A friend of mine is a high school teacher in the NYC public school system and it’s been weird. Even with in-school teaching, their system is part-remote – they have stationary teachers who are teaching 1 class period and supervising the classroom during other periods, where a teacher teaches the class over zoom. The goal is to limit movements + mixing.

This obviously sucks for all parties and makes school more like jail. But like all disasters it can be a forcing function for change. Remote teaching, if it survives the lockdown, enables something that was never possible before – matching students to the subject + teacher they need right now, rather than whoever is available in the building.

In particular, this enables a hybrid of self-paced and classroom-driven learning that is impossible with the classic on-site, long-semester, lecture-driven model. Read on.

Self pacing and efficiency

From a state’s perspective, school is a good investment because it allows you to have an advanced economy. But that means you have to decide how to split resources between kids who are on their way to PhDs, vs STEM bachelors, vs other. America somehow isn’t rich enough to give all three tracks opportunities, especially given the fast rate of inflation in education. India definitely isn’t, and hence has a few really good schools catering to the top of the scoring distribution.

From a student’s perspective, school is a good investment because society treats them as criminals without it.

Also from a student’s perspective, school is jail. Even if you’re really good at a part of school, school is full of opportunities for shame and punishment: What if you’re a music genius getting held back for being bad at math? What if you’re killing it on tests and getting graded down for bad attendance? One of the lessons of 2020 is that half-baked policies backed by state coercive power can escalate to fascism in an eyeblink.

Fix all this by making school self-paced. If you assume every kid is 50% faster at learning one subject than all their others, they should be college level in that subject by 9th grade, or career-ready by the end of high school. The fact that we’re not allowing kids to do this is in essence robbing them of 4 years of life. It’s also robbing them of 4 years of youth during which they could start their career with more flexible brains.

From the state’s perspective, the argument is almost the same – if you can reduce the cost to train skilled professions, and if that leads to increased yield, and if that leads to increased GDP growth, 💥.

My friends who work in education have always objected to the self-pacing model because it requires too many teachers on site or splits their attention too many ways. It’s hard work to prepare for a lesson module, and you can’t do it times ten students. But with video wall teaching, ‘on site’ is no longer a constraint.

Granular learning modules

Instead of 3 to 6 month semesters where everyone learns in lockstep, students can have 1-4 week granular learning modules which they exit by mastering the material (on a test) rather than ‘aging out’ of it. If someone only needs 1 week, they move to the next one right away.

This has a couple of advantages: first it wastes less of everyone’s time. You exit the system either early, or having learned more, or both.

Second it enables what’s called mastery learning, which means that you keep at something until you get it. Although this in theory means that some students will move through the system more slowly, that in its own way is more efficient: if the prequisite model is designed well, people won’t advance to classes they’re not prepared for, fail, and require extra help.

Self-pacing also creates simpler and better incentives for students – the reward for success is getting to advance, i.e. GTFOing. (And who doesn’t want to GTFO). And the punishment isn’t a bad grade, it’s getting left behind.

It also changes the nature of grading – instead of getting scored on how well you did at a small list of broad subjects, you get a record of how many narrow subjects you mastered.

Skipping grades is regarded as bad because it separates kids from their social peers. If school is mixed-age by default, that’s no longer a problem. If most students end up in a mix of above- and below-grade classes, age matching matters even less.

Better metrics

Granular modules also enable granular data collection.

We can answer question like:

  • who’s best at teaching a given narrow topic
  • where does this student need help or extra coaching, and is there another student who just destroyed that subject and can pitch in
  • what is this teacher particularly bad at teaching, and who can coach them

Curriculums are already, in the old system, standardized enough that this information is easy to categorize. It’s just hard to collect.

Using tests to ‘graduate’ learners to the next granular module creates a much faster cadence of standardized testing – multiple times per month instead of per semester or per year.

I think more frequent, smaller testing is good. It’s silly and unfair for teachers to be in such strict control of grades, which partly assess students but also assess the teachers. Giving teachers less assessment power also makes students the customer – the interaction is now ‘I need your help passing this test’, not ‘I’m sucking up to you so I can GTFO’.

Our current metrics system is goodhart’s law in action. Test scores and grades are metrics but also targets. The incentives are to cheat or teach to the test, rather than to cultivate capable kids who are satisfied customers of education

Granular modules also give us compatibility scoring – is teacher X bad for student Y? But this is tricky and could expose bias (good) or reinforce it (bad), so maybe we should hold off.


Semester length is, I suspect, an area where there’s no experimentation happening right now. If kids are able to learn twice as fast, we wouldn’t know that. I think it makes sense to find out.

Granular modules also enable all kinds of other experiments. Multi-month lockstep semesters mean that any kind of experiment is (1) expensive and (2) potentially causes great harm. Shorter modules let us run more experiments and pick up the pieces more cheaply when they fail.

Our failure to innovate in education is sad and cruel. It also may be a civilization-ending catastrophe, but it’s too soon to tell. Like healthcare, school is a technologicaly product that should be as cheap as cell phones but instead is inflating faster than the basket.

Governments are grossly underinvesting in technology for independent learning. Napkin math, we pay more as a country for textbooks every year than the cost of building the video game HALO. Why not build HALO learning environments instead of textbooks? It’s bullshit.

Granular modules let us test all this and more. Zoom learning enables granular modules. This is technology we have today. What are we waiting for?