From ChatGPT Back to Clay & Cuneiform: A Start at Rethinking Pedagogy for the Age of "AI"
The future of education depends on what we expect students to remember and do, not just what they can prompt Chatbots to generate. Thus forget banning AI. Instead teach students what it can’t do...
The future of education depends on what we expect students to remember and do, not just what they can prompt Chatbots to generate. Thus forget banning AI. Instead teach students what it can’t do for them. AI is simply the latest abstraction layer improving our information technology, a further step forward in the progression that started when we replaced clay with papyrus. Such abstractions are tremendously valuable, and productive. And so they become indispensable—until they break, and are no longer so…
From calculators to ChatGPT, every educational revolution has sparked fears about lost skills—justifiable fears—but simply yelling “stop!!” has never been productive. So it is with education now: “AI” is unstoppable. We must, instead, figure out how to roll with it.
But AI doesn’t spell the end of education. It does, however, demand a radical reboot. The old ways of certifying learning are dying. With them dies the assumption that essays or exams measure a form of understanding worth having. The question isn’t “how to stop cheating” but rather “how to teach what students need to remember and be able to do to manage and flourish in the future information ecology.”
One of the most productive ways to think about “AI” is the way Steven Sinovsky does: as an abstraction layer. Abstractions help us manage complexity. They are thus essential as complexity grows, and with growing complexity comes growing power and capability. The first key is to design them properly.
But there is also a second key: Abstraction layerss always leak. As “AI” reshapes how students write and how educators teach, we must confront the structural shortcomings of relying solely on upper-layer tools without understanding anything about leaky foundations. Only then do we have a chance of navigating this new cognitive landscape responsibly.
We have the very sharp Matthew Yglesias observing from the bleachers:
Matthew Yglesias: Who are “the groups”? <https://www.slowboring.com/p/who-are-the-groups>: ‘The whole education system needs to get out of a mindset where they conceptualize the use of AI as “cheating,” and try to devise ways to get people to not cheat. You can devise assignments that involve not using a computer—blue book exams, oral exams—but you also have to devise assignments where the intention and expectation is that students will be using AI. It’s similar to the way calculators have been integrated into math and science courses for a long time now. We don’t just act like calculators don’t exist and then try to stop people from “cheating” by using them…
And the sharp Johan Fourie reportign from inside the scrum:
Johan Fourie: AI ate my homework <https://www.ourlongwalk.com/p/ai-ate-my-homework>: ‘My students can now turn in passable essays having spent less than ten minutes writing them…. We should approach this with curiosity and ambition…. A university is… a disciplined, systematic and public commitment to scientific inquiry… [that] binds professors and undergrads… in the pursuit and transmission of knowledge… not… as a method: rational, testable and endlessly open to revision…. Can LLM help us teach that method?….
This year… [my] average attendance hovered around 20[%]…. I enjoyed the lively discussions with the committed few…. But is the counterfactual not better: students having access to an AI-Johan… endlessly patient, never dismissive, able to respond to any question, and happy to converse in the language of their choosing?… Student essays? They won’t last another year or two. Traditional assessments? Joshua Gans recently posed… [the] question: should universities assess at all? Should we not simply prepare students as best we can in the scientific method?… Spend more time providing one-to-one guidance, not less….
Progress happens… through conversation, challenge and collaborative interrogation of what is known and what remains uncertain…. Human judgement and mentorship… will determine which paths are worth following, which answers withstand scrutiny, and which questions are worth asking in the first place…. The greatest contribution a university can make is not to certify what students know, but to guide them towards a framework for how to think…
The lesson that I am starting to take away from this is that I need to think much more profoundly about learning outcomes that I have been throughout my career. I need to start approaching every class as I designed it with these questions in mind:
What do I want my students to remember five years from now about this class?
What do I want my students to know how to do after they have taken this class?
Let me drop back into the syllabus for the course I just finished teaching: Econ 113: American Economic History.
Its notional syllabus was one of, week by week, addressing these questions in order:
How is American economic history exceptional in ways that make studying American economic history not a waste of our time?
How is American economic history exceptional with respect the surrounding historical and comparative context of other time and other eras?
What are the pre-1860 causes of the United States’s now being a relatively and absolutely prosperous largely English-speaking country of 340 million people?
