Mind your manners11 minSam Altman says polite users have cost OpenAI tens of millions of dollars. Lab studies show rude prompts can knock accuracy down by double digits, while over-the-top flattery doesn't help either. Here's what manners actually do inside a language model, what they cost, and the one habit worth keeping.Read now →

AI Is Doing More of Your Thinking. The Research on What That Costs Your Brain Is Getting Hard to Ignore
Brain scans of students writing with ChatGPT, a flood of 'workslop' at the office, a sneaker company that became an AI stock overnight: the evidence that we're outsourcing judgment, not just typing, keeps piling up. Here's what the research actually shows, why groupthink makes it worse, and a practical plan to keep your mind your own.
History is full of moments that make no sense in hindsight. Smart people making disastrous decisions. Whole groups convincing themselves of things any outsider could see were wrong.
We usually look back and ask: what were they thinking?
Maybe that's the wrong question. Maybe the better one is: were they thinking at all?
Critical thinking isn't a permanent possession. It's a skill. Like any skill, it weakens when you stop using it. And right now, a growing body of research suggests many of us are handing more and more of it to machines, often without noticing.
This isn't an argument against AI. It's an argument for knowing what you're giving away.
Groupthink: the old bug in human thinking
Start with a much older problem.
Groupthink is the psychological pattern where the desire for harmony inside a group overrides clear thinking. Under pressure to fit in, people stop questioning. They go along. And to protect the consensus, groups sometimes join together to silence those who disagree.
That's how individually intelligent people end up making collectively irrational choices. Historians have used conformity pressure, alongside propaganda and grievance, to help explain how ordinary people embraced catastrophic movements in the 20th century.
You might expect that, with more education and information than ever, we'd have outgrown it. The argument here is the opposite: groupthink may be accelerating, and technology that makes it effortless to stop thinking is pouring fuel on it.
Cognitive offloading: useful, until it isn't
Humans have always used tools to lighten mental load. Psychologists call it cognitive offloading.
- Writing meant we didn't have to remember everything.
- Calculators spared us tedious arithmetic.
- GPS let anyone navigate anywhere without studying maps.
- Search engines meant we store fewer facts, because we know we can look them up.
Offloading is often a good trade. It frees capacity for harder problems. But taken too far, research has linked it to weaker memory and less independent reasoning. GPS users, for example, tend to form poorer mental maps than people who navigate on their own.
None of those tools comes close to what AI now offers. A calculator does arithmetic. A chatbot will write the essay, structure the argument, summarize the research and suggest what to conclude.
That's not offloading a task. It's offloading the thinking that the task used to train.
What the brain scans showed
The most-discussed evidence comes from the MIT Media Lab.
In a study titled "Your Brain on ChatGPT," researchers had 54 participants write essays over four months in three groups: one using ChatGPT, one using a search engine, and one using only their own brains. EEG headsets measured brain activity throughout.
The findings:
- The brain-only group showed the strongest and most widely distributed neural connectivity.
- The search group showed moderate engagement.
- The ChatGPT group showed the weakest connectivity.
- ChatGPT users reported less ownership of their essays and had poorer recall of what they'd written, sometimes struggling to quote their own work.
The researchers coined the term "cognitive debt": the convenience you gain now is paid back later in weaker engagement and understanding.
The important caveats
This study deserves careful handling, and its critics have made fair points. It was released as a preprint, not initially peer-reviewed. The sample was small. A later published comment raised concerns about EEG statistics and how connectivity differences were interpreted.
So it doesn't prove that ChatGPT damages brains. What it does suggest, consistent with decades of learning research, is that when a tool does the effortful part, you don't get the benefit of the effort.
Why writing is thinking
That's the heart of the education worry.
Writing an essay isn't just producing text. It's feeling your way through an argument, discovering what you actually believe, noticing the gaps. That kind of sophisticated thinking is hard to do any other way.
A student who has an AI write the paper misses that development entirely. And let's be honest, most aren't doing a rigorous edit either.
Some argue students are learning a new skill: prompting. That's a weak defense. Chatbots are designed to be usable by everyone with minimal effort, and each model update makes good results easier to get with less skill. Prompting is closer to learning a smartphone than learning to reason.
Those students will carry the habit into the workplace, assuming AI has left entry-level jobs for them to fill.
At work: outcomes over understanding
Workplaces care about outcomes more than process, which is a big reason AI fits so neatly into white-collar jobs.
But there's a hidden cost. A worker who has AI produce their work absorbs about as much of it as the student absorbs from an AI essay. That shows up in three ways:
- They can't explain it. If the output needs defending or adapting, they're stuck.
