The Core Challenge of Public Education

Why school is the way it is

I wrote an essay for a “Non-Book Review Contest” hosted at the Astral Codex Ten Substack. It ended up being a finalist. The essay was published anonymously as part of the contest back in July. I’m publishing a different version here, with the same basic argument but some modifications and updates. If you read the original version, everything from the “What About Tracking” heading onward is new, including a new section on Alpha School.

“Democracy is the worst form of Government except for all those other forms that have been tried from time to time.” — Winston Churchill

What Do Schools Do?

Imagine for a moment that you visit 100 random classrooms in 100 random schools across the U.S. You’ll be impressed by some teachers; you won’t think much of others. You will see a handful of substitute teachers struggling to manage their classrooms. You’ll see some schools where the energy is positive and students seem excited to learn, and others where learning feels like pulling teeth. Two commonalities you might notice: first, in the vast majority of classrooms, the students are grouped by age and taught the same content. And second, you might notice that the learning isn’t particularly efficient. Many students already know what is being taught. Others are struggling and would benefit from a much slower pace and more support. You will see plenty of sitting around waiting for the next thing to happen, or activities that seem designed to take up time rather than maximize learning.

What do schools do? Your first thought might be that schools exist to maximize learning. Observing 100 random classrooms may disabuse you of that notion. It sure doesn’t seem like school is doing a good job of maximizing learning.

Thesis

Here’s my thesis. On the surface, it seems that the primary function of school is to transmit knowledge, and so the bottleneck to a more efficient education system must be knowledge transmission. But that’s not the case: the true bottleneck in schools is motivation.

This might seem like a silly thing to say. During those 100 classroom visits you might have seen a lot of classrooms with a lot of students who don’t look very motivated. The core design of our schools — age-graded classrooms where all students are expected to learn more or less the same curriculum — is the worst form of motivation we could invent…except for all the others. While school is not particularly effective at motivating students, every other approach we’ve tried manages to be worse. Schools are a giant bundle of compromises, and many things that you might intuitively expect to work better simply don’t.

The important thing to remember is that when I talk about school, I’m talking about tens of millions of students and a few million teachers in the US. You might say to yourself, “I wasn’t very motivated in school.” Sure, I believe you. The goal isn’t to motivate you, it’s to motivate as many students as possible, and to do it at scale. If you have a boutique solution that works for your kid in your living room, that’s nice, but it isn’t likely to scale to the size at which we ask our education system to operate.

There Must Be a Better Way

Educational thought leaders have long argued that we can do better than our current system. A common theme has been personalizing learning: allowing students to go at their own pace, rather than forcing everyone to learn at the same speed.

The push for individualized instruction dates back to the early 20th century. Sydney Pressey, a psychologist at Ohio State University, built the first “teaching machine” in 1924. His device, a mechanical testing apparatus resembling a typewriter, allowed students to answer multiple-choice questions and receive immediate feedback. It was largely dismissed as a gimmick. In the 1950s, B.F. Skinner revived the idea with his own teaching machine, designed around operant conditioning principles. Skinner argued that traditional classrooms were ineffective because they delayed feedback and forced all students to move in lockstep. His machine broke lessons into tiny steps, rewarding correct answers instantly. Like Pressey, Skinner believed technology could revolutionize education. But his machine was never widely adopted and mostly forgotten. In the 1960s and 70s, computer-assisted instruction systems tried to teach math through mainframe terminals but were expensive and impractical. In recent years, some of the boldest claims have come from Khan Academy. Founded in 2006, Khan Academy began as a set of lecture videos created by Sal Khan and has grown to include practice exercises with feedback, full curricula, and an AI chatbot tutor. Unlike earlier personalized learning tools, Khan Academy has seen broader adoption in real classrooms. It is a common element in personalized learning programs, which have been popular with tech billionaires who like to donate to education causes.

These are just a small sample of the myriad efforts to personalize learning over the years, to help meet students where they are and pull students from the lockstep of whole-class age-graded instruction. With all that enthusiasm, what were the results of the push for personalization?

Personalized Learning Has Failed

Intuitively, it’s reasonable that an education at your level and meets you where you are will result in more learning than just following the prescribed course of study for 6th grade or whatever. All else equal, it’s certainly true that instruction at your level will result in more learning. The thing is, we can’t hold all else equal. Schooling is a massive enterprise, and we can’t give every student instruction at their level without rethinking that enterprise. In general, when schools have tried, they have failed.