Why was the United States a country that both was late to abolishing slavery and managed to run the most productive and surplus-extracting slave economy ever?
How did the United States transform itself from a resource-abundance frontier-settler economy to the one that most successfully explored scientific, technological, industrial, and educational productive frontiers?
Why did the United States master the technologies of the post-1870 Applied-Science Era more effectively and thoroughly than any other economy?
Who gained and who lost from America’s becoming and remaining the world’s most immigrant-friendly society?
Why and how did patriarchy erode in America?
Why did the United States master the technologies of the post-1910 Mass Production Era more effectively and thoroughly than any other economy?
Why and how did the United States break its earlier political-economy mold—that of classical liberalism, or, rather, pseudo-classical semi-liberalism—and become the social-democratic New Deal Order society it was in the generation after World War II?
Why did the United States master the technologies of the post-1970 Information Era more effectively and thoroughly than any other economy?
Why and how did the United States, a land of opportunity and frontiers—first land, resource, and settlement; later scientific, technological, industrial, and educational—nevertheless suffer through not just one but two separate Gilded Ages of extraordinarily high inequality?
Why did the experience of the Great Depression of the 1930s not do a better job of innoculating the United States against a near-repetition of that disaster in the years after 2007?
Will the United States succeed in mastering the technologies of the future Attention Info-Bio Tech Era more effectively and thoroughly than any other economy?
Is the age of growth-promoting at least semi-rational governance that began in northwest Europe starting in the 1500s at its end?
What does the Fermi Paradox have to tell us about the past, present, and future of human history?
Should I—if I do teach this class again, probably in the spring of 2027—explicitly run each week by saying:
What is our question?
How would you go about discovering the best answer to it, at least to your satisfaction?
How would you then persuade somebody else that the answer you have come up with is the most likely answer?
Could that serve as a scaffolding for education in the age of AI?
I think it could. By focusing on “how would you discover?” and then “how would you persuade?”, we could avoid all of the traps involved in the overuse of LLMs.
And I also ran across Steven Sinovsky, with the cognitive bull case:
Stephen Sinovsky: 232. From Typewriters to Transformers: AI is Just the Next Tools Abstraction <https://hardcoresoftware.learningbyshipping.com/p/232-from-typewriters-to-transformers>: ‘AI is a tool-driven revolution… [with] the typical reaction to new tools—fear, skepticism, even rejection…. My freshman year (1983)…. Faculty were worried: if students didn’t handwrite first drafts, would they learn to write at all?… We ran an experiment…. [But] Macintosh launched. Apple pushed them onto campuses. What the faculty hoped to study was rendered moot…. The tool leapfrogged the debate…. Word processors offloaded spell check, formatting, and editing—freeing us to focus on content….
I was a computer science major… among the first CS majors who didn’t take physics or EE—and some argued we’d never truly understand computers. They were wrong. That too was an abstraction.
AI is the next abstraction layer…. Lke all previous… it’s criticized… [for] loss of fundamentals…. That’s true. But also true: they can do far more than previous generations. Abstraction is about not needing the old tricks. No one misses manually hyphenating or footnoting on a typewriter. [And for] lack of understanding…. That’s a weak argument. When a carpenter uses a nail gun, do we say they no longer understand roofing?… People say students will get lazy, not “really” understand, or miss what “matters.” The same was said about word processors. And Macs. And dropping EE courses….
It is not surprise that we're seeing so much writing about concerns—writers are the ones who are directly challenged. Just as electrical engineers were challenged by software abstracting out hardware…. Some will say… that AI needs more scrutiny sooner and that we should slow down before we understand. The challenge is the future doesn’t just wait around…. It arrives with new tools in hand…. AI is here. It’s already happening.…
Sinovsky is, mostly right. He is 100% right that the future is now. He is 80% right on the bull case, I think.
There is also a bear case. It applies not just to modern “AI”, but to all cognitive strategies that rest too heavily on the usefulness of abstraction layers. However, I cannot find a good-enough encapsulation of it this morning.
So this is how I would make it:
In our pursuit of understanding everys single complex system—economies, computation procedures, or sociotechnical ecologies—we msut rely on abstraction layers. There is no other way to even half-manage cognitive and organizational complexity. Yet there lies a treacherous truth beneath this methodological elegance of sweeping all lower-layer details under the rug. It is thus: all abstraction layers leak.