- They understand their field less. Domain knowledge grows from doing the work.
- They connect with clients less. Sending AI-written emails means never really internalizing what customers need.
The less you engage with your work, the less it becomes part of you.
Microsoft and Carnegie Mellon researchers found a related pattern in a 2025 survey of 319 knowledge workers: the more confidence people had in generative AI, the less critical thinking they reported applying to its output. People who were more confident in their own abilities thought more critically.
The rise of "workslop"
There's now a name for what happens when that habit spreads: workslop.
You've probably experienced it. A coworker sends you something to review, maybe a public-facing email. The sentences are fine individually, but the whole thing feels oddly hollow, as if you're reading an imitation of communication rather than the real thing. You wonder whether you're the problem. You ask the author questions they can't answer. Eventually they admit AI wrote it, without much embarrassment. You end up rewriting it almost from scratch, losing hours you hadn't planned on.
Researchers at BetterUp Labs and the Stanford Social Media Lab surveyed 1,150 US full-time desk workers. 40% said they had received workslop in the previous month.
The core failure is a lack of critical thinking: trusting an AI to do the work and not reviewing it. Careful editing of AI output can take as long as writing it yourself, so under time pressure, the review is the first thing cut.
And management often reinforces it, pushing AI adoption to justify headcount cuts or chase productivity metrics.
Groupthink, meet the AI boom
Why would leaders accept lower-quality work? Back to groupthink.
The business world has embraced AI almost universally, sometimes regardless of results. The clearest symbol: in April 2026, struggling sneaker brand Allbirds announced a pivot to AI compute infrastructure and a rebrand as NewBird AI, backed by a $50 million financing facility to buy GPUs. Its shares jumped more than 600% in a day.
In many offices, criticizing AI has become a social risk. People who question it worry about being labeled as backward. Notice how often criticism starts with an apology: "Look, I'm not against AI, but..."
That's the hallmark of conformity pressure, and it persists even against sobering evidence. MIT's 2025 "GenAI Divide" research concluded that about 95% of the organizations it studied saw no measurable return from generative AI pilots. Meanwhile, some large employers have cut jobs while citing AI efficiency, even as workers report that cleaning up AI-generated "code slop" adds to their workload, and that training systems to replace them is demoralizing.
The bigger questions get little airtime. If AI replaces huge numbers of workers, who buys the products? What happens without some form of income support? Industry leaders sometimes describe a future where intelligence is metered and sold like electricity, a utility you need to pay for. The key word is need: critical thinking as an external service rather than a capacity you carry.
We were conditioned for this
AI isn't our first mass experiment in outsourced attention. Social media got there first.
Early social networks were about friends and photos. Over time, algorithms began ranking content by the reactions it provoked, optimizing for time on platform and ad views. Constant notifications chopped up attention. Doomscrolling fed an endless stream of alarming news, leaving many people anxious and less able to sustain focus.
Accuracy suffered too. A UNESCO survey of digital content creators found 62% said they don't systematically verify information before sharing it.
Algorithms also tailor feeds to what we already believe, creating echo chambers that feed confirmation bias and polarization. Online tribes form, mock each other, and enforce conformity within their own ranks, with real fear of public shaming for stepping out of line.
Add the casual spread of clinical language, calling ordinary people "narcissists" or ordinary conflict "trauma," and you get labels that sound authoritative and short-circuit questioning. Language with an air of expertise tends to suppress dissent, a classic groupthink ingredient.
Now layer AI on top: feeds filling with AI-generated content, posts written by AI, opinions shaped by the same few models. It can start to feel like living in a hive mind. The recent TV series Pluribus, from Breaking Bad creator Vince Gilligan, dramatizes exactly that: nearly all of humanity merges into one unified mind, and a few individuals who remain themselves are horrified by its perfect consensus.
Even brilliant people aren't immune. Robert McNamara, a celebrated intellect and architect of America's war in Vietnam, kept advocating it long after it was failing, a textbook case of the sunk cost fallacy overpowering clear evaluation.
What critical thinking actually is
Definitions vary across philosophers, educators and psychologists. But they agree on one thing: critical thinking takes effort, and it doesn't happen automatically.
The American philosopher John Dewey, who helped define the concept, believed it benefits both individuals and society: more personal fulfillment, less collective waste. He argued teachers should respect students' demand for reasons, encourage self-sufficiency and ground them in disciplines like history, science and math, preparing them to be capable democratic citizens.
