Category Archives: blog

What I’ve Learned About Motor Learning

The ball bounces up and down on your foot and it feels perfect. You’re connected with the ball somehow—just by tapping it and letting gravity retrieve it. You’re concentrated but you can’t really describe how you’re controlling the ball. And it wasn’t so automatic years ago. How did it get to be this easy?

Think like a roboticist. Imagine programming a machine to learn foot juggling. How would you solve this problem?

You might ask a robot to try various poses to lift its legs so it’s most stable. Maybe you want it to judge the spin of the ball in the air. Maybe the trajectory. Maybe you want it to find the optimum leg angle of contact for each particular environment. Do you use trial and error or gather more data to run fast simulations after each kick?

Humans learn movements (a.k.a. motor skills) in two ways:
By revising models of ourselves, other objects, and the environment
By revising commands to optimize accuracy, speed, and energy

The first mechanism we use is to create predictive models for our movement. We have models of ourselves, objects, and other living things in our impressive model repertoire. Every time you lift that coffee cup to take a sip, you’re drawing upon the model you’ve made of your arm, your wrist, your tired fingers, and your particularly thick porcelain coffee cup. You’re even using a model of how much your last sip weighed in order to expertly drink again. Of course, none of these models slip into your consciousness because you’d be overwhelmed at breakfast—It’s automatic.

To learn is to refine your models to have less error.

In a study, people watched a video of a basketball free throw shot and were asked whether the ball would go into the basket. Unsurprisingly, expert basketball players did better than novices. Players predicted outcomes significantly better than chance from footage of a free throw that stopped before the ball was even released. These players have created predictive models based on body movement not on ball trajectory—a set of models that give them an edge when a millisecond advantage will get them possession of the ball.

For me, this redefines the expertise of an expert athlete. Instead of robotically sculpting movements and beefing up muscles to execute commands in an inactive brain, an expert athlete is comparing and contrasting previous models of him/herself, other athletes, and the environment to make decisions in a fully engaged cerebrum. Simply watch Lionel Messi dribbling by world-class defenders in slow-motion—he’s quite literally playing a different game than the players around him because he’s observing and processing information differently.

The second mechanism we use to find our movement solution is to employ a stockpile of movement commands. These are our baseline actions we issue with each shift in our seat, each turn of the head, and each opening of our mouths. Over time, these commands are altered and perfected to be more nuanced and more complex. This kind of learning is more in line with my idea of an expert athlete who has a large repertoire of easy commands.

To learn better commands isn’t to reduce our error, but to make them faster, more consistent and use less energy—our movements become effortless.

To recapitulate:
In the first mechanism, we reduce our error over time and we call our error reduction, learning. In the second mechanism, we learn when we improve our speed, accuracy, and energy efficiency in our commands to our limbs.

To learn motor commands and models is to make memories that are necessarily unstable. Memories are strengthened and distorted each time we recall them. There’s even a term in biomechanics called the “forgetting factor” that begins to eat away at your motor abilities just seconds after the action. We can pick up oft-executed actions again after long inactive spans with ease because we’ve built up such a plethora of commands and refined models that remain not as a single memory but a group of memories waiting to be remembered. Somehow we’re able to jump back on the bike, ski down the mountain, or throw a ball after years of inactivity.

MODIFYING COMMANDS/MODELS

We gradually learn new movements by fine-tuning our models of ourselves and the world and by making better commands (see the above descriptions of these mechanisms). How do we go about doing this fine-tuning? We must first make errors. An error alerts our brain that we must change something to have a desired result. But what drives this change in our movement?

Before we can say anything, we have to throw down our qualifiers: the movements scientists study are from very controlled/contrived learning situations—mostly standing on a platform, moving a robotic arm with a hand and with this movement, hitting a target on a screen. Scientists change the feedback though the screen and through the robotic arm. The results from these situations are early and not as generalizable as you might like either. For starters, the reaching movements they’re testing are fairly novel to a largely college-aged population in the middle of America. However, there are some interesting results:

1. Our subconscious undermines our conscious control of movements. If you perceive an error resulting from your movement, you will automatically adjust and explicit, top-down commands from your conscious control simply get in the way of this subconscious and automatic adjustment. This informs training in an interesting way: you simply need to let your body go through the movement to allow itself to correct. Instruction from coaches and yourself might give you different ways of trying to correct movement errors, but it’s your subconscious that needs to recognize and correct your movement. Stroke patients illustrate this concept nicely. It’s common for people who’ve suffered strokes to walk asymmetrically. When they walk on a special split-belt treadmill that speeds up under one of the feet, the stroke sufferers adjust and walk somewhat normally. Then when the treadmill runs normally for both feet, the stroke patients walk normally because their subconscious has recognized and corrected for movement errors that aren’t registered in everyday life for some reason.