The implications of this leakage are profound, and the failure to recognize it can lead to catastrophic misunderstanding and mismanagement. So we must equip ourselves with a rigorous grasp of the fundamental dynamics that abstractions are designed to mask. No, we should not have to start our w ite-collar workdays by spending two hours mixing and smoothing our clay for our cuneiform styluses. We do not have to go overboard. But we do have to know enough about lower-level layers to provide us with cognitive insurance.
The phrase "all abstraction layers leak" originates from the realm of computer science, particularly from Joel Spolsky’s canonical essay on the Law of Leaky Abstractions. It states that while abstractions simplify, they cannot fully encapsulate the complexity of the underlying systems they represent. In economics, we see thiss whenever we take monetary aggregates, national income accounts, and stylized growth models as the reality. But the true underlying reality is made up of the patterns in human behavior and institutional structure. Abstractions will fail to capture the full richness of social reciprocity and institutional nuance. Thus, for example, Gillian Tett's insights on the anthropological underpinnings of economic exchange are extremely valuable. But someone who takes the abstraction letters of the economics discipline as the true reality is incapable of comprehending them.
The leaking of abstraction layers is not a flaw to be corrected; it is a feature of complex systems. But it is a feature that imposes an obligation: those who wish to intervene in, manage, or even merely survive such systems must dive beneath the abstraction. They must learn not just what the system does, but how and why it does so. That is the only way to avoid what Alfred North Whitehead called the fallacy of misplaced concreteness. This fallacy becomes fatal when abstraction layers leak in systems where stakes are high.
To learn the fundamentals beneath any abstraction is not simply to "go deeper"; it is to inoculate oneself against the myopia of model-worship and the complacency of surface reasoning. It is, in a very real sense, the only way to grasp reality in its full, leaky, splendor. And as we slouch towards whatever utopia or dystopia awaits, it is the only intellectual ethic that offers a chance of understanding the path.
References:
Fourie, Johan. 2025. “AI Ate My Homework}. Our Long Walk.
Sinovsky, Stephen. 2025. 232. “From Typewriters to Transformers: AI is Just the Next Tools Abstraction”. Hardcore Software.
Spolsky, Joel. 2002. "The Law of Leaky Abstractions".
Tett, Gillian. 2021. Anthro-Vision: A New Way to See in Business and Life.
Whitehead, Alfred North. 1925. Science and the Modern World.
Yglesias, Matthew. 2025. “Who Are the Groups?” Slow Boring.
There is a lot of question-begging going in all this analogizing to the technological advances of the past. Oh yes, to be sure *if* AI functioned as a nail gun does to a hammer, as a compiler does to an assembler, or as a calculator does to pencil and paper, then indeed the problem would be to master the new, higher level of abstraction. But in actuality, in order to integrate AI "similar to the way calculators have been integrated into math and science", we mainly have to account for the fact that the calculator gets a completely wrong answer about 15% of the time. The compensating skill to be developed would not be abstraction but rather estimation, being able to check whether the calculator's answer is in the right ballpark. But this is an advanced skill!
Once your students understand this point, will they really be so enthusiastic about AI? Will they be happy to leave the interesting bits to automation and take on the dull drudgery of copy-editing and correction themselves? It seems unlikely. I would remind you that you yourself, with your remarkably open and flexible mind, have publicly made several attempts to coax useful output from AI without any notable success so far.
In terms of teaching, in fields where you have knowledge to transfer, one should put oneself imaginatively in the position of the student, understand where they are coming from, and then figure out how to analytically continue them from their state to the state of knowledge you'd like them to be in. This cannot be done by teaching axiomatically the first time around.
From Barenblatt's book on scaling:
"Of special importance is the following fact: the construction of models, like any genuine art, cannot be taught by reading books and/or journal articles (I assume that there could be exceptions, but they are not known to me). The reason is that in articles and especially in books the 'scaffolding' is removed, and the presentation of results is shown not in the way that they were actually obtained but in a different, perhaps more elegant way. Therefore it is very difficult, if not impossible, to understand the real 'strings of the work: how the author really came to certain results and how to learn to obtain results on your OWn."