Cognitive Debt and AI: What Kids Lose When AI Does the Thinking

Every time your kid asks a chatbot to outline an essay, they’re making a trade: less effort now, possibly less skill later. Researchers at MIT gave that trade a name in 2025: cognitive debt. The idea is that work you hand to AI is “paid for later,” in skills you didn’t practice. Cognitive debt and AI have become a common worry for parents since. But handing thinking to tools isn’t new or always bad. Writing a grocery list is offloading. So is a calculator, a calendar reminder or tilting your head to read sideways text. Psychologists call it cognitive offloading, and they’ve studied it for years. The real question isn’t whether kids should offload thinking. It’s which thinking. The evidence so far, which is real but mostly correlational or from adults, points to a usable rule: offload what you’ve already learned or don’t need to learn; keep the parts of a task that are the learning. Here’s the mechanism, the evidence on both sides, and activities to teach your kid where the line is.

Key Takeaways

• Cognitive offloading is “the use of physical action to alter the information processing requirements of a task so as to reduce cognitive demand” (Risko and Gilbert, 2016). It’s everyday and often useful.
• Cognitive debt is the MIT Media Lab’s term for the cost of offloading the thinking that builds skill. Its source study is a small, adult, not-yet-peer-reviewed preprint.
• In a 666-person survey, frequent AI use was linked to lower critical-thinking scores through more offloading, and younger participants relied on AI most (Gerlich, 2025). The design is correlational.
• In 319 knowledge workers, higher confidence in AI went with less critical thinking, and higher self-confidence with more (Lee et al., 2025).
• Meta-analyses also find AI assistance can improve performance and higher-order thinking, mainly when it scaffolds a learner’s effort rather than replacing it.

Cognitive Debt and AI: What Offloading and Debt Mean

Cognitive offloading has a precise definition. In a 2016 review in Trends in Cognitive Sciences, Evan Risko and Sam Gilbert describe it as “the use of physical action to alter the information processing requirements of a task so as to reduce cognitive demand.” Their examples are ordinary: tilting your head to perceive a rotated image, or programming a smartphone to remind you of an appointment. Offloading frees working memory, the small mental workspace where we hold and manipulate information, for other things.

Cognitive debt is newer and less settled. Nataliya Kosmyna and colleagues at the MIT Media Lab coined it in “Your Brain on ChatGPT,” a 2025 preprint in which adults who wrote essays with a chatbot showed weaker brain connectivity, felt less ownership of their work and struggled to quote it. As a July 2026 paper in Frontiers in Developmental Psychology summarizes, cognitive debt is “the idea that work handed to AI is paid for later.” We cover that study, and the critique of it, in our article on the MIT study.

The difference matters. Offloading is a behavior. Debt is a claim about its long-term cost. The first is well established. The second is plausible, partly supported and still being tested.

How Offloading Can Turn Into Debt

The mechanism runs through practice. Skills get built by doing the effortful part repeatedly: retrieving a fact, structuring an argument, debugging a mistake. Two well-established findings make this concrete.

You remember what you generate. In 1978, Norman Slamecka and Peter Graf showed across five experiments that people remember words they produced better than words they read, the generation effect. When AI generates, the child is mostly reading.

Effort feels worse even when it works better. In a 2019 PNAS study, Louis Deslauriers and colleagues found physics students learned more in active classes but felt they learned less, reading the effort as a sign of poor learning. That’s why kids happily offload the effortful part: it feels like a good deal.

For adolescents there’s an extra layer. Jorge Pereira Campos and Tatiana Koff argue in Frontiers in Developmental Psychology that executive functions such as planning and weighing evidence mature rapidly between about 10 and 20, driven by practice. They call it the “cognitive offloading paradox”: AI is most useful for exactly the tasks whose effortful practice builds those skills. They present it as a hypothesis.

The Evidence For and Against

Evidence that worries researchers

Gerlich, 2025. In a mixed-methods study of 666 participants published in Societies, Michael Gerlich found “a significant negative correlation between frequent AI tool usage and critical thinking abilities, mediated by increased cognitive offloading.” “Younger participants exhibited higher dependence on AI tools and lower critical thinking scores.” Limits: it’s a survey with interviews, so it shows association, not cause. People who think less critically may also lean on AI more.