Last year, Laurence Holt published an excellent article summarizing the core challenge of today’s education technology. There is no shortage of fancy online programs that claim to teach kids math. Khan Academy was the first to gain widespread popularity, but it’s actually used much less now than some newer entrants like iXL and i-Ready. Each program commissions some study showing that students who use their program with fidelity learn more than those in a control group. Holt digs into the data, and it turns out that the group who used the programs with fidelity was often around 5%. The article is called “The 5 Percent Problem.” These programs do seem to help a subset of students, but don’t do much for the rest. While Holt’s article focuses on math education, education technology has had a similarly lackluster impact on achievement in English classes. We know that reading on screens leads to less learning than reading on paper, and the personalized learning apps have a similarly disappointing track record as in math.

This phenomenon isn’t limited to schoolchildren. Remember the MOOC craze of the early 2010s? Universities started releasing free or low-cost versions of their coursework online. Briefly, it was all the rage: MOOCs were supposed to democratize knowledge and disrupt higher education. Instead, completion rates were low, and the MOOC mostly died an unceremonious death. Numbers vary depending on the source but 10% completion is a generous median, the same order of magnitude as the 5% problem in K-12 education technology. The vast majority of people who sign up for a course never finish. Many MOOCs are still online and get plenty of views on YouTube, but we’ve learned that most people need more than course content posted online in order to learn. The big difference between MOOCs and school is that if you don’t finish that MOOC on the U.S. constitution, life goes on. If a kid doesn’t learn to read, they’ll be at a disadvantage for the rest of their lives.

The core problem with these online programs is having every student work independently, without any connection to what the students around them are learning. That just doesn’t motivate many students. Psychological research consistently shows that humans are conformist creatures. We instinctively align our behaviors to group norms. Classic studies like Asch’s line experiments, where 75% of participants denied obvious truths to match group answers, remind us that humans are social and prioritize conformity. This tendency isn’t just peer pressure, it’s evolutionary wiring. For our ancestors, conforming boosted survival by maintaining group harmony and reducing conflict. Today, this manifests in classrooms where most learners thrive on collective routines. Conformity explains why personalized learning often fails. Most students need the social scaffolding of lockstep instruction, even when it’s inefficient. Conformity isn’t a perfect solution, but it’s the best one we have.

One form of learning that has been shown to be particularly effective is deliberate practice. Deliberate practice involves being pushed outside of your comfort zone, focusing on specific, concrete goals to improve performance, and getting consistent feedback. One common characteristic of deliberate practice is that it isn’t particularly fun. Most contexts where deliberate practice is common, like sports and music training, involve expert, individualized coaching. The coach is mostly there for motivation. The coach does other things as well, but the most important thing a coach can do is motivate you to train. Deliberate practice isn’t common in school learning, but it’s a good reminder that motivation is the key to lots of forms of learning in and out of school. Learning isn’t always going to be a ton of fun. In the absence of one-on-one tutoring for every student, conformity is the best tool we have to create the motivation necessary for learning.

But Personalization Works for Some Kids?

Let’s explore personalization a bit more. Clearly this personalized learning thing works for some students. Maybe 5%. Maybe 10%. Why?

Here’s a broad generalization about learning. Let’s take the basics of learning to read as an example, where there is a wealth of data to back up the generalization. Some students will learn to read no matter what they experience in school. Often their parents teach them, or an older sibling or neighbor, and they pick it up quickly. Others more or less teach themselves. This group doesn’t benefit much from organized school, at least in terms of learning to read. We might call these “no-structure learners.” A second group needs the structure of school, but the quality of the teaching doesn’t matter too much. As long as they get the basics, a solid exposure to reading, and some support from a teacher, they will learn to read just fine. We might call these “low-structure learners.” Then there’s a third group. This group will struggle to learn to read. For this group, the quality of teaching matters enormously. Some will be diagnosed with dyslexia, though a strong course of synthetic phonics will reduce that number. Many will learn the basics but still struggle with things like multi-syllable words for years to come if they aren’t taught well. A carefully-sequenced, well-taught curriculum can make a large difference for these students. We might call these “high-structure learners.”

What does structure looks like? The details vary a bit depending on the age and the subject, but there are a few common elements. Structure means breaking complex tasks down into small pieces, focusing on one piece at a time, providing accountability in the form of humans who care that you are learning and intervene if you are not learning, giving frequent feedback, and providing a lot of practice. The human piece is what’s missing in too many personalized learning programs, and often the other elements can’t do enough to compensate for the lack of accountability.

