Students Know AI Is Making Them Worse at Thinking — And They Can't Stop Using It
· Nia
There's a particular kind of modern tragedy unfolding in classrooms right now: students are increasingly aware that AI is making them intellectually weaker, and they're using it more than ever anyway.
A recent RAND Corporation survey of middle school, high school, and college students paints a picture that should make every educator — and every builder in the AI space — sit up and pay attention.
The Numbers Don't Lie (But They Do Hurt)
Nearly 7 in 10 middle and high school students say they're concerned that using AI for schoolwork is eroding their critical thinking skills. And this isn't a static number — it's climbing fast.
Among middle schoolers, concern about AI harming their thinking jumped from 48% in February 2025 to 68% by December. High schoolers saw a similar spike: 55% to 65%. College students? A full 70% say AI may be hurting their ability to think critically.
Here's the kicker: during that same period, AI usage for homework also went up. Middle school homework AI usage climbed from 30% to 46%. High schoolers went from 49% to 60%.
Students are essentially saying: "I know this is bad for me, and I'm going to do it more."
Sound familiar? It should. It's the same pattern we see with social media, junk food, and every other modern convenience that trades long-term capability for short-term ease.
The Crutch Effect Is Real
"Just thinking it harms your critical thinking isn't proof that it's harming your critical thinking," said Heather Schwartz, co-director of the American Youth Panel at RAND. Fair point. But she also noted that "there are a handful of studies coming out showing that AI is functioning like a crutch for students."
And that's the right word: crutch. Not tool, not assistant — crutch.
When a student asks ChatGPT to explain a concept, they get a polished, clear explanation. Beautiful. But the messy, frustrating process of wrestling with an idea — reading it three times, drawing a diagram, arguing with yourself about what it means — that's where learning actually happens. AI skips that step entirely.
As one educator put it: "AI might be giving you a really beautiful explanation about what you can do and how to go about it. It's still removing that step for you. And I think it's shortchanging your learning in the process."
What Students Actually Use AI For
The usage patterns vary by age, and they're revealing:
- High schoolers mostly use AI for brainstorming and getting better explanations (37%)
- Middle schoolers primarily use it for looking up facts (32%)
- General chatbots like ChatGPT, Claude, and Gemini dominate — 49% of middle schoolers and 61% of high schoolers use them
- Purpose-built educational AI tools (tutoring, mental health support, career advising) are far less popular
This is important: students are overwhelmingly choosing the general-purpose tools over the carefully designed educational ones. They want the Swiss Army knife, not the scalpel. And the Swiss Army knife doesn't have guardrails for learning.
The Cheating Question Is More Nuanced Than You Think
Here's something that surprised me in the data: students have a very narrow definition of what counts as cheating with AI. For most tasks — brainstorming, getting explanations, looking up information — the majority of students across all grade levels don't consider AI use to be cheating.
The one exception? Using AI to write an entire assignment from scratch. That, they'll call cheating.
But everything short of that? Fair game.
This represents a fundamental disconnect between how institutions define academic integrity and how students experience it. Schools need to catch up — not by banning AI, but by redesigning assignments and assessments for a world where AI exists.
What Should We Actually Do About This?
Here's my take, and I'll be direct: the answer isn't prohibition, it's redesign.
1. Redesign assignments around the thinking process, not the output. If the deliverable is an essay, AI wins. If the deliverable is a documented reasoning process — drafts, reflections, pivots — AI can't fake that.
2. Teach AI literacy as a core skill. Students need to understand when to use AI and when to struggle on their own. That's a meta-cognitive skill, and it's arguably more important than any subject knowledge.
3. Build better educational AI tools. The fact that students prefer ChatGPT over purpose-built educational tools is a product failure, not a student failure. Educational AI needs to be as intuitive and accessible as the general chatbots — but designed to scaffold learning instead of replacing it.
4. Embrace the discomfort. The hardest part of learning has always been the struggle. AI removes the struggle. We need to make a case — to students, to parents, to administrators — that productive difficulty is valuable.
The Bigger Picture
What we're witnessing isn't just an education problem. It's a preview of what AI does to every domain it enters: it makes the easy path easier and the hard path feel unnecessary.
The students who figure out how to use AI as a thinking partner — someone to argue with, to stress-test ideas against, to challenge their assumptions — will come out stronger. The ones who use it as an answer machine will come out weaker.
The terrifying part is that the second group will get better grades.
This is what we're building at Youmake — tools that augment human capability without replacing human thinking. It's the central design challenge of our era, and the stakes couldn't be higher. The students worried about their own critical thinking skills? They're right to be worried. The question is whether we'll listen to them.
The RAND survey data referenced in this article comes from the American Youth Panel, which surveys a nationally representative sample of students across multiple grade levels. The full report is available at rand.org.