WORLD’S FIRST TITANIUM CHASENFollow the tradition — pass on the ware.Take the 1-question survey →
Follow Studio Fungwa

Does Using AI Make You Think Less? Cognitive Offloading, Explained

Cognitive offloading and thinking with AI

You ask an AI to draft an email, summarise a report, or solve a problem you could have worked through yourself. The result arrives in seconds. What happened to the thinking you skipped — and does skipping it, repeatedly, change what you are able to do without help? Psychologists call the handover cognitive offloading. In 2025 two widely reported studies put numbers on it, and in 2026 the debate about what those numbers actually prove is still active. This guide separates the findings from the headlines.

Offloading is old. The scale is new

Humans have always offloaded memory and calculation: notebooks, calculators, calendars, search engines. Offloading itself is not a flaw. Writing a phone number down frees working memory for the conversation you are actually having. The question has never been whether to offload, but what.

Generative AI changes the category of thing that can be handed over. A calculator offloads arithmetic, a step inside a larger task you still direct. An AI assistant can offload the framing of the problem, the first draft, the comparison of options and the conclusion — the parts of a task where reasoning and learning normally happen. That is why researchers moved quickly, and why the results deserve a careful rather than a panicked reading.

What the MIT essay study found

In 2025, researchers at the MIT Media Lab released a preprint, often cited as “Your Brain on ChatGPT,” in which 54 participants wrote essays in three conditions: with ChatGPT, with a search engine, or with no tools, while EEG recorded brain activity. The ChatGPT group showed the lowest measured neural connectivity of the three groups, and a striking number of its members struggled to quote a sentence from an essay they had just produced. The authors used the term “cognitive debt” for effort borrowed now and repaid later in weaker independent performance.

The limits matter as much as the result. Fifty-four people is a small sample, the task was timed lab essay writing, and the paper was released as a preprint rather than after full peer review. It shows what happened during assisted writing in that setting. It does not, by itself, prove that using AI causes lasting damage to intelligence — a distinction our guide to how IQ tests work would insist on for any single study.

What the Microsoft and Carnegie Mellon survey found

A second 2025 study, presented at CHI 2025 by researchers from Microsoft and Carnegie Mellon University, surveyed 319 knowledge workers about 936 real examples of using generative AI at work. Its central pattern was about confidence. The more a worker trusted the AI, the less critical thinking they reported applying to its output. The more they trusted their own ability, the more critical thinking they reported. Critical thinking did not simply vanish; in many accounts it shifted — from producing an answer toward checking, editing and taking responsibility for one.

This was a survey built on self-reports, so it measures what workers perceived and remembered, not a direct test of their reasoning before and after. A third 2025 study of 666 participants reported a strong negative association between frequent AI tool use and critical-thinking scores, with offloading as the proposed link and younger participants showing the strongest pattern — again correlational, meaning heavy use and lower scores travel together without proving which causes which, or whether a third factor drives both.

The honest summary as of 2026

Put together, the evidence as of October 2026 supports a conditional claim, not a sweeping one. When people hand over the steps of a task where learning happens, and then do not check or engage with the result, they practise less, remember less of what was produced, and report thinking less critically. No long-term longitudinal study yet tracks whether regular AI users lose measurable cognitive ability over years. Anyone who tells you the science has settled that question — in either direction — is ahead of the evidence.

There is also a counterweight in the same literature. Studies of AI tutors that give hints instead of finished answers have found the learning penalty can disappear, and sometimes reverse, when the tool is designed to keep the learner working. Design and habit, not the mere presence of AI, appear to decide the outcome. That is consistent with what we found testing how to use AI to get smarter: the prompt is the pedagogy.

A practical rule: offload the step, keep the skill

Before handing a task to AI, ask two questions. First: can I already do this well without help? If yes, offloading is leverage — you are spending saved time elsewhere, the way a strong mental arithmetician may still use a spreadsheet. Second: am I supposed to be learning this right now? If yes, offloading the core step borrows against the skill you came to build. Use the AI around the step instead: to quiz you, to generate practice problems, to critique your draft, or to explain a point you got stuck on after trying.

  • Draft yourself first, even badly, when the writing is the thinking. A rough paragraph of your own gives you something to compare the AI version against.
  • Read output as a reviewer, not a recipient. Check one claim against a source. Argue with one recommendation. That checking step is where critical thinking now lives.
  • Close the loop from memory. After an AI-assisted session, write down what you learned without looking. If you cannot, the session produced output, not learning.
  • Keep some reps unaided. Musicians practise scales without accompaniment. A short weekly task done entirely without AI keeps your baseline visible — to you.

None of this requires using AI less overall. It requires noticing which mental steps you are handing over, in the same way brain training research asks which ability an exercise actually practises. The tool is not the habit. The habit is the training.

Go deeper with AI

This article is the starting point. Copy any of these prompts into your favorite AI assistant to learn more about cognitive offloading and AI use:

🔎 Use AI as a tutor instead of an answer machine

I am learning [topic]. Do not give me finished answers. Act as a tutor: ask me one question at a time, give hints when I am stuck, make me explain my reasoning back to you, and only show a worked example after I have attempted the problem myself. Start with a diagnostic question.

🧭 Audit a task before you hand it to AI

Here is a task I am about to give to AI: [describe it]. Split it into steps and label each step: (1) safe to offload because I already know how to do it and it is mechanical, (2) risky to offload because doing it builds a skill I need, or (3) offload the draft but keep the checking myself. Explain each label in one sentence.

💡 Rebuild a skill you have been outsourcing

I have been using AI for [skill — e.g. writing first drafts, mental maths, summarising articles] and I worry I am getting worse at it unaided. Design a 2-week practice plan: short daily exercises without AI, one AI-assisted session per week where I critique the AI output instead of copying it, and a self-test at the end to measure whether I improved.

Tip: replace the bracketed parts with your own situation — the more specific your prompt, the more useful the answer.

The short summary

Cognitive offloading — handing mental steps to a tool — is as old as the notebook, but generative AI lets people hand over framing, drafting and judging, not just storage and arithmetic. A 2025 MIT Media Lab preprint (54 participants, EEG) found the lowest brain connectivity and weakest recall of their own text in the ChatGPT essay group; a Microsoft / Carnegie Mellon survey of 319 knowledge workers (CHI 2025) found higher trust in AI tracked with less reported critical thinking, while self-confidence tracked with more. Both are short-term, small or self-reported evidence, and no longitudinal study yet proves lasting ability loss. Tutor-style AI that withholds answers removes the penalty in related work, which points to habit and design as the deciding factors.

The single most important point: offloading a skill you already own is leverage, but offloading a skill you are trying to build is decay — so keep the learning steps, and hand over only the mechanical ones.

Want better prompts, in your inbox?

One sharper AI prompt + one new brain training and IQ topic every week. Free forever, unsubscribe anytime.

We never share your email. One email a week, that's the deal.

Try the IQ Challenge

Take the free 25-question timed quiz and get your estimated IQ and percentile instantly.

Start the Challenge


Keep exploring: see how IQ tests actually work, read about working memory, and check whether brain training works.

For entertainment only — this guide is for fun and general information, not a clinical or professional assessment.

© FUNGWA 峯華 LLC (fēng huá — “peaks of splendor”)