Motivation is a spectrum. There aren’t sharp divisions between these three groups, and motivation depends on context, so students may have a different motivation profile for a different subject or a topic outside of school. Plenty of students fall in between those groups. Still, the broad heuristic of no-structure, low-structure, and high-structure learners is helpful in understanding how to structure public education. This is the answer to why personalization only works for 5% of students. Students need different levels of structure in order to succeed. Personalized, go-at-your-own-pace learning works for some kids who have more motivation, but doesn’t provide enough structure for most students.

The same phenomenon shows up in pandemic learning loss. Learning loss was concentrated mostly in the lowest-achieving students. Many high-achieving students did fine; these are students who didn’t need the structure of school, or for whom the minimal structure of online coursework was enough to keep them moving forward. The high-structure students who already struggle in school lost the most ground.

Here’s an anecdote from my personal experience. Online charter schools have spread rapidly in the last ten years. In my state it’s not very hard for students to enroll. I’ve had a number of students unenroll from our local public school and start at an online charter school. Most are back within a few months. They generally say one or both of two things: first, that they are bored learning on their own and they miss having people around. And second, they just weren’t motivated and didn’t learn much. Now to be clear, I’m in favor of having some online charter schools. They are a great option for some students — students who can summon the motivation, and students with outside-of-school circumstances that make attending school challenging. It’s the 5 percent problem. There’s that 5 percent, those no-structure learners or students in other unique situations, who benefit from options like online charter schools. But the vast majority do best in the schools we already have.

No-structure learners thrive anywhere. Low-structure learners need any coherent system. High-structure learners are much more sensitive to the quality of teaching, but trying to meet each student where they are doesn’t work very well. Lumping everyone together and asking them all to learn the same curriculum seems to work better at scale than anything else we’ve tried. These are the core challenges of education.

What About Tracking?

One theme of this post is that different students need different things to learn. A logical response is to say “Hey, what if we track students, and put them in leveled groups to do a better job of meeting them where they are?”

Here’s the core challenge of tracking:

That’s a bell curve divided into three equal areas. The most common way to do tracking is to divide students up into three equal-sized groups based on prior achievement, and try to adapt teaching to those students’ needs. The bell curve gives a hint as to why that doesn’t work as well as you might think. The middle group is clustered pretty closely together. You can teach pretty much as before. No problems there. But the groups on the left and the right include a huge range of achievement. The core problem I’m describing in this post is that schooling isn’t very efficient. If you walk into a typical classroom you’ll see a bunch of students who are struggling with the content and need more support, and a bunch of students who already know most of what’s being taught and are bored. Tracking can help a bit when it’s done well, but that shape of that bell curve means it can’t solve the basic problem: we have a huge diversity of students, most of them aren’t motivated to learn without the structure of a classroom and a teacher, and in that classroom we haven’t figured out a better strategy than teaching everyone the same thing.

In general in the U.S., schools rarely track in elementary school, they often but not always track in middle school, and usually track in high school. There are plenty of exceptions. But that structure follows for logical reasons: in elementary school, differences in achievement between students are relatively small. They balloon in middle school, and by high school they are so large that many schools have no choice but to track in some form. At the same time, student motivation falls off a cliff in high school. You can interpret this phenomenon in different ways depending on your priorities. You might see tracking as a way to segregate students and leave struggling students behind, or homogeneous grouping as a cruel sacrifice of the most talented students in the name of equity. The choices we make around tracking are values-based and there’s no right answer. But it’s worth looking at the central challenge of tracking as a problem of motivation. Tracking can help to solve the challenge of teaching students with a wide range of prior knowledge, but it only goes so far, and tracking without thoughtful support systems tends to leave struggling students even less motivated.

I think thoughtful forms of tracking have a lot of promise. Unfortunately, thoughtful forms of tracking are uncommon. I wrote a longer post on this topic here if you’d like to read more. But I’ll say that empirically, tracking doesn’t seem to make much of a difference. There are a wide range of studies on this topic, many of which are ideologically motivated in one way or another. But one commonality is that the effects are small. Here’s one representative meta-analysis looking at tracking and finding an effect near zero. You can look for more, and the effects are always small. I don’t doubt that tracking can be done well, and is done well in some places. But it’s not a magic bullet, and the core challenge of motivating students remains.

