With frequent use of neural networks, we become not so much stupid as lazy
A group of researchers from universities in the US and UK conducted a series of experiments whose results are hard to call encouraging. Just ten minutes of working with an AI assistant was enough for people to perform worse on tasks without it, and, even more alarmingly, they lost the desire to even try. The study has not yet been peer-reviewed, but its scale and methodology deserve attention.
How the experiment with the AI assistant was designed
The paper was published on the arXiv platform in April 2026. A total of 1,222 people participated in the series of randomized controlled trials. In the first experiment, 350 Americans were given a set of fraction problems. Half of the participants were given access to a specialized chatbot based on GPT-5 from OpenAI, while the rest solved problems on their own. Midway through the test, the AI group’s access to the chatbot was turned off.
The result was dramatic: after the AI was disconnected, participants in that group began giving significantly more incorrect answers, and many abandoned the tasks entirely. The experiment was repeated with 670 participants with the same result. Finally, a third trial, this time with reading comprehension tasks rather than math, confirmed the same pattern.
The “boiling frog” effect in working with neural networks
The study’s authors use the metaphor of the boiling frog. It’s a well-known parable: if you throw a frog into boiling water, it will jump out, but if you heat the water slowly, it won’t notice the danger. Real frogs, fortunately, don’t work that way, but as a description of human behavior, the analogy proved frighteningly accurate.
The point is that prolonged use of AI gradually erodes motivation and persistence — qualities that are fundamental to learning. A person gets used to receiving instant answers and stops going through difficulties on their own. According to the authors, these effects accumulate, and by the time they become noticeable, reversing them is already difficult.
When AI is turned off, people don’t just give wrong answers — they refuse to try. Persistence drops.
Why 10 minutes with AI changes human behavior
One of the most unexpected aspects of the study is the speed at which the effect manifests. Researchers documented that dependence on AI forms in literally ten minutes of use. This is not a months-long habit but a nearly instantaneous recalibration of expectations.
Scientists suggest the cause lies in the very architecture of modern AI systems. A good mentor doesn’t just give answers — they help you learn, track progress, and sometimes say “no.” AI chatbots are designed differently: they are optimized for instant and complete answers, never refuse (except on safety issues), and don’t consider the user’s long-term goals.
In essence, AI trains the brain to expect that the answer is always instantly available, depriving it of the experience of independently overcoming difficulties. It is precisely this experience, as decades of cognitive research show, that lies at the foundation of learning and development.
Hints instead of ready-made answers: how AI can avoid making users dumber
The study did find one encouraging nuance. Participants who used the neural network to get hints and explanations coped much more easily with tasks after the AI was turned off. Meanwhile, those who simply asked the bot to provide a ready-made answer found themselves in a significantly worse position.
This is a fundamental difference. It suggests that the way one interacts with AI is critically important for the consequences. If you use a neural network as an assistant that guides your thinking rather than as a cheat sheet with ready-made solutions, the negative effect can be mitigated. However, the authors emphasize that even in this case, the long-term consequences remain to be studied.
A threat to education and the “generation without perseverance”
The study’s authors openly warn about the dangers of thoughtlessly integrating neural networks into schools and universities. Mass use of AI in education could lead to the emergence of an entire generation of people who will never know what they are capable of. And this would undermine innovation and creativity on a societal scale.
This is not the only study with similar conclusions. Earlier research showed that the use of AI in schools is associated with poorer social and intellectual development, and children who regularly work with neural networks show worse results on tests.
The problem is compounded by another phenomenon — so-called mental overheating from AI. One study found that about 14% of workers who actively use AI experience cognitive overload: brain fog, slowed thinking, and headaches. Moreover, those who had to simultaneously manage multiple AI tools reported 12% greater mental fatigue.
How AI use affects the thinking of an ordinary user
It’s important to note that the study on perseverance and AI has not yet been peer-reviewed. It is a preprint — a preliminary publication that has yet to be verified by independent experts. This is not a reason to ignore the results, but a reason to treat them with caution.
Nevertheless, the direction of the trend is becoming clearer