Gather 'round, for the scrolls of machine learning grow heavier by the moon, and most are filled with the same weary boast: bigger models, longer benchmarks, another half-point on a leaderboard nobody outside the guild can read. But every so often a paper arrives not to flatter the machines, but to ask an awkward question about us. Mariana Lenharo's recent dispatch in Nature is one such tale, and it carries an uncomfortable moral: the early evidence suggests AI may be sanding down the very skills it was summoned to assist.

The Endoscopist Who Forgot the Hunt

Consider the parable of the polyp-hunters. In a study that sent a small tremor through the medical guilds, researchers tracked endoscopists - the brave souls who steer cameras through the human colon in search of trouble - before and after they began using AI detection tools. The AI was good. Genuinely helpful. It caught things. And then, when the machine was switched off and the doctors were left alone with their own eyes once more, their detection rates had dropped (Budzyń et al., The Lancet Gastroenterology & Hepatology, 2025).

Read that again. Not "stayed the same." Dropped. The clinicians had, in the span of months, grown a little worse at the one thing they'd trained years to do well. It's the cognitive equivalent of letting your knight ride into battle so often with a magic shield that he forgets how to raise his own arm.

This has a name older than ChatGPT: automation complacency. When a reliable helper handles the watch, the human watchman drifts off to sleep. Pilots have known this dragon for decades - lean too hard on autopilot, and manual flying skills wither in the dark.

The Cognitive Debt Collector

The bards of MIT's Media Lab told a parallel legend. They wired up volunteers writing essays - some with a large language model whispering in their ear, some with only a search engine, some with nothing but their own beleaguered brains - and watched the electrical storms of thought (Kosmyna et al., "Your Brain on ChatGPT," arXiv:2506.08872, 2025).

The findings, while early and small, were the stuff of cautionary fable. The LLM-assisted scribes showed weaker neural engagement, remembered less of what they'd "written," and felt less ownership over their own words. The researchers coined a phrase that ought to be carved over every server farm: cognitive debt. You borrow ease today; the interest comes due in the skills you quietly stopped practicing.

It's the same bargain your sense of direction made with GPS years ago. Quick - without your phone, could you navigate your own city? Exactly. The map app didn't steal your wayfinding; you handed it over, one painless turn-by-turn at a time.

Before You Smash the Loom

Now, let us not become the village mob with torches. The same Nature piece is careful, and so should we be. None of this proves AI makes you dumber in some grand, irreversible sense. The studies are young, the samples modest, the long-term effects unmapped. A calculator didn't end mathematics; it freed mathematicians to chase bigger quarry. The question is not whether the tool helps - it plainly does - but whether we're using it to amplify skill or to replace the practice that built the skill in the first place.

The heroes of this saga aren't the doctors who used AI. They're the researchers who thought to measure what happens when the AI walks away. That's the trial that matters: not "can the machine do it," but "can the human still do it afterward."

The Moral of the Scroll

So here is the wisdom, traveler, plain as an innkeeper's bill. AI is a magnificent assistant and a treacherous crutch, and the difference lives entirely in how you lean on it. Let it spot the polyp, then check the work yourself. Let it draft the essay, then rewrite it until the thoughts are yours. Use it to skip the drudgery, not the learning - because the muscle you never flex is the muscle you eventually lose.

The machines are not coming for your skills. But they will happily hold them for you, indefinitely, until you forget you ever asked for them back.

References

• Lenharo, M. "Is AI ruining our skills? Early results are in - and they're not good." Nature (2026). DOI: 10.1038/d41586-026-01947-1 | PMID: 42315587
• Budzyń, K. et al. "Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy." The Lancet Gastroenterology & Hepatology (2025).
• Kosmyna, N. et al. "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Tasks." arXiv:2506.08872 (2025).
• Parasuraman, R. & Manzey, D. H. "Complacency and Bias in Human Use of Automation." Human Factors (2010). DOI: 10.1177/0018720810376055

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.