You might think that we’ve found the solution to tracking. We just need to get all the no- and low-structure learners together and let them move much faster. Here’s the issue. The no-structure learners will always be bored, as long as we are committed to putting them into classrooms where everyone learns the same thing. And those classrooms where everyone learns the same thing are exactly what the low-structure learners need. As soon as you create a higher track, there will be a ton of demand for it. Parents will insist that their kid join. And as it grows, it won’t be able to accelerate very quickly. The bell curve is just too spread out. You still need the structure of a classroom where everyone is learning the same thing, and that just isn’t a very efficient way to teach because that system is designed to solve the bottleneck of motivation, not the bottleneck of knowledge transmission.

What About Alpha School?

Wait, what about that article I read last week. Haven’t we solved all this with AI? Isn’t there some private school with no teachers, where students learn twice as fast in only two hours a day?

Alpha School has been in the news the last few months, and I think it’s a good case study in the importance of motivation in education. Here’s a representative quote from an article about Alpha School:

Fourth and fifth graders at Alpha School in Austin, Texas, aren’t just learning — they’re pioneering education’s new frontier. Every click and every keystroke is guided by artificial intelligence.

Students spend only two hours in the morning on science, math and reading, working at their own speed using personalized, AI-driven software.

On the surface this looks a lot like the hype cycle from the personalized learning tools we’ve seen for decades. Every generation is convinced that the latest technology will revolutionize education, and comes up with some promising-sounding anecdotes to support their hypothesis. The assumption is the same one education “experts” have made before: the bottleneck to learning is knowledge transmission, and technology has finally found a way to solve that bottleneck and unlock educational excellence, or something.

If you haven’t, I really recommend reading the Alpha School review from the same contest I submitted the original version of this essay to. The review is written by a parent of several Alpha School students. It skips the glossy marketing copy and gets into the nitty-gritty details of how Alpha School works. I think teachers should understand it, because the Alpha School model is compelling for lots of people inside and outside of education.

The first thing to understand about Alpha School is that for the most part it uses education software available and commonly used in schools across the US. A large chunk of the curriculum is on iXL, which claims on its website that 17 million students already use it. My district pays for iXL, though I don’t use it in my classroom. Not exactly cutting edge innovation. These tools are wrapped inside a broader system designed by Alpha School to do things like sequence topics and provide spaced repetition. The “every click and every keystroke is guided by artificial intelligence” bit is nice marketing copy, but at least from the perspective of that reviewer there is literally zero generative AI. While there is absolutely machine learning happening under the hood of these tools, it’s the same machine learning that has been common in schools for a decade.

The second thing to understand about Alpha School is how they approach motivation. They don’t just sit kids down and say, “ok, get to work on iXL.” The school has a tiny teacher guide-to-student ratio, an incentive system that more-or-less pays students for working quickly through the personalized curriculum, and a support system where students can book one-on-one tutoring calls if they got stuck with a lesson. This is the real secret sauce to Alpha School, the key to their eye-popping claims about fast learning. It’s also the part that rarely gets discussed in the articles about the school.

The reviewer writes about how his children weren’t able to enroll in Alpha School right away for logistical reasons, so he tried to have them work on the exact same online tools that Alpha School used to keep up with academics (in this quote, GT refers to the Gifted and Talented branch of Alpha School. Alpha School has several campuses in Austin with different themes):

We tried getting the kids to work on it for about an hour per day, but it was a fight every time. It was the same content they would be doing at GT, but without the GT structure, and it did not work.

But once the kids started at GT, those same iXL lessons became a game for them. I remember taking the kids to the park one day after school. They asked me, “Instead of playing can you set up a hotspot so we can do a few more lessons? I want to earn more GT-Bucks!”.

Students do in fact learn twice as fast as the typical student at a traditional school (as measured by the NWEA MAP assessment). The reviewer notes that practically speaking, students are often working for three hours a day, not the two hours that Alpha School claims in its marketing – and Alpha School radically increases the amount of “GT-Bucks” offered for learning over the summer break. But the results are still impressive. What can we learn from this?

First, a final important anecdote. Later in the review, we learn that Alpha School shares its software with homeschooling families. Students still do academics for two hours a day, but in the homeschool program the typical student learns at the average speed of a student at a traditional school, as measured by the MAP assessment. That still might seem impressive — finishing an entire school day in two hours! But if you talk to a few homeschooling families, you will quickly discover that two hours of academics a day is pretty normal. That doesn’t surprise me. A typical school might have students learning core subjects for four hours a day (the rest is breakfast, homeroom, specials/electives, lunch, recess, etc). Could a typical student learn twice as fast if you took away all the other students in the room, let them work at their own pace, and found a way to motivate them? Sure. There’s lots of inefficiency in schools. That’s the nature of mass public education. But even this supposedly cutting-edge technology can’t get beyond a modest speedup without the incentive system surrounding it.

