Education Was Already Broken. AI Simply Revealed It.
Ezra Klein is asking the wrong questions about AI and education.
In his latest podcast, Ezra Klein muses on education and AI with Rebecca Winthrop, the director of the Center for Universal Education at the Brookings Institution and coauthor of The Disengaged Teen: Helping Kids Learn Better, Feel Better, and Live Better.
Education is a crucial topic, but Ezra approaches it from the wrong angle. From the get-go, he asks all the wrong questions:
I have a 3- and a 6-year-old. And one of the ways that my uncertainty about our A.I.-inflected future manifests is this deep uncertainty about how they should be educated. What are they going to need to know?
I don’t know what the economy or society is going to want from them in 16 or 20 years. And if I don’t know what it’s going to want from them, what it’s going to reward in them, how do I know how they should be educated? How do I know if the education I am creating for them is doing a good job? How do I know if I’m failing them? How do you prepare for the unpredictable?
As a parent of a 6- and 8-year-old, I sympathize. I also ask myself “How do I know if I’m failing them?” on a daily basis—a lot of the work I’m sharing here is introspection on how to do better for my kids.
But Ezra’s framing misses a deeper, more fundamental point about what learning actually means. Maybe it’s just how he frames the interview to tee up the conversation, but to me he comes across as an aspiring top-down regulator.
The podcast interview title is “We Have to Really Rethink the Purpose of Education,” but what comes out of it is not a new view of education, but a rehashing of the current one, with new guardrails around the much-feared AI.
The conversation highlights two interconnected issues. The first is a view of education as a process of filling students’ minds with knowledge and skills. From that view derives the second issue, the use of essays and other assignments as a measurement for the progress in education.
The bucket theory of the mind
Questions such as “what are my kids going to need to know?” or “how do I educate my kids to satisfy what the economy will want from them in 20 years?” implicitly assume that education is a simple matter of filling a kid’s brain with facts and skills.
This idea is based on what Karl Popper dubbed the bucket theory of the mind—“that we obtain knowledge just by opening our eyes and letting the sense-given or god-given ‘data’ stream into a brain that will digest them1.” And it’s plain wrong.
Our minds are not containers to be filled with facts, or booklets where we collect achievement stickers for skills we’ve mastered.
Learning is more akin to building a network of connected ideas. And the only way that learning happens is from within, with conjecture and refutation.
We learn when we face a problem, guess at a solution, and verify it through criticism. In an ideal world, schools would cater for the problems each student is interested in and help them create the knowledge to solve them—along the way developing those crucial skills of knowing how to learn, knowing how to criticize ideas, knowing how to detect and try to correct errors.
The writer Louis L’Amour had it right when he wrote in his memoir Education of a Wandering Man:
The idea of education has been so tied to schools, universities, and professors that many assume there is no other way, but education is available to anyone within reach of a library, a post office, or even a newsstand.
And:
Education should provide the tools for a widening and deepening of life, for increased appreciation of all one sees or experiences. It should equip a person to live life well, to understand what is happening about him, for to live life well one must live with awareness.
And again:
Acquiring an education has many aspects, of which school is only one, and the present approach is, I believe, the wrong one. Without claiming to have all the answers, I can only express my feeling that our methods of instruction do much to hamper a child in learning. Our approach is pedestrian. We teach a child to creep when he should be running; education becomes a task rather than excitement. Yet each of us can remember one or two teachers who made learning an adventure, which it surely is.
Learning is an adventure, a choose your own adventure.
Winthrop gives a hint of this with the story of Kia, a teen who was “totally disengaged, doomscrolling in middle school” until she joined a self-create class. Kia developed an escape room themed on the Lincoln and JFK assassinations and learned history and science through it.
Alas, building escape rooms was no true escape from the system. It was only a means to get Kia re-engaged with the rest of the curriculum.
That’s disappointing but not surprising. When so-called educators approach schooling as filling a bucket, then the ultimate goal is always to make sure they can tick the box for every item that a committee has decided belongs in the bucket.
Generative AI and homework
The bucket theory of the mind leads directly to the stance that using generative AI to write school essays and other homework is cheating.
Let’s leave aside the contradiction of adults simultaneously worrying that AI will be crucial in the future economy while at the same time refusing to teach kids how to use it. Let’s focus instead on whether using it is cheating.
Considering using AI to generate an essay as cheating implies the goal of the homework is to verify the student’s capacity to write an essay on their own. I guess the idea behind that is that writing an essay teaches you to think. I stand behind that, reading and writing are the best cognitive calisthenics. But anyone who has gone through school and has been forced to produce essays and other assignments can understand the difference between learning to think by applying yourself to something you care about and writing an essay to show you know grammar and how to adhere to the introduction–thesis–antithesis–conclusion structure.
Someone decided that assigning an essay on a topic is a good way to teach kids to think. But it’s a barely adequate proxy—and one that works only in the bucket-theory world.
Think back to Kia and her escape rooms. When we understand that learning is problem solving, then whether or not someone uses AI to solve the problem has little to do with the outcome.
Give any kid a problem they care about and they’ll genuinely learn something to solve it, with or without AI.
So, yes, using LLMs to write an essay is cheating, but it’s cheating in a game that should not be played in the first place.
Back at university, I had to give more than one exam on coding by writing the code on a piece of paper. That exercise did little to prepare me for my job as a software developer where the more tools I can use to help me write good code, the better.
If education should prepare kids for the future, then why is there so much emphasis on things with little correspondence in the real world?
Where to go from here
I don’t pretend to be an education expert, and like L’Amour I don’t claim to have all the answers.
What concerns me is that influential voices such as Ezra Klein at the New York Times approach the problem with a top-down regulatory mindset that, despite all the refined vocabulary and progressive ideas, doesn’t seem to have moved on from how education was thought of in the early 1900s: a means of preparing the masses to be productive in the economy. So much for really rethinking the purpose of education.
Ezra and I both worry about preparing our children for the unpredictable. But I don’t think the answer lies in regulating AI or better top-down design of the curricula.
What if we discarded the bucket theory of the mind and built schools that help children develop their own lattices of knowledge, taking the driver’s seat on their own education?
Maybe the way to prepare kids—and adults—for the unpredictable does not rest on a committee of experts. Maybe the skills needed are more basic: how to learn, how to spot errors, how to solicit and parse criticism, how to collaborate with other humans, how to leverage existing technology and get up to speed with new ones.
There are a myriad of ways to acquire those skills. I believe the way forward lies in accounting for this, and that needs a flexible infrastructure of decentralized, personalized learning.
Maybe the way not to fail our kids is to stop trying to build an education for them—and start building it with them.
1 - From the 1989 lecture Towards an Evolutionary Theory of Knowledge, available in the All Life Is Problem Solving collection.
Shared a similar perspective watching this. Part of the issue is that there's only so much to reimagine when the scope is as narrow as repackaging public school in its current bureaucratic and one-size-fits-all form. This is why the best path forward is to create more opportunity for choice across the board: 1. more diverse models of schools and innovative education opportunities in communities, 2. more choice for families to select and have access to and 3. more personalized learning at the student level which is enhanced by 1. and 2. More choice = more agency in the learning process, and ideally a more holistic understanding of how we view education as a society.
We're seeing a growing movement around alternative education, building new systems outside of the legacy system, in pursuit of this desire for choice and more curated education experiences.