Home / Blog / AI and Your Mind

Field Note 03

AI Is Quietly Making You a Worse Thinker. Here Is How to Stop It

The more you trust the model, the less you think. That is not a worry. A 2025 study of knowledge workers measured it.

By Adrian Dunkley6 min readYour Mind

You feel sharper using AI. You are probably getting duller. Both can be true, and that gap is the trap. The output looks better while the muscle that produced your own output wastes, the way a calculator makes the arithmetic faster and your mental arithmetic worse at the same time.

In 2025, researchers from Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about 936 real tasks where they used generative AI. The finding, presented at the CHI conference: the more confidence a person had in the AI, the less critical thinking they reported doing. People with more confidence in their own skills kept thinking. People who trusted the tool offloaded the work to it. And across the group, AI use produced a less diverse set of outcomes, which the researchers read as a flattening of independent thought.

A tool that thinks for you is also a tool that lets you stop.
75%of knowledge workers use AI at work (Microsoft)
319workers studied on critical thinking (MS + CMU)
↓less critical thinking as AI confidence rises

Cognitive offloading is real, and it compounds

Psychologists call it cognitive offloading: when the brain learns it can lean on an external aid, it stops doing the work itself. We already offload phone numbers to our phones. Nobody minds. The difference with AI is what you are offloading: not a list of digits, but reasoning, judgement, the act of forming a view and defending it. Outsource that often enough and you do not have a view anymore. You have a prompt and a printout.

This matters more now because the tool is everywhere. Three in four knowledge workers use AI at work, per Microsoft's 2024 Work Trend Index, and most started in the last year or two. The exposure is daily and rising. The erosion is quiet, which is exactly why you have to defend against it on purpose.

The one move: draft with it, defend without it

Banning AI is not the answer, and you would not stick to it. The answer is to keep ownership of the thinking even when AI does the typing. Run a draft-then-defend rule.

Draft, then defend
1

Draft

Let the model produce a first pass if it helps. Speed is fine here.

2

Close it

Shut the tab. Now the answer has to come from you.

3

Defend

Rebuild or argue the key points in your own words. Check every fact yourself.

4

Keep what survives

If you cannot defend a claim, you do not understand it yet. Cut or learn it.

The model can write the draft. It cannot be the one who stands behind the work. That part stays yours.

Then add a floor: keep one real decision a week that you make start to finish with no AI in the loop. A hiring call, a strategy, a hard email, a household choice. Not because the model would get it wrong, but because the day you cannot do it without the model is the day you have handed away the thing that makes you worth asking.

Do this now

Pick your next meaningful piece of work. Use AI for the draft, then close it and write the three load-bearing sentences yourself, from memory, checking the facts by hand. Notice how much you actually knew. That gap, what you could not reproduce, is the muscle you have been outsourcing. Train it.

The compounding upside

This is not nostalgia for doing things the hard way. People who keep their judgement sharp are exactly the ones AI pays out for, because they can direct it, catch its mistakes, and own the result. The flattening the study found is a market signal: if everyone's output converges on the same model-shaped answer, the rare independent thinker is worth more, not less.

The takeaway

  • A 2025 Microsoft and Carnegie Mellon study found higher AI confidence linked to less critical thinking, and less diverse output.
  • The mechanism is cognitive offloading: lean on the aid, lose the skill.
  • Draft with AI, then close it and defend the work in your own words. Check facts by hand.
  • Keep one AI-free decision a week. The independent thinker gets more valuable, not less.

Frequently asked questions

Does using AI actually make you dumber?

It can erode specific skills through cognitive offloading. The 2025 Microsoft and Carnegie Mellon study found that workers who trusted AI more reported doing less critical thinking. Used deliberately, with a draft-then-defend habit, AI does not have to cost you your judgement.

Should I stop using AI for work?

No. Three in four knowledge workers use it, and avoiding it has its own cost. Keep using it for drafts and grunt work, but keep ownership of the thinking: defend the output yourself and keep one AI-free decision a week.

What is cognitive offloading?

It is the brain's habit of relying on external tools instead of doing a task itself. Useful for phone numbers. Costly when what you offload is reasoning and judgement, because those skills fade when unused.

This is not anti-technology panic.

It is the opposite of doom. The point is to use AI hard and stay sharp, on purpose. Survival Guide for the AI Apocalypse shows you how to protect your attention and judgement while still getting the speed, written by an insider who builds these systems. Keep your mind. It is the one tool you cannot reinstall.

Order on Amazon →
About the author

Adrian Dunkley has spent close to two decades building, deploying, and governing AI systems. He writes the briefing he gives friends and family. More at the book and adriandunkley.net.

Sources

  1. Lee, H.-P. et al. (Microsoft Research and Carnegie Mellon University), "The Impact of Generative AI on Critical Thinking," Proceedings of CHI 2025: survey of 319 knowledge workers across 936 tasks; higher AI confidence linked to less critical thinking and less diverse outcomes.
  2. Microsoft and LinkedIn, 2024 Work Trend Index (May 2024): 75% of knowledge workers use AI at work.