Student Question of the Week: Does AI Kill Student Thinking?

New Series: Questions from Future Educators

A few days ago, I had the opportunity to discuss AI with a group of college students preparing to become secondary-level educators. These aren’t just observers of the AI revolution as they’re about to walk into classrooms where AI is already reshaping how students learn, research, and think. The questions they asked weren’t theoretical. They were urgent, practical, and deeply thoughtful.

Rather than give rushed answers in the moment, I asked them to write down their questions. What came back was a collection of concerns that I suspect many of us share: How do we teach critical thinking when AI can think for students? How do we assess learning when AI can complete assignments? How do we prepare students for a world we can barely imagine ourselves?

These questions deserve more than quick responses. They deserve experimentation, evidence, and honest conversation. So over the next several weeks, I’m going to tackle them one at a time, not with definitive answers, but with frameworks, experiments, and resources that will help all of us think more clearly about AI’s role in education.

This week, we start with the question that underlies almost every concern about AI in schools:

The Question:

“How can AI be a good thing when it does the ‘thinking’ for the students? Helping with lesson planning, curricula, and simple tasks sure, but isn’t ‘artificial intelligence’ antithetical to the very skills I’m supposed to be teaching as a Social Studies teacher? Teachers have functioned just fine without it for so long, and unlike other new technologies that would supposedly ‘revolutionize learning’ (TV, Radio, the Internet, etc), Artificial Intelligence takes the cognitive load off of students (instead of just being a new medium in which information is presented), so how can we expect students to learn any better with a tool that lowers the amount of thinking students do?”

My Response

This question cuts to the heart of why so many educators feel uneasy about AI and it should! But I want to challenge one assumption embedded in your question: that AI necessarily “takes the cognitive load off students.”

Consider this: Does a calculator reduce mathematical thinking, or does it shift where that thinking happens? A student who uses a calculator to check 127 × 84 isn’t thinking about multiplication, but a student who uses a calculator to model compound interest over 40 years is thinking about something far more complex than arithmetic. (note: does anyone else take pictures of their numbers scribbled on paper to have AI calculate and solve in quick ways? no? Just me, then carry on)

The question isn’t whether AI reduces cognitive load. It’s where we want students’ cognitive energy to go. A student who spends 20 minutes formatting a bibliography isn’t thinking about history, they’re thinking about commas and italics. A student who uses AI to generate a first draft and then spends 40 minutes interrogating its arguments, identifying its biases, and rewriting its weak evidence? That student is thinking harder than they ever did before.

Let me show you what I mean as I know some of you are not quite in agreement with me which is prefectly fine.

The Ammonite or the Octopus?

Here’s a useful metaphor: 66 million years ago, an asteroid struck Earth with catastrophic force. In the oceans, the ammonite, a shelled creature that had evolved gradually over millions of years, went extinct. Its rigid shell, perfectly adapted for stable environments, became its downfall when the environment radically changed.

The octopus survived. Why? It was adaptable. Its soft body could change color instantly, squeeze through tight spaces, and even edit its own RNA in hours to adjust to temperature changes. When chaos hit, rigidity failed and flexibility thrived.

AI is our asteroid moment in education. We can be ammonites clinging to rigid structures that worked in stable times or we can be octopuses, adapting our approaches while staying true to our core mission: developing thinking humans.

The Bad Version (So You Know What to Avoid)

Here’s what cognitive offloading actually looks like:

Student version:

• “Write my essay on the Industrial Revolution” ← Student learns nothing

• “Give me the answer to question 5” ← Student learns nothing

Teacher version:

• “Create a lesson plan for me on the Civil War” ← Teacher learns nothing

• “Give me 5 discussion questions for my class” ← Teacher learns nothing

Notice the pattern? When AI does the final product, thinking disappears. When AI acts as a thinking partner during the process, thinking deepens.

The difference is whether you’re using AI as a vending machine (input request, receive finished product) or as a sparring partner (input thinking, receive challenge).

A Real Warning from Research:

There’s a genuine danger here that researchers call “cognitive sloth” when our natural preference for easy answers over mental effort. A 2025 study from Zhejiang University found that while AI made tasks faster and more effective, it actually undermined workers’ intrinsic motivation, leaving them bored and disengaged even after the AI was removed.