2. We learn quickest with punishments, but rewards help learning and are better for retaining knowledge. Like most learning, we can speed it up and retain it better with rewards and punishments. Some of you reading this might perk up, thinking you can gain an edge in your child’s soccer or violin development. However, again I hasten to warn you that this hasn’t been studied in complicated movements nor is there a clear way to reward or punish a soccer player while training. BUT, for the immediate improvement of both our modeling and commands, penalty seems to be the winner. Interestingly, we retain our motor learning better when it’s reinforced by rewards. Punishment is a stronger signal because of humans’ loss aversion: a loss stings more than the equivalent gain.

Bias of aped autobiographical learning

I thought learning done at school was straightforward when I started teaching nine years
ago. I knew I could help students from low socioeconomic backgrounds grasp the joy of learning.

But I didn’t know that I was working with a very narrowly defined sense of learning: the process of gaining linguistic, historical, scientific, and mathematical information. I didn’t really think other types of information belonged inside school, nor could they help disadvantaged students.

I was wrong (obviously). But why was I biased toward this kind of learning? I couldn’t articulate this question even when I researched learning for Sift.

Screen Shot 2015-08-26 at 3.34.45 PM
To understand my bias, let’s start by defining “learning.” Roughly, learning is the act of gaining new knowledge. And there are different types of knowledge: motor and autobiographical.

Babies learn both types of knowledge very clearly – they learn how to roll over (motor knowledge) and learn to dislike a shape of food (autobiographical knowledge). Adults learn both types of knowledge as well.

Our extreme autobiographical talent separates us from other animals and artificial intelligence. We’ve evolved to extract certain patterns from complex stimuli in our environment. My claim is that because we are so good at autobiographical learning, we treat it very oddly: We revere and fear it. This claim will take a book worth of argument and evidence to be convincing. However, let’s start with my bias toward a narrow version of autobiographical knowledge.

The explicit curricula now taught in schools service standardized tests – the gates of universities, well-paid careers, etc. Standardized tests are very rough attempts to measure narrowly defined autobiographical knowledge. I say this because they measure your ability to recall a laundry list of facts and standard interpretations that have simply accrued in textbooks over the last century. There are a few grammatical and logical skills thrown in as well without any evidence showing they deserve to be learned. Thus, without realizing it, the message given to teachers and students is that our ideal educational product is someone who can give clear reasons for the Holocaust and recite the steps of DNA transcription and translation. This imposed curriculum is where I got my bias toward a stilted idea of learning.

 Standardized learning objectives and tests only ape autobiographical talent. Students don’t extract patterns from complex stimuli to gain new knowledge directly relevant to their life. They develop ways of recalling, constructing a tenuous self-interested motivation based on a future promise of “a good university,” which needs another set of justifications to excel in and “a good job” after. Good teaching is often tricking children into connecting themselves to the content through various activities or even a personal relationship to the teacher. In other words, the “auto” in autobiographical is stripped out of schools.

Who would do better at a given task: a person who is directly motivated by self-directed curiosity or a person who is indirectly motivated by promise of a future reward? I would think the directly motivated person would do better, but indirect motivation is powerful. I don’t know which one is more psychologically healthier. Again, a promised future reward is a very real motivator for many people whether they are aware of it or not. However, there isn’t authentic chance for self-directed autobiographical learning in school curricula.

Our educational system loves to judge a student’s ability and compare her to another student and for good reason: University admissions officers need ways by which they admit students. The true paradigm-shifting educational innovation would be a way of determining a students’ autobiographical learning ability. Standardized tests attempt this, but manage to cultivate a much different skill set of indirect and aped autobiographical learning.

There is still a lot of hope in educational reform. MIT has thrown out lecture-style classes in favor of inquiry-based, peer-guided classes and has improved learning outcomes. They have put the student’s curiosity back into their learning. Pedagogical trends in high school now favor self-directed behavior but are hamstrung by overly stuffed curricula. Perhaps change is coming?