Lee et al., 2025. Microsoft Research surveyed 319 knowledge workers, who shared 936 examples of using generative AI, for a CHI 2025 paper. “Higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking.” AI also shifted the kind of thinking people did, toward “information verification, response integration, and task stewardship.” Limits: adults, self-reported.

Bastani et al., 2025. In a field experiment with nearly a thousand Turkish high school math students in PNAS, students given an unrestricted GPT-4 tutor did better during practice but 17% worse than never-users once access was removed. They used it “as a crutch.” A version designed to give hints largely avoided the harm.

Evidence that complicates the story

Lower brain activity can mean efficiency. Campos and Koff note that “lower neural activity need not mean worse cognition; in studies of expertise it often marks more efficient processing.”

Children’s output didn’t suffer in one study. They also describe a small preprint in which children’s brain control networks were less connected than adults’ while co-creating with ChatGPT, yet “their creativity scores… were no different from the adults’.”

Meta-analyses find benefits. Campos and Koff cite a meta-analysis of 69 experiments finding AI assistance “tends to improve academic performance and higher-order thinking,” and another of 29 finding a moderate positive effect, stronger for students who already self-regulate well. Their summary: benefits “are clearest when the AI scaffolds a student’s thinking rather than doing the work for them.”

So the research is mixed, and the mix has a pattern. Harm shows up when AI substitutes for the thinking that is the point of the task. Benefit shows up when AI supports that thinking.

When Offloading Is Fine, and When It Costs

Task	Is this skill the learning goal right now?	Offload to a tool?	Example
Arithmetic in a physics problem, after mastering arithmetic	No	Yes	Calculator frees attention for the physics
Learning multiplication facts in third grade	Yes	No	Practice is the point
Remembering a dentist appointment	No	Yes	Calendar reminder
Planning the structure of an essay	Yes, for a student learning to write	Not at first; AI feedback after a draft is fine	Outline yourself, then ask AI what’s missing
Spell-checking a finished story	Usually no	Yes	Tool catches typos; the story is theirs
Debugging their own code	Yes	Hints only	Ask AI to explain the error, not fix it
Summarizing a book for a book report	Yes	No	Reading and summarizing is the assignment
Looking up a fact to check a claim	No	Yes, and verify it	Search or AI, then confirm with a second source

The rule underneath the table: offload what you’ve already mastered, or what isn’t what you’re trying to learn. Keep the part that is the learning. That’s the scaffold-versus-substitute line in one sentence.

How to Teach Your Kid About Cognitive Debt

Ages 5–8: The “brain jobs” sort

Write cards for everyday thinking jobs: remembering a phone number, adding up snack money, spelling your name, choosing a birthday present for a friend, remembering where your shoes are, making up a story. Sort them together into two piles: “jobs a tool can do for me” and “jobs that make my brain stronger.” Then talk about the tricky ones. Most kids decide the story and the present are brain jobs. That’s the lesson.

Ages 9–12: Test the generation effect on yourself

Make two lists of 10 words. For list one, your child just reads each word. For list two, they generate each word from a clue with some letters missing (“h_t, opposite of cold”). Wait 15 minutes, then ask them to write down every word they remember. Most kids remember more of the generated list. Then ask: “If an AI writes your sentences for you, which list is that like?”

Ages 13+: A one-week AI log and a verification challenge

For one week, every time your teen uses AI, they jot down three things: what they asked, what thinking they handed over (planning, recalling, explaining, deciding), and whether they could have done it themselves. At the end, they sort the entries into “fine to offload” and “I needed to do that.” Then a challenge: ask an AI five questions in a subject they know well and find any errors or weak spots. The Microsoft study found verification is where critical thinking goes when people use AI. Practicing it builds the self-confidence that study linked to more critical thinking.

The question to ask: “When is it smart to let a tool remember or work something out for you, and when does it cost you something you’ll need later?”

