Is AI making us think less?

What if the most pressing danger of artificial intelligence is not the loss of jobs or a wave of layoffs, but a quieter erosion that takes place in our heads. A new study by researchers from Microsoft and Carnegie Mellon proposes an unsettling idea. When people trust AI to carry more of the load, they tend to do less of their own heavy thinking. Over time, that habit could blunt the very skills that make us human. The paper surveyed 319 knowledge workers who already use AI weekly. Participants shared 936 real work examples, from looking up facts to drafting emails to summarizing reports. The more confident people felt in AI’s ability to finish the task, the less they engaged the six classic layers of critical thought. Those layers include remembering ideas, understanding them, applying them to real contexts, breaking them apart, combining them in new ways, and judging them for quality. The researchers warn that reliance on AI can shrink “routine opportunities to practice judgment,” which risks leaving users unprepared when unusual cases land on their desks.

This is not an abstract debate playing out among engineers. It is a lifestyle question that touches how we learn, decide, and relate. The modern workday already nudges us toward convenience. We use maps rather than memorize streets, we autopay bills rather than reconcile line items, we rely on feed algorithms rather than curate ideas. AI promises more convenience still. It drafts emails and proposals, cleans up code, builds slides, and writes summaries. These are useful. Yet there is a trap inside the usefulness. If convenience becomes the goal, rather than a bridge, our mental muscles stay on the couch.

The Microsoft and Carnegie Mellon team did not measure brain scans or run clinical trials. They gathered self-reports from real workers about perceived effort and the kinds of thinking they engaged while using AI. We should treat self-report carefully. Still, the pattern they surfaced is both intuitive and consequential. People saved time, felt less cognitive strain, and often skipped deeper steps like cross-checking or interrogating assumptions, especially for routine or lower-stakes tasks. That choice can feel harmless on Tuesday afternoon. Multiply it across months, and a different picture emerges. Habits form. The default becomes to accept the first plausible output and to reserve our effort for the rare crisis. That looks like comfort. It behaves like atrophy.

Anxiety about AI usually starts with jobs, not judgment. The fear is widespread. Surveys this year found that many American workers worry about the long-term effect of AI on their opportunities, and a significant share believe technologies could make their jobs obsolete or reduce openings in their field. Even if the exact numbers vary by poll, the trend is clear. Concern is not fringe. It has become a mainstream sentiment across offices and shop floors.

The twist is that, while the labor market may adjust over years, our minds adjust right now. We do not need a pink slip to feel the shift. We only need an autocomplete box that is very good at sounding right. The moment we outsource a judgment we could have shaped ourselves, we get an answer, but we also lose a rep. Just as calculators changed how students practiced arithmetic, AI changes how adults practice analysis. The risk is not that we stop thinking altogether. The risk is that we stop rehearsing the moves that keep us strong, the same way a runner who skips easy, routine miles loses base fitness and then struggles on race day.

There is a second twist. The companies building AI tools also shape the incentives and the context in which we work. Microsoft’s research arm produced the critical thinking study, and the company is investing at a scale that reorders infrastructure. It has publicly detailed tens of billions in planned spending on AI data centers for the current fiscal year. That level of investment accelerates adoption across industries. It makes AI cheaper, more available, and more embedded in daily workflows. The more AI is everywhere, the more our habits will be tested.

So, what should we do with this tension. The answer is not to reject AI. The answer is to use it in a way that leaves our judgment stronger, not softer. A few lifestyle shifts help, and they fit in the cracks of a normal day.

Turn prompts into thinking drills. A good prompt is a hypothesis. Rather than “Write my email,” try “Draft two versions of this email that argue for X, one concise and one diplomatic, then list three weaknesses in each.” You have asked the tool to produce, but also to expose tradeoffs. Now you can compare, revise, and decide. The act of choosing becomes a workout, not an afterthought.