Suqing Wu et al., “Human-Generative AI Collaboration Enhances Task Performance but Undermines Human’s Intrinsic Motivation,” Scientific Reports 15, no. 1 (April 29, 2025): 15105, https://doi.org/10.1038/s41598-025-98385-2

This is why the HOW matters so much. We’re not just adopting a tool; we’re choosing whether that tool makes thinking visible or invisible.

Before You Try These Experiments...

When thinking about AI in your classroom, ask yourself these three questions:

1. What won’t humans do? (Tedious work that gets ignored or is impossible to scale like giving every student personalized feedback on every draft)

2. What shouldn’t humans do? (Error-prone, repetitive tasks like grading multiple-choice questions or checking citation formats)

3. What can’t humans do? (Processing patterns across thousands of student essays to identify common misconceptions)

These questions help us see AI not as a replacement for thinking, but as a tool that redirects cognitive energy toward higher-order skills.

Now, let’s test this with real experiments.

Try This Week: 3 Experiments to Test Your Thinking

Don’t just take my word for it. This week, I challenge you to try these three experiments. Pay attention to how your brain feels during each one such as are you relaxing or are you working harder?

Experiment 1: The Socratic Mirror (Logic)

The Concept: This is the ultimate “Anti-Cheating” use case. The AI refuses to give you answers. And yes the study modes of Gemini and ChatGPT already do this very thing, but there is something to prompting it yourself, modifying the prompt, and treating it like a digital play sandbox.

The Task: Pick a topic you are studying right now.

The Prompt:

I am studying [Topic]. I want you to test my understanding. Ask me one question at a time. If I get it wrong, give me a hint but do not give me the answer. If I get it right, ask a harder follow-up question. Do not lecture me; just quiz me.

The Shift: You are now in the “hot seat.” The cognitive load is entirely on you.

After you try this: How did it feel to be questioned rather than answered? Did you struggle to articulate what you thought you knew? That struggle is your brain working and this is what “productive struggle” looks like in real-time. And this is what we want our students to learn how to navigate and feel the sensation when we actually do know what we are talking about.

Experiment 2: The “Blind Spot” Check (Metacognition)

The Concept: We often assume our work is “complete.” This forces you to identify what is missing.

The Task: Paste a recent essay, assignment, or lesson plan you wrote into ChatGPT.

The Prompt:

Acting as a critical scholar in [Insert Subject], identify three major gaps, biases, or counter-arguments that I failed to address in this text. Don’t fix them; just list them.

The Shift: You aren’t asking the AI to write; you are asking it to audit your thinking. You have to evaluate if its critiques are valid.

After you try this: Did the AI’s critique make you defensive? Did you disagree with some of its points? That defensiveness is your brain working and you’re evaluating, not passively accepting. This is metacognition in action.

Experiment 3: The Perspective Simulator (Empathy & Bias)

The Concept: Understanding history or literature requires stepping into foreign mindsets.

The Task: Choose a controversial historical event (e.g., The Industrial Revolution, Westward Expansion, the New Deal).

The Prompt:

I want to interview a coal mine owner in 1850 and a labor union organizer in 1850. Simulate a conversation between the two of them where they disagree on the concept of “progress.” Let me ask them questions.

The Shift: Instead of reading a static textbook definition of “Industrialization,” you are actively navigating the emotional and economic logic of the people involved.

After you try this: Notice how much effort it took to formulate good follow-up questions. Did you have to think about what each person would value? What they would fear? That’s the cognitive load andyou’re synthesizing perspectives in real-time, not just memorizing facts.

More importantly, which I know this is more meta, but how do we know how to process this engagement to know if valid and factual or is our base knowledge on said topic not sufficient enough to detect where the algorithms are wrong?

A Cautionary Tale from Medicine

A 2023 study by researchers at Harvard Business School examined how AI affected physician performance. They discovered something surprising: AI working alone was actually MORE effective at diagnostic accuracy than radiologists working WITH AI.

Why? Because doctors often undervalued AI input compared to their own judgment even when AI was correct. They exhibited what researchers call “automation bias in reverse.”

But here’s the critical twist: AI was far less effective than humans at gathering patient information in initial consultations, frequently failing to ask follow-up questions and missing contextual clues.

The lesson? Neither humans nor AI should work alone. The key is using each for what they do best and teaching students to develop the critical judgment to know the difference.

Agarwal, Nikhil, Alex Moehring, Pranav Rajpurkar, and Tobias Salz. Combining Human Expertise with Artificial Intelligence: Experimental Evidence from Radiology. NBER Working Paper Series, no. w31422. Cambridge, Mass: National Bureau of Economic Research, 2023.

Prompt of the Week: “The Sparring Partner”

Copy/paste this to turn ChatGPT into a debate opponent rather than an answer machine.

Context: You are a debate expert and a historian specializing in [Insert Topic/Era].

Task: I am going to make an argument about [Insert Topic]. Your goal is to disagree with me. You must find a logical fallacy, a weak piece of evidence, or a missing perspective in my argument.

Rules:
1. Do not be polite. Be firm but academic.
2. Keep your responses short (under 100 words).
3. End every response with a question that forces me to clarify my position.

Let’s begin. My argument is: [Insert Your Argument Here]

Sidebar: Recommended Reading

Where to go if you want to nerd out on the science of “Cognitive Extension.”

The “Homework Apocalypse” (Ethan Mollick)
A must-read article from Wharton Professor Ethan Mollick. He argues that we shouldn’t ban AI, but rather assume that all homework will eventually be done with AI, and ask “how does that change what we assign?” And while posted back in July of 2023 it still holds water.
→

The Homework Apocalypse

Harvard’s AI Pedagogy Project
An incredible library of assignments created by Harvard researchers designed to deepen critical thinking using AI, rather than replace it.
→ metaLAB (at) Harvard

SchoolAI
If you want to see what a “safe” AI sandbox looks like for K-12 students, this tool allows teachers to build “Sidekicks” that are programmed to never give answers, only guidance.
→ SchoolAI.com

Watch This (10 Minutes That Will Change Your Mind)

Ethan Mollick’s keynote on “Co-Intelligence” demonstrates exactly what we’re talking about.

Your watching task: As you watch, count how many times Mollick describes AI increasing rather than decreasing the complexity of student work. (Spoiler: It’s most of the video.)

→

The Surprising Upside: AI Can Make You (and Your Students) Luckier

Here’s something unexpected: Organizations that use AI well don’t just get more efficient, they get luckier. Not by magic, but by systematically changing the odds in their favor.

The same principle applies to classrooms. When students use AI as a thinking partner, they develop what researchers call “strategic serendipity” or the ability to create more opportunities for breakthrough moments. They:

• Leverage help more effectively (AI can provide 24/7 tutoring and feedback)

• Use connections better (AI can surface sources, perspectives, and patterns they’d never discover alone)

• Control chaos (AI can help organize complex research and manage multi-step projects)

• Know what’s missing (AI can identify gaps in arguments and unexplored angles)

Luck isn’t random, it’s about creating conditions where good things are more likely to happen. That’s what thoughtful AI use does: it expands the possibility space for learning.

Think of it this way: A student researching the Civil Rights Movement who asks AI to “find three historians who disagree about the effectiveness of nonviolent protest” isn’t being lazy as they’re practicing the kind of intellectual curiosity that leads to deeper understanding. They’re making their own luck by asking better questions.

I Want to Hear From You

After you try one of these experiments, hit reply and tell me:

• Which experiment did you try?

• What surprised you?

• What still concerns you?

I’ll feature the most interesting responses in next week’s newsletter (with your permission, of course). Let’s learn from each other.

Final Thought

You wrote: “Teachers have functioned just fine without it for so long.”

This is true. But we’ve also “functioned just fine” without a lot of things that transformed education: the printing press, the library, the scientific method itself.

Consider this historical parallel: In 2009, an HP employee recognized the potential of touchscreen technology and tried to sound the alarm about the coming tablet revolution. Leadership ignored him, choosing to “wait and see” what Apple would do with the iPad. They could fast-follow, they reasoned. Today, Apple is worth 60 times HP.

The cost of waiting isn’t neutral. It’s the difference between shaping the change and being shaped by it.

The question isn’t whether we need AI, it’s whether AI, used thoughtfully, can help us teach the skills that matter most: critical thinking, perspective-taking, and metacognition.

I don’t think AI is the answer to everything. But I do think it’s a mirror. It shows us what we actually value. If we’re afraid students will use it to skip thinking, maybe we need to design assignments where thinking is unavoidable and where AI makes the thinking visible instead of invisible.

That’s what these experiments are designed to do.

Try one this week. Let me know what happens.