So what can we learn from Alpha School? I honestly feel a bit sad when I think about it because it might genuinely be innovative. We know that there’s a group of students who are motivated to learn on their own. It seems like Alpha School figured out a way to increase that number, to get more students to learn quickly and largely on their own with personalized software. There is absolutely a selection effect happening here. The group of students who can pay tens of thousands of dollars for private school tuition is not representative of your average public school. But there’s still something traditional schools can learn here.

What makes me feel sad is that so many people get sucked up in the hype for new, “innovative” school models, and places like Alpha School insist on more-or-less lying about what makes their program successful. I wish we could be honest. Alpha School found some clever ways to motivate students. But if you read the articles covering Alpha School, they rarely mention the incentives. Everyone wants to believe that technology and personalization are coming to save us. And I understand why. If you walk into a typical school, you will be struck by how inefficient everything feels. You see students who need more support and aren’t getting it, and bored students who are ready to move faster. It’s rational to see that and assume there must be a way for technology to fix it, to solve the bottleneck of transmitting knowledge to each student efficiently. But that’s not the actual bottleneck limiting students from learning faster.

The interesting question to me is whether Alpha School can scale. What percent of students are able to learn more effectively with this model. Is it 10%? 20%? 40%? I doubt we will ever know, because no one wants to be honest about the model and Alpha School costs a fortune so we won’t be getting a representative sample of students any time soon. If another school tries to copy their “success,” they’re likely to copy all the wrong stuff.

Where Does This Leave Us?

School can be a bummer. Kids are bored. Kids bully each other, sometimes persistently and cruelly. Some schools struggle to manage behavior and resort to draconian restrictions. Some teachers are petty or vindictive. I don’t doubt any of this; I’ve seen all of it.

I want schools to be better. The point of this post isn’t that schools are perfect as they are. But it seems to me like everyone who wants to reinvent schools is focused on solving the wrong problem. The core problem, the bottleneck to mass education, is motivation. Every generation thinks that technology has finally arrived to solve the problem of mass education. But they are solving the problem of mass knowledge transmission, not mass motivation, and mass motivation is the core challenge we need to solve.

Here’s my bigger hypothesis: maybe the basic structure of education is fine. Not because it’s perfect, but because for the vast majority of students it’s the best option we have. If we accept that, we can work to improve schools within those constraints. Maybe that means finding ways to support independent learning for no-structure students who are self-motivated, or experimenting with ways to track students that will do a better job of meeting students where they are while still motivating them. Maybe we can design better personalized learning software that takes motivation more seriously than some silly points and badges. I think we have a lot to learn. But I doubt that the best solution to educating everyone is some futuristic technology-driven utopia. Education is messy. We will probably never find a perfect solution.

Here’s something you have to remember. It’s easy to cherry-pick in education. If you want to start a school to prove that penguin-based learning is the future, that penguin meditation and penguin-themed classrooms are superior to the stuffy, traditional, obsolete schools we have now, you can. It’s simple. Find a way to only accept no-structure and very low-structure learners. Then start your school. Do your penguin meditation, make sure there’s a basic structure for learning core academic skills, and you’re set. The results will be great, you can publish articles about the success of your method, if you’re lucky you’ll get some of that sweet sweet philanthropy money.

Honestly, I’m pessimistic. I think the key to improving education is understanding that the true bottleneck of mass public education is motivation. And more than motivation, it’s understanding that different students need different types of motivation. If we understood that, we could start to learn the best way to motivate students at scale. Too many motivation strategies rely on personalizing learning speed or appealing to student interests. And those can work in some cases, but they don’t scale well. We need motivation strategies that scale. That’s where we should be experimenting and pushing the boundaries in education. But that’s not where our current education system puts time and energy. Instead we’re captivated by clever-sounding solutions, always wanting to believe that technology is coming to save us. Technology isn’t coming to save us, for basic reasons of human psychology that are unlikely to change any time soon. So I’ll head back into my classroom, and keep grinding away trying to make progress with my students, ignoring the latest hype cycles and “innovation.” None of it addresses the core challenges to public education.