Habits That Keep Offloading From Becoming Debt

Do the first pass yourself

First draft, first attempt, first outline: the child’s. AI comes in second, for feedback or questions. This one habit keeps most homework on the scaffold side.

Use hint mode by default

The PNAS experiment showed design matters. Set chatbots to study or tutor modes where available, or teach the prompt “Give me a hint, not the answer.”

Build self-confidence through real wins

The Lee study linked self-confidence to more critical thinking with AI. Kids build that confidence by succeeding at hard things themselves, which is a reason to keep some projects AI-free.

What not to do

Don’t treat all offloading as cheating. Calculators, spell-checkers and reminders are legitimate tools. Teaching your child to tell the difference is more useful than banning tools, and more likely to survive the teenage years. For a wider look at the research, see is ChatGPT making kids worse at thinking?

What to Watch For Over the Next 3 Months

• Week 4: Can your child name one thing they let AI do this week that was fine, and one that wasn’t? Is the first draft of homework theirs?
• Month 2 red flags: Reaching for AI before trying; inability to start a task without it; homework that’s fast and polished but can’t be explained.
• Month 3 self-check: Repeat the generation-effect test or the AI log. Has your child started making the offload-or-not call on their own? That judgment is the skill.

Frequently Asked Questions

What is cognitive debt?

Cognitive debt is a term from MIT Media Lab researchers for the idea that thinking you hand to AI now is paid for later, in skills you didn’t build. It comes from a 2025 preprint on adults writing essays with ChatGPT. The idea is plausible and partly supported, but it hasn’t been tested directly in children.

What is cognitive offloading?

Cognitive offloading is using an action or tool to reduce the mental effort a task needs, such as writing a list, setting a reminder or using a calculator. Psychologists Evan Risko and Sam Gilbert defined it in a 2016 review. It’s normal and often helpful; the cost comes when you offload the part you’re trying to learn.

Does AI make kids worse at critical thinking?

The evidence is mixed. Some studies link heavy AI use to lower critical-thinking scores, especially among younger users, but those are correlational. Experiments show AI can hurt learning when it gives answers and help when it gives hints. The way AI is used matters more than whether it’s used.

When is it okay for my child to let AI do the thinking?

When the skill isn’t what they’re trying to learn, or when they’ve already mastered it: checking spelling on a finished essay, looking up a fact, doing arithmetic inside a harder problem. It’s not okay when the offloaded part is the assignment’s purpose, such as planning an argument or summarizing a book.

How can I tell if my child is building cognitive debt?

Watch for reaching for AI before trying, difficulty starting tasks without it, and work that looks polished but can’t be explained. Closed-book tests that slip while homework improves are another sign. Asking your child to explain their reasoning is the simplest check.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Risko, E. F., & Gilbert, S. J. (2016). “Cognitive Offloading.” Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
2. Kosmyna, N., Hauptmann, E., Yuan, Y. T., et al. (2025). “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” arXiv (preprint, not peer-reviewed). https://arxiv.org/abs/2506.08872
3. Pereira Campos, J., & Koff, T. (2026). “Your brain on ChatGPT, but whose brain? The missing adolescent in AI-cognition research.” Frontiers in Developmental Psychology, 4. https://doi.org/10.3389/fdpys.2026.1885225
4. Gerlich, M. (2025). “AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking.” Societies, 15(1), 6. https://doi.org/10.3390/soc15010006
5. Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers.” CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3706598.3713778
6. Bastani, H., Bastani, O., Sungu, A., et al. (2025). “Generative AI without guardrails can harm learning: Evidence from high school mathematics.” Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2422633122
7. Slamecka, N. J., & Graf, P. (1978). “The generation effect: Delineation of a phenomenon.” Journal of Experimental Psychology: Human Learning and Memory, 4(6). https://doi.org/10.1037/0278-7393.4.6.592
8. Deslauriers, L., McCarty, L. S., Miller, K., Callaghan, K., & Kestin, G. (2019). “Measuring actual learning versus feeling of learning in response to being actively engaged in the classroom.” Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.1821936116