Those are exactly the habits now under pressure. Teachers have increasingly sounded alarms about students' reading stamina. But there are counter-signals too. Long-form content is thriving: people are choosing long podcasts, books, lengthy articles and longer videos. Surveys have suggested younger audiences are going back to movie theaters more than many predicted. Those may be signs that people feel the emptiness of endless feeds and are pushing back.
How to keep your mind your own
Practical advice comes from Gloria Mark, a professor of informatics at UC Irvine who studies attention.
- Admit scrolling has little value, then replace it with something that does. Thirty minutes in a book leaves a memorable experience; thirty minutes of short clips often leaves a vague, uneasy feeling and little memory.
- Probe your impulses. Before reaching for your phone, ask: why am I checking this? Is now a good time, or am I just bored?
- Schedule breaks deliberately. Your mind needs rest after focused work. When you're fatigued, you're more likely to act compulsively.
- Work with your rhythms. Track when your focus peaks and do your hardest thinking then.
And for AI specifically:
- Think first, then prompt. Write your own outline or answer before asking the AI.
- Use AI as a sparring partner, not an author. Ask it to critique your argument instead of writing it.
- Never send what you can't explain. If you couldn't defend it in a meeting, don't ship it.
- Keep some tasks fully manual, like writing that matters to you, mental math and navigation on familiar routes, to keep the skills alive.
- Seek disagreement. Ask the AI for the strongest counterargument, and talk to people outside your usual circles.
- Practice restraint. Learn when to disengage from online fights designed to have no winner.
There's an old Buddhist parable about a man struck by a poisoned arrow who refuses treatment until he knows who shot it and why. The point: deal with the urgent problem first. Here, that means action, not just awareness.
Why this matters
AI is the most powerful cognitive tool humans have ever built. That's exactly why its effect on thinking matters. A tool that can do your reasoning will, if you let it, do your reasoning.
The stakes go beyond productivity. Democracies depend on citizens who can weigh evidence, resist manipulation and disagree productively. Businesses depend on people who understand their work well enough to notice when something's wrong. A society that offloads judgment becomes easier to steer.
There's also a deeper point. Human intelligence evolved largely as social intelligence: modeling other minds, cooperating, competing, building institutions that outlast individuals. Intelligence increasingly lives in the relationships between minds. If more of those relationships run through the same few AI systems, our collective thinking could converge in ways nobody chose.
Counterpoints
- The evidence is early. The MIT study is small and contested. Much other evidence is survey-based and self-reported.
- Every new tool sparked this fear. Socrates worried writing would weaken memory. Calculators didn't destroy math. Some offloading genuinely frees people for higher-level work.
- AI can strengthen thinking too. Used as a tutor or critic, it can explain concepts, challenge assumptions and expose people to opposing views.
- Workslop is a management problem. Clear quality standards and review norms can fix much of it.
- Correlation isn't causation. People who lean heavily on AI may differ in other ways from those who don't.
FAQ
Does using ChatGPT damage your brain? There's no proof of damage. An MIT Media Lab study found weaker brain connectivity and poorer recall when people wrote essays with ChatGPT, but it was small, initially not peer-reviewed and has drawn methodological criticism.
What is cognitive offloading? Using external tools, from notes to GPS to AI, to reduce mental effort. It's often helpful, but heavy reliance can weaken the skills the tool replaces.
What is workslop? AI-generated work that looks polished but lacks substance, shifting the burden to whoever receives it. In one survey, 40% of US desk workers said they'd received it in the past month.
How can I use AI without losing critical thinking? Think before you prompt, use AI to critique rather than create, never send work you can't explain, and keep some important tasks manual.
The bottom line
AI doesn't make anyone stop thinking. It just makes not thinking extremely convenient.
The research is early and imperfect, but it all points the same way: the effort you skip is the growth you skip. Combine that with groupthink at work and algorithmic echo chambers at home, and outsourcing your judgment becomes the default rather than a choice.
Use the tool. Just don't let it use up the part of you that asks whether it's right.
Sources
- Your Brain on ChatGPT: Accumulation of Cognitive Debt (MIT Media Lab)
- Comment on "Your Brain on ChatGPT" (arXiv)
- AI-Generated "Workslop" Is Destroying Productivity (Harvard Business Review)
- Workslop research (BetterUp Labs)
- The Impact of Generative AI on Critical Thinking (Microsoft Research)
- Allbirds shares jump over 400% on plans to pivot to AI from sneakers (Reuters)
- From wool sneakers to GPUs: Allbirds' AI pivot and 600% stock surge (Fortune)
- MIT report: 95% of corporate generative AI pilots fail to deliver returns (Computing)





