can grow from simpler drills, while experienced players who have already refined their skills need to focus on being able to perform those skills in complex, realistic environments.
This is something I've thought about previously, but I'd come up with a different way to conceptualize it. I've often seen this simple diagram of our comfort zone, learning zone, and panic zone:
For people to grow, they need to push outside their comfort zone and challenge themselves with skills they're not as familiar with. But go too far beyond their capabilities, and they'll go back to not being able to learn—things are just too confusing and too overwhelming.
This is a balance that we need to consider at the same time we consider the balance between "high transfer from realistic situations" and "low transfer from simple drills". Yes, the transfer should be better if our players are playing the game itself, but when that game is too complex, when it puts them in the panic zone, the benefits of realism are wasted on players who are a little too confused to learn effectively.
The problem of tired legs
Another weakness I see in Gray's arguments is that he never mentions anything about the problem of tired legs. As an old guy, I like throwing practice because I get reps without needing to put miles on my knees. I don't really care if the transfer is 10%, because I don't feel comfortable playing more sports, but I do feel comfortable with lots of throwing practice.
Although he mentions other sports like tennis and soccer, many of Gray's examples come from baseball. The stats revolution (sabermetrics) happened much earlier in baseball than in other sports because it is so easily broken down into its component parts. It's easier to determine if a hitter adds value to a team because they are up at the plate by themselves, and they either make good contact with the ball or they don't.
"Drilling" baseball is not much different from actually playing baseball—you can do lots of realistic batting practice without drastically increasing the risk of tearing your ACL. The same is not true of throwing the frisbee in real game situations. We need lots of real-game experience, but a more mature analysis would discuss the tradeoffs—throwing practice has less transfer, but we can do much more of it safely.
Motor skills vs. strategies
As I covered in the first section, I'm sold that Gray's technique is useful for learning motor skills. Where I'm not sold is that the same applies to the more strategic aspects of sports. Gray argues against a top-down decision making process, and uses the example of watching film:
A common example of this in sports is the use of video to help an athlete learn what to “look for”. A football quarterback watches unfolding plays and is asked to give a verbal response about what they would do: “I would throw deep” or “I would dump the ball off to the running back”...But there seem to be some fundamental flaws with this...model.
This goes very strongly against my experience as a sports-player and sports-watcher. Am I really supposed to believe that all the time Tom Brady, Peyton Manning, or LeBron James have spent studying game film have nothing to do with their success? Of course there's going to be an integration process—watching film is effective only when we combine it with lots of actually playing the sport.
We can imagine motor skills vs. strategic thinking as a spectrum—near one end is a game like chess, which has intuition but also involves players making explicit mental calculations of possible future states. On the other end are pure motor skills, like hitting a 95 mph fastball, that are best done without any explicit thinking. Playing field sports like frisbee or football would fall somewhere between these two extremes: best done in a flow state where you're not really thinking too much, but complex enough to benefit from explicit "book learning"—at least once you've had time to absorb and integrate those lessons.
It disappoints me that Gray isn't more of a "spectrum guy". I feel like "everything is a spectrum" thinking would fit perfectly with his "everything is variable" and "everything is complex" philosophies. But instead of transfer being a spectrum, he shoves everything into the "transfers" and "doesn't transfer" boxes. Instead of strategizing being a spectrum, he shoves everything that isn't reading a book into the "no thinking allowed" box.
In my recent coaching experience, it has been obvious that some explicit discussions can transfer to implicit flow-state knowledge. I've seen this pattern often:
First, I have a discussion on the sideline with a player: hey, here's a cool skill I'd like to see you try in a game. They try it out on the sideline a few times.
Second, they get in the game and need a little push from me to try out their new skill (they only see the opportunity after I yell from the sideline "hey [X], do the thing!")
Third, they're able to use the skill independently in a game.
Gray doesn't believe that mental models exist, saying:
For all...of these learning processes, we do not need to infer some internal processing or computation inside the performer’s head. All that is occurring is change in the relationship between the performer and their environment... in sum, the control of actions can be explained much more simply, more parsimoniously, as the establishment (and adaptation) of information-movement control laws in which the performer directly picks up some action-relevant information from the environment and uses it to regulate their actions. No need for prediction, information processing or assessing memories of previous actions.
But again, his examples are usually simple motor skills: heading a soccer ball, hitting a baseball, landing your foot on the board when long jumping. His arguments make sense for these motor skills, but can't agree the same is true for the more strategic aspects of sports: a quarterback deciding where to throw, or a defender deciding how to guard a player based on their known tendencies. Whether or not the knowledge is fully explicit, if we have access to it, it must be stored somewhere in our brain. He says there's "no need to prediction", but I think it's obvious that players like LeBron succeed on being able to see a future that their opponents haven't seen yet.
Final Thoughts
Although there are some areas where I've disagreed pretty strongly with Rob Gray's opinions, overall there's a lot of places I agree with him.
Where we agree:
There is no "one true throw". Expert skill involves the adaptability and fluidity to create successful outcomes in an infinity of (slightly) different situations.
Variation in practice is more effective than trying to do the same exact thing every rep.
Transfer is a big challenge, in sports and other learning situations.
No matter how much we (for example) practice throwing, we need to do lots of throwing in real games, too, because there will always be aspects of practice that don't transfer.
Drills are dangerous, and require careful thought to make sure they actually represent the information we're using in a "real game"
Where we still seem to disagree:
It's obvious to me the "traditional method" has enough success stories that it can't be that much worse than the new way (ignoring that the new way is more fun).
He picks and chooses saying this will transfer, that won't transfer. I think we can optimize transfer, but even "repetitive" practice will have some level of transfer. Everything is a spectrum, including transfer.
Most of his examples involve simple motor skills (throwing a ball, hitting a ball, catching a ball) but he seems to believe the same conclusions apply to the more complex/strategic situations in sports. I think it's obvious that "book learning" can help us with things like "reading a defense".