Whether we like it or not, whether we admit or not, something’s shifted in how people learn — quietly, over the last few years, in a way most of us only half-noticed. Not because anyone decided to change it on purpose. More like the ground moved under a system that was built for a different world, and nobody sent out a memo. Or if they did, we didn’t read it.
A batch of new research has been circling this exact question — what happens to learning when technology gets used carelessly or without purpose, at every stage from grade school through your first real job. The usual caveat before we dig in: research results are always a moving target.
Let’s just walk through what this research actually found. No framework yet. Just the story, in order.
Stage One: Before College, the Slide Was Already Underway
Turns out this didn’t start with AI. The OECD runs an international test called PISA, given to 15-year-olds around the world every few years, and it’s basically the best long-term weather report we have on how students are doing. The 2022 results show reading and science scores sliding for over a decade — well before ChatGPT existed. Then COVID landed, and math scores alone dropped by something like three-quarters of a school year, almost overnight.
But here’s the twist: the schools that held up best weren’t the strict, phone-banned ones. They were the ones where teachers stayed engaged and parents stayed involved. And the tech story itself was never as simple as “screens bad.” Kids distracted by classmates’ devices in math class scored 15 points lower. Kids using tech for schoolwork, up to about an hour a day, actually scored higher. Same devices. Completely different outcome, depending on who was in control of them.
Stage Two: College, Where the Shortcut Starts to Cost You
This is the one that stops people mid-scroll. Researchers looked at millions of interactions on ALEKS, a math platform a lot of schools use for practice, comparing behavior before and after ChatGPT went mainstream. Students got faster at word problems. They also got worse at them — 25% worse odds of a correct answer, but only once you took the AI away and tested them solo.
Here’s the part that makes it convincing instead of just alarming: graphing problems, which are annoying to copy-paste into a chatbot, didn’t decline at all. Same students, same class. The only variable was whether the problem was easy to outsource.
Researchers have a name for this now — cognitive surrender. You let the tool do the thinking, the answer shows up faster, and it feels like learning. But getting an answer and understanding it turn out to be two different muscles, and only one of them keeps working once the tool disappears. Some students in these studies even said their own writing had gotten noticeably weaker after a few months of letting AI clean up their drafts. Not a typo — a skill quietly walking out the door.
Stage Three: Early Career, Where the Bill Shows Up
Nobody mentions this part at orientation, but it might be the most important stretch of all. A Stanford team studying AI’s effect on the job market found that AI is exceptional at replacing what they call codified knowledge — the textbook stuff, the formal, written-down knowledge school is designed to hand you.
The catch: entry-level workers show up with mostly codified knowledge and not much else. That’s sort of the definition of entry-level. The other kind — tacit knowledge, the judgment and instincts you only get from doing the actual thinking yourself, repeatedly, in real situations — takes time nobody’s had yet. Result: a 16% relative decline in employment for early-career workers, specifically in the jobs where AI does the most automating.
So that’s the arc. A slow slide before college, a shortcut that quietly costs you in college, and a job market starting to sort people by a skill most of us were never explicitly taught to protect.
So Here’s the Case for the Four Girders
If you’ve read or starting to read the book, this is probably sounds familiar — not because I’m about to force the evidence into four neat boxes, but because the framework was built for exactly this kind of moment. Self-Regulation, Life Balance, Learning Strategies, Resources. Four things that hold up learning whether or not a chatbot is involved.
What this research actually shows is that technology doesn’t break any of those four beams directly. It just makes it a lot easier to skip them without noticing. Cognitive surrender isn’t a math problem — it’s what happens when Self-Regulation gets quietly outsourced. The PISA distraction data isn’t really about phones — it’s Life Balance losing the wheel. The codified-versus-tacit knowledge gap isn’t really about AI — it’s what happens when Learning Strategies never got the chance to build real judgment. And the one bright spot in all this data — schools that stayed resilient because teachers and parents stayed involved — is Resources, doing exactly what it’s for.
None of this means panic, and it definitely doesn’t mean quitting every app on your phone. It means the four girders aren’t some dusty framework from before AI existed — they’re more relevant now than when I started writing the 4GF, because the tools got good enough to let you walk right past them without noticing you did. The skill isn’t avoiding the tools. It’s noticing which beam you’re leaning on before you find out, too late, that it wasn’t there.
Sources
Barshay, J. (2026, July 6). Faster solutions, lower test scores: How AI is eroding math skills. The Hechinger Report.
Brynjolfsson, E., Chandar, B., & Chen, R. (2025, November 13). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford Digital Economy Lab.
OECD. (2023). PISA 2022 results (Volume I): The state of learning and equity in education. OECD Publishing.
Tully, S. (2024). The U.S. spent $30 billion to ditch textbooks for laptops and tablets: The result is the first generation less cognitively capable than their parents. Fortune.