AI was promised as an expert in every pocket, an assistant and tutor on call at all times. Research on what months and years of use do to people finds it can act like a prosthesis instead, one that saps the strength of the person relying on it and changes the way they work.
It starts with what the models say. A 2025 paper called Artificial Hivemind gathered 26,000 real, open-ended questions, the kind with no single right answer, and found that different AI models give strikingly similar responses to them. Someone who switches between ChatGPT and Claude to hear a range of views is steered toward the same narrow band either way.
The same narrowing shows up in science. A study of 41.3 million research papers, published in Nature, found that scientists who use AI publish 3.02 times as many papers, receive 4.84 times as many citations and become research project leaders 1.37 years earlier than those who do not. Across science as a whole, though, AI adoption shrank the range of topics studied by 4.63% and cut scientists’ engagement with one another by 22%.
Individual skills can weaken too. At four endoscopy centers in Poland, doctors who had been working with an AI tool that spots growths during colonoscopies were measured on the procedures they did without it. Their rate of finding precancerous growths fell from 28.4% before the tool arrived to 22.4% after.
Confidence can rise while the skill falls. In a randomized trial with 111 medical students, those shown a wrong AI explanation diagnosed correctly 9.2% of the time, against 21.0% for students given no explanation at all. Any AI explanation, right or wrong, raised the students’ confidence, and among those given wrong explanations, confidence no longer told right answers from wrong ones.
Skills like these work like a muscle. They weaken without use and cannot be put away in a closet and taken back out when needed. A concert pianist can come back from a six-month break, but a six-year break ends the career.
That raises the question of cost. When a company requires its people to work with AI and their skills wear away as a result, the company has taken something from them, and none of the studies puts a figure on what that loss should cost it.





