In an experiment conducted last year at the Massachusetts Institute of Technology (MIT), more than fifty students from different universities in Boston were divided into three groups and asked to write essays in the style of the SAT test, on broad topics such as: “Should our achievements benefit others in order to bring us true happiness?” One group had to rely solely on their own mental abilities. The second group had access to the Google search engine to find information. The third group was allowed to use ChatGPT — the powerful artificial intelligence-based language model that can generate complete passages of text.
While the students worked, all three groups wore a device with electrodes on their heads that measured brain activity. The results, according to MIT Media Lab researcher Nataliya Kosmyna — one of the study’s authors — were startling: ChatGPT users showed significantly lower brain activity than the other two groups. Neurological analysis showed fewer connections between different regions of the brain, fewer alpha connections (associated with creativity) and fewer theta connections (associated with short-term memory). Some of the participants who used ChatGPT said they had “no sense of ownership” over the texts they had written and, in one round of testing, 80% were unable to quote a single phrase from what they had “written.”
This is among the first studies to scientifically measure what Kosmyna calls the “cognitive cost” of relying on artificial intelligence to carry out tasks that once required direct human thought.
Another surprising finding was that the texts created with the help of ChatGPT were very similar in content and vocabulary. Although the essay topics were designed to encourage different points of view, using AI led to a homogenization of thought. “The output was remarkably similar, even though the students were different, had come on different days, and were addressing personal or social issues,” Kosmyna said.
On the question about true happiness, ChatGPT users were more likely to focus on personal success or career. In another question about philanthropy, they uniformly argued in favor of helping the less fortunate — unlike the other groups, which also included criticism of the concept of charity.
“There are no divergent opinions when you use ChatGPT,” Kosmyna added.
“It’s like one big average, all at once.”
AI works according to the logic of averages: large language models are trained to find recurring patterns in vast quantities of data. Their responses tend toward consensus — in writing style (full of clichés and banality) and in the substance of ideas. Of course, earlier technologies such as computers or SparkNotes have also influenced the writing process, but AI has an unprecedented power to replace independent thought. Anyone using ChatGPT to write a wedding speech, a contract, or a university essay is taking part in an experiment similar to the one at MIT.
According to OpenAI CEO Sam Altman, we are on the cusp of what he calls a “gentle singularity.” In a recent post, Altman wrote: “ChatGPT is more powerful than any human who has ever lived. Hundreds of millions of people rely on it every day for increasingly important tasks.” In his vision, humans are merging with machines. But the long-term consequences of this symbiosis are not yet known — and if early experiments like this one are any indication, the benefits may come at a profound qualitative cost.
In April, researchers at Cornell published another study that identified similar homogenizing effects from AI. Two groups — one from the U.S. and one from India — answered questions such as: “What is your favorite food and why?” or “What is your favorite holiday and how do you celebrate it?” Some of the participants used a tool that automatically suggested words generated by ChatGPT whenever they stopped writing. The results showed that the responses from both groups became more similar — more oriented toward Western norms. Favorite food: pizza. Favorite holiday: Christmas. Even when chicken biryani was mentioned, the description avoided specific ingredients such as nutmeg or lemon pickles and replaced them with general phrases like “richly spiced flavors.”
Of course, theoretically, a writer can reject AI’s suggestions. But these tools have a hypnotic influence, in which the unceasing flow of ideas replaces one’s own voice. Cornell professor Aditya Vashistha says: “It’s as if a teacher is standing behind you all the time saying, ‘This is the best way.’”
Over time, he says, “you lose your identity, your authenticity, and your confidence in your writing.”
His colleague Mor Naaman adds that AI “works covertly to change not only what you write, but also how you think.”
Mental independence, then, is fading. And this harmless average is not as innocuous as it seems. Writer and journalist Vauhini Vara, in her book “Searches,” notes that the mediocrity of AI texts creates “an illusion of safety and harmlessness,” but in fact “reinforces cultural hegemony.” OpenAI has a clear interest in making its responses as acceptable as possible to as many people as possible — in order to expand the number of users who pay for it.
Many platforms already promise to help us “overcome creative block” and “unleash our imagination,” but studies suggest the opposite.
An experiment at Santa Clara University tested the use of AI in two types of creative tasks: improving products and predicting bizarre consequences. Users who used ChatGPT provided answers that were much more similar to one another than those who used “Oblique Strategies” — a set of enigmatic cards created by Brian Eno and Peter Schmidt in 1975 to stimulate creativity.
Researcher Max Kreminski, who now works for the startup Midjourney, says that users initially contribute original ideas, but over time they shift into a “curation mode” — increasingly accepting the ideas offered by AI. The influence is one-way: “Human ideas don’t have much influence on what the machine produces,” he says. Over time, as the AI’s memory fills up, the model begins to repeat itself and becomes even less original.
The studies at MIT, Cornell, and Santa Clara are small in scale, and the results are only the beginning of a broader picture. Meanwhile, in the Meta AI app you can see content generated by millions of people: a surreal stream of polished images, filtered clips, and texts for everyday tasks like “write a professional email to reschedule a meeting.”
One user, @kavi908, had asked Meta’s chatbot to analyze “whether AI might one day surpass human intelligence.” The response contained only positive scenarios — that AI would improve and humanity would benefit. No scenarios involving failure or risk. It was the average, programmed with a pro-technology bias.
But to believe that answer completely, you would have to switch off all your brain activity.









