> ## Content Index
> Fetch the complete content index at: https://ai-public-health.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Theory That Waited: Why Old Ideas Fit the AI Era
- URL: https://ai-public-health.com/the-theory-that-waited-why-old-ideas-fit-the-ai-era/
- Published: 2026-11-16T14:00:00.000Z
- Updated: 2026-11-16T13:59:59.000Z
- Author: Min Wu
- Tags: AI in Education, Metacognition, Higher Education, Public health, AI Literacy

**Case of the Week,** Min Wu, PhD · ai-public-health.com

*While I was writing my textbook, the loudest question around AI in higher education was cheating. Students were using AI to write their homework, and nobody seemed to know what to do about it. University leaders issued policies, researchers debated detection tools, and faculty traded stories in hallways. I did not hear much clarity from any of them, and I was not sure I had any myself.*

*What I kept noticing was a confusion underneath the debate. People were treating fluent writing as if it were proof of thinking. For most of my career, it more or less was. A clean, well-organized essay usually meant a student had worked through the ideas. AI broke that link. Fluency became cheap, and the arguments about cheating were really arguments about what the fluency still meant.*

*I came to a position slowly: accept the fluency as it is, and grade the thinking. Then I needed a name for homework built that way, and the name came from my own field.*

**Key Takeaways**

- **Meta-AI homework** rests on metacognition, a theory from the 1970s: grade how students think while using AI, not how fluent the output is.
- Old theories often wait for their problem. When a new technology confuses everyone, the human side of it has usually been studied before.
- Before writing new rules for AI, ask which old question it reopens, and look for who already answered it.

### **Data About Data, Thinking About Thinking**

As an informatics professor, I work with metadata every day. Metadata is data about data: not the record itself, but what tells you how to read it. In the older electronic medical record, that meant a short list: the problem, the time, the place. In my recent work on AI-powered personal health records, I proposed a richer set of eight kinds of metadata, such as when something happened, in what context, and in what pattern of behavior, because an AI agent needs that context to act safely on a person's data ([doi:10.2139/ssrn.7518960](http://dx.doi.org/10.2139/ssrn.7518960?ref=ai-public-health.com)). Metadata is how we index, find, and trust information.

Thinking about the homework problem, I saw the same structure one level up. If metadata is data about data, then what I wanted to grade was thinking about thinking: whether a student knew what they understood, noticed where the AI might mislead them, and could say how their own judgment changed the output. Trained as a physicist, I found the symmetry irresistible. I named the approach meta-AI homework before I had looked for any theory behind it.

Then I went looking, and I found that the theory already existed. It had been waiting for about fifty years.

I honestly do not remember whether I found it through a literature search or because an AI tool suggested it. I like both possibilities. Either way, the match was immediate.

### **A Theory Built for Children's Memory**

The idea is called metacognition. The developmental psychologist John Flavell named it in the 1970s while studying how children come to monitor their own memory and understanding: how a child learns to know whether they really know something. In 1979 he laid out the framework: what we know about our own minds, the feeling of knowing or not knowing, and the strategies we use to check ([doi:10.1037/0003-066X.34.10.906](https://doi.org/10.1037/0003-066X.34.10.906?ref=ai-public-health.com)). Researchers who built on his work later described it as three moves: planning, monitoring, and evaluating.

Flavell was not thinking about machines that write. He was thinking about a seven-year-old deciding whether they had studied a list of words long enough. Yet those three moves map almost exactly onto what a student needs when working with AI. Before the tool, what do I already know, and where might it mislead me? During, is this output consistent with what I know? After, where did my judgment change the result, and where did I let the result change my judgment? In "Homework Is the Rehearsal," I described how those three phases became the structure of meta-AI homework, which I set out in full in Chapter 10 of my textbook ([doi:10.1007/978-3-032-15872-7\_10](https://doi.org/10.1007/978-3-032-15872-7%5F10?ref=ai-public-health.com)). What I did not say there is how little of it was new.

Flavell lived to see metacognition become one of the most cited ideas in educational psychology. He died in March 2025, at 96, a little more than two years after generative AI reached the public. He saw his idea spread through classrooms. He did not have time to see it become one of the most useful tools we have for the hardest teaching problem AI has created.

### **Some Ideas Wait Even Longer**

Flavell at least saw his theory adopted. Others never did.

The Soviet psychologist Lev Vygotsky proposed that people learn best just beyond what they can do alone, supported by someone who knows more and who gradually steps back. He died in 1934, at 37\. His work was suppressed in his own country for years and reached English readers only decades after his death; his best-known English collection, *Mind in Society*, appeared in 1978\. Today, people designing AI tutors argue in his terms: should the AI be the one who knows more, and when should its support be taken away? Vygotsky is not part of my framework, and I did not cite him in my textbook. He is simply the starkest case of the pattern. Some ideas outlive their authors and wait for a problem that did not exist when they were written.

### **Why Old Theories Fit New Tools**

This is not a coincidence. AI is new, but the human side of using it is not. People have always needed to judge whether they understand something, decide when to trust an outside source, and notice when someone else's fluency is standing in for their own reasoning. The technology changes the speed and scale of those problems. It does not change what they are.

That is why I am wary of treating every AI question as if it needs a theory invented this year. The cheating debate felt new and urgent, and many of the responses were new and urgent too: detection software, honor pledges, bans. Most of them were aimed at the output. The older question, whether the student was thinking, had a fifty-year-old answer sitting in the literature, waiting to be applied.

### **What I Am Still Unsure About**

A theory that fits well can also fit too easily. Metacognition was built on children's memory tasks in laboratories, not on adults working with tools that are fluent, confident, and sometimes wrong. I am applying it in a setting its author never tested, and the mapping still needs evidence from real classrooms. I also cannot rule out that I am drawn to it partly because the name was so satisfying. The symmetry came first, and the theory second.

Still, I would rather build on an idea that has survived fifty years of scrutiny and test whether it stretches than invent a new one that has survived none.

If you teach, train, or lead people through the AI transition, here is one question to ask before writing the next policy: what old problem is this new tool reopening, and who already studied it?

***Waiting Makes the Foundation***

---

*What old idea from your own field turned out to fit the AI era? I would love to hear. Reply or leave a comment.*

Want to read the earlier essays in one place? They’re collected in *AI and Public Health, Volume 1*, in [ebook](https://books2read.com/u/m2pdWO?ref=ai-public-health.com) and [paperback](https://www.amazon.com/dp/B0HLXLTGHL?ref=ai-public-health.com).

*As a member of the Springer Nature Author Affiliate Program, I may earn a small commission from purchases made through this affiliated link to my book. Support the author by checking out* *my textbook\*\*,* Artificial Intelligence in Public Health: [https://tidd.ly/4mH9389](https://tidd.ly/4mH9389?ref=ai-public-health.com)