Practice verification as a ritual, not a chore. When AI hands you a fact, treat it like a note passed in class. It might be correct, it might be confident and wrong. Build a simple routine. Open one authoritative source and one source you would not usually pick. Cross-check dates, names, and numbers. If the task is higher stakes, add a third source. Place a short time box on the ritual. Most checks will take a few minutes, not an hour, but the habit matters. The Microsoft and Carnegie Mellon study observed that workers often used critical thinking to verify outputs against external references. Lean into that impulse. Make it consistent.

Use tiers of trust: Not every task deserves the same scrutiny. Decide your tiers in advance. For low-stakes drafting, ask AI to generate and then you fix tone and details. For medium-stakes analysis, ask AI to outline competing angles, then you write the synthesis yourself. For high-stakes decisions, treat AI as a stakeholder you must challenge. Ask for a counterargument, sources, and uncertainty ranges. Then you go get the raw materials and judge them.

Write your reasoning, even briefly: You do not need a lab notebook. You need a sentence or two that captures why you chose a path. “Chose Option B because it fits user constraints A and C, and the downside D is manageable with step E.” Those micro-notes build a trail. They also force a moment of reflection before you click send. Reflection is the gym of judgment.

Schedule manual reps: Pick one recurring task that you often hand to AI. Do it manually once a week. Summarize a report yourself. Draft the first paragraph yourself. Sketch the data story yourself before you ask for a chart. This keeps pattern recognition fresh. It is the equivalent of playing scales if you are a musician.

Invite disagreement on purpose: Ask a colleague to be your designated skeptic on projects that matter. If you are solo, ask AI to generate the strongest opposing case and to cite it. Then test those claims in your own sources. Disagreement focuses the mind.

Mind the vibe, not just the output: People often describe a sense of ease when AI does the first pass. Ease feels good. It can also be a signal to slow down. If the output feels too ready, look for the seam that is most likely to split. Ask, what assumption here would embarrass me if it broke. Check that one.

Keep curiosity alive off-screen: Read outside your lane. Cook from a new cuisine. Learn a physical skill that requires feedback, like a dance step or a tennis serve. Cross-training keeps your brain adaptable. When work tasks get smoother, make life a little rougher in the best sense.

A common objection is that many tasks really are routine. Why waste scarce attention on another paraphrase or another status email. The answer is to be intentional. Save attention where it is wise, then spend it where it matters. The danger the researchers describe is not any single use of AI. It is the drift, the long glide down the slope of convenience. The goal is not to fight every slide. The goal is to build guardrails so you stay on your feet.

There is also a social angle worth naming. If experienced professionals stop practicing the basics, they become less able to coach juniors. Teams then rely on the same tool to both do the work and to teach the work. That cycle weakens bench strength. The result is an organization that is fast on a sunny day and brittle when storms arrive. The study’s “exception-handling” warning lives here. When systems fail in a new way, you need people who remember how to reason from first principles. That memory is easier to keep if you never stopped using it.

None of this denies the upside. AI can accelerate analysis and unlock time for creative work. Leaders across tech and policy circles are debating the scale and timing of labor disruption, yet many also point to the potential productivity boost. Both can be true. The challenge for individuals is to capture the upside without outsourcing the core. That is a lifestyle skill. It looks like small habits repeated. It sounds like better questions. It feels like a bit more friction today so that tomorrow’s self carries more strength.

Here is a final way to frame it at home. Imagine you want to stay fit. You buy a top-of-the-line treadmill. It adjusts incline automatically, recommends workouts, and even keeps your stride efficient. It is a wonderful tool. If you let it carry the run for you, you will become great at starting the treadmill. If you use it to push your own lungs and legs, then step off and do hills outside once a week, you will stay a runner. AI is the same. If you let it, it will make you great at starting prompts. If you use it to push your mind, then step off and do some hills on your own, you will stay a thinker.

The study that sparked this conversation is not a verdict. It is a mirror, held up at a moment when AI is spreading through every corner of the workday. Look in that mirror and decide how you want your mind to look one year from now. Set rituals that serve that goal. Keep the machine close. Keep your judgment closer.