Is AI Making Us Worse at Thinking? What the Evidence Shows


Short answer

Is AI making us worse at thinking?

Nobody has shown that AI is making people worse at thinking over the long run. Surveys show people expect it, which is a forecast. The first randomized trials found something narrower: after a short stretch of AI help, people did worse and gave up sooner once the help was taken away.

Picture a woman writing a birthday note to her sister. This is an illustration, not a real case. She asks a chatbot for help, and in seconds she has three warm paragraphs. She sends one. Later she wonders whether the words were still hers.

That small doubt is the big question in miniature. Many people now ask whether Artificial Intelligence (AI) is making us worse at thinking, at making things and at being close to each other. Most of what they hear is a forecast. A little of it is a measurement.

Is AI making us worse at thinking?#

Not in any study that settles it. Not over the long run. What exists is a large body of opinion and a small body of experiment, and the two are easy to blur.

The opinion is everywhere. In a February 2026 survey by Pew Research Center, the nonpartisan research group, four in ten U.S. adults said AI will have a negative impact on society over the next 20 years. Far fewer expected a positive one.

That number describes a fear, and fear is information. It came from answers people gave online or by phone, not from a test at a desk. It tells you what people expect. It does not tell you what happened to anyone’s mind. A forecast is not a finding.

The experiments are newer. They are smaller. They are more careful. They do not ask people what they fear. They give people a task, take the tool away, and measure what is left. Their answer is more specific than the headlines, and more useful.

Read Also:  What Medicare Actually Pays For

Does AI make critical thinking worse?#

In short experiments, leaning on AI for answers made people worse at the same task once the AI was gone. That is the clearest measured result so far.

A team of researchers from Carnegie Mellon University, the University of Oxford, the Massachusetts Institute of Technology (MIT) and the University of California, Los Angeles ran a set of randomized trials with 1,222 participants.

They published the results as a preprint, meaning a paper shared before formal review: AI Assistance Reduces Persistence. People solved fraction problems or answered reading questions. Some had an AI assistant and some did not. Then everyone worked alone.

The researchers say that just 10 to 15 minutes of AI help led to significant drops in independent performance and in persistence, the willingness to keep trying. People who had used the assistant solved fewer problems on their own and skipped more of them.

The detail that matters most is how people used it. In the second experiment, 61 percent of the people with the assistant said they mainly used it to get answers. Those who used it for hints or clarifications held steady against their own earlier scores.

The paper says this split was self-reported, so it does not prove cause. Still, the drop sat mostly with people who used the tool for answers.

What do people say AI does to their creativity and relationships?#

Chatbot users say chatbots help creativity more often than they hurt, and that it mostly leaves their relationships alone. Pew asked this in a survey of 5,119 adults in February 2026, and the creativity and relationship questions went to the people who use chatbots.

Chatbot users asked whether chatbots help or hurt. Source: Pew Research Center, February 2026.

Among chatbot users, 43 percent said chatbots help how creative they are, and 22 percent said they hurt. On relationships with other people, 13 percent said help and 15 percent said hurt. Most, 72 percent, said chatbots neither help nor hurt their relationships.

Age matters. Pew also reports shares among all U.S. adults, counting non-users as their own group. On that base, adults under 30 were about evenly split on creativity, while older groups leaned toward help. It also found fairly small shares of people 65 and older who think chatbots help their creativity at all.

Read Also:  Who Gets It, Who Pays

These are feelings. They are reports of how people feel at a keyboard, a kitchen table or a desk, written on a survey form. A person who says a chatbot makes them more creative may be right, or may be enjoying a faster draft.

Neither the survey nor anyone answering it measured creativity. Like a bathroom scale that reads your mood instead of your weight, a survey on this question measures the wrong thing well.

Is AI damaging to the brain?#

None of the studies in this article shows lasting damage. The study behind most of the alarming headlines is small, and its own authors frame it as a first look.

Researchers at MIT ran a study, Your Brain on ChatGPT, that asked people to write essays with an AI tool, with a search engine, or with nothing. They measured brain activity with electroencephalography (EEG), sensors on the scalp that read electrical signals. They report that the AI group showed the weakest brain connectivity while writing. The AI group also felt the least ownership of its essays.

Read it for what it is. It is small. A total of 54 people took part, and only 18 came back for the final session. It is a preprint, not a reviewed journal article.

Weaker activity while writing with help is also what you would expect from any shortcut, the way your legs work less on an escalator. It does not show the brain is harmed, only that it was less busy.

The ownership finding is the one worth keeping. It is the birthday note again. When the words come from the machine, they feel less like yours.

What is cognitive offloading?#

Cognitive offloading means handing a mental task to a tool so your own mind does not have to carry it. A grocery list is offloading. So is a calendar alert, or a calculator at the hardware store.

Offloading is old. It is not bad. People have always moved memory and arithmetic onto paper and tools, and gained room for other work. The question is what you lose when you hand over the part that was building the skill.

Read Also:  AI Safety Is Not What They Say. It Is What They Gave Up. · Hanh D. Brown

The randomized trials suggest a mechanism. The researchers argue that when a tool finishes tasks in seconds, unaided work starts to feel slower and harder than it is. Each handoff makes the next one more tempting. They describe it as self reinforcing.

That fits a pattern this site has written about before: people learn from far fewer examples than a model needs, but only if they do the learning. A skill is kept the way a path through tall grass is kept. Walk it, or it closes over.

Where does the real choice sit?#

In what you ask the tool to do. The research points one way: asking for the answer costs more than asking for help along the way.

That is not advice to avoid AI. It is a distinction any user can make on any task. Three kinds of request sit on a line from teaching to replacing:

  • Ask for a hint, and you still do the hard part.
  • Ask for a check of your own work, and you still did the hard part.
  • Ask for the finished result, and you save time but skip the practice.

All three are reasonable choices. Only the first two keep the skill.

The same choice shows up in how people write with these tools. When the person sets out what they want first, the tool serves the thinking instead of replacing it, which is the point of writing the brief first. It shows up for children, too, in what a kid should learn when the tool can do the heavy lifting.

Bring it back to the birthday note. The woman writing to her sister could have asked for three warm paragraphs. She could also have written two clumsy lines of her own and asked the chatbot to help her say them better. The second note takes longer. It is worth it.

Her sister would never know the difference. She would. She always would. The words would still be hers, and so would the habit of finding them.

Source: Pew Research Center, Americans and AI 2026.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top