An excerpt from Practicing in the Age of AI: Essays on Medicine, Meaning, and Machines.

I recently bought a children’s book for my four-year-old granddaughter. Tucked beneath the copyright page was a sentence I had never seen before: “No part of this book may be used or reproduced in any manner for the purpose of training artificial intelligence technologies or systems.”

It wasn’t a medical text. It wasn’t a journal article or a clinical guideline. It was a Dr. Seuss book.

That single sentence lingered more than the story itself, and not in a good way.

The warning reflects a growing anxiety about large-scale AI scraping and intellectual property. While AI companies argue that training models on copyrighted material constitutes “fair use,” numerous lawsuits from authors, artists, and media organizations, such as The New York Times, allege copyright infringement. As of mid-2025, courts have produced mixed rulings, with some suggesting that training on legally acquired data is fair use, while the unauthorized use of pirated content is not.

Publishers have reason to worry about how creative work is absorbed, reused, and monetized by machines that do not ask permission. But encountering that language in a children’s book made me realize something deeper was at stake, something less legal and more educational. Because this isn’t just about protecting content. It’s about how learning happens.

What happens when machines are not allowed to read?

Medicine has always been a cumulative discipline. We learn by reading what came before us, by hearing stories retold on rounds, by watching mentors reason out loud, and by borrowing language until it becomes our own. Knowledge moves forward because it is shared, reshaped, and reused. That has always been true for humans. Now it is becoming true for machines.

That is what makes these new prohibitory statements different from traditional copyright notices. They are not just legal guidelines. They shape what machines are allowed to read, and, in doing so, what clinicians will one day learn with those machines. From an educational standpoint, this is not simply a matter of access. It is a matter of what knowledge survives and what is excluded.

The risk of a thinner learning environment

AI systems trained on incomplete or selectively restricted material may develop blind spots. If landmark reviews, reflective essays, or ethically complex case discussions are excluded, what remains may be thinner, flatter, and more procedural. The result is not just a less informed machine. It is a shallower learning partner.

An AI that has absorbed only checklists and billing-friendly language may be adequate for rote tasks, but it will not model uncertainty well. It will not reflect ethical concerns. It will not recognize the gray zones where medicine actually lives. Those are precisely the places where trainees and patients need the most help.

There is also an equity issue hiding in plain sight. Elite institutions and proprietary vendors may negotiate access to restricted materials, while freely available AI tools and the under-resourced learners who rely on them are left with less complete knowledge bases. We risk recreating an old hierarchy in a new form: not who has the best teachers, but whose AI has access to the best books.

Protection versus circulation

To be clear, the impulse behind these warnings is not malicious. Many authors worry, rightly, that their work, especially narrative and interpretive writing, will be condensed, stripped of attribution, or repurposed without consent. AI does not read with empathy. It does not understand intention. It patterns language without experiencing its weight.

But protecting medicine’s human core by walling off its stories may backfire. Medicine has never learned well from summaries alone. We learn through repetition, metaphor, rhythm, and surprise. We learn from stories we do not fully understand the first time we hear them. We learn by sitting with ambiguity before we know what to do with it.

This tension is no longer abstract. Popular writing platforms such as Medium now offer writers a choice: limit AI training on their work, or preserve broader reach and attribution through AI-driven discovery. The trade-off is subtle but consequential. Protecting content may increasingly mean limiting its visibility, not only to machines, but to the human audiences those machines now help guide.

How physicians actually learn

As a medical student, I was told something simple and lasting: If something is worth learning, you will hear it more than once. Medicine repeats what’s important on rounds, in lectures, in stories retold, until meaning finally sinks in. Which brings me back to Dr. Seuss.

He was never training a machine. He was training minds through rhyme, absurdity, misdirection, and joy. He was teaching pattern recognition before children knew what patterns were. He was preparing readers to tolerate nonsense long enough for meaning to emerge. Those are the same cognitive muscles physicians rely on every day.

• When a student hears a strange constellation of symptoms and does not yet know what to make of them
• When a resident senses something is wrong but cannot articulate why
• When an attending pauses, revises a story, and says, “Let’s think about this another way”

That kind of learning does not come from locked vaults. It comes from exposure.

A closing reflection

Ray Bradbury understood this instinct well. In Fahrenheit 451, the urge to suppress books never extinguished the hunger for stories; it merely drove them underground, memorized, shared piecemeal and carried forward by people who refused to stop learning.

Medicine has always depended on open circulation: of ideas, of failures, of stories told before they are fully understood. If AI is going to participate meaningfully in medical education (it already is) it needs access to that full tradition, not just the safest or most monetizable fragments. Warnings alone will not get us there. Neither will forcing AI to learn in the dark.

Thoughtful collaboration between publishers, educators, and AI developers might. Transparency might. Clear attribution might. However it happens, medicine cannot afford to muzzle its own learning process in the name of control. If we do, we will not just be teaching machines less medicine. We will be teaching ourselves less, too.

Arthur Lazarus is a former Doximity Fellow, a member of the editorial board of the American Association for Physician Leadership, and an adjunct professor of psychiatry at the Lewis Katz School of Medicine at Temple University in Philadelphia. He is the author of several books on narrative medicine and the fictional series Real Medicine, Unreal Stories. His latest book is Nobody Told Me There’d Be Days Like These: Hard Truths from Physicians—and What They Mean for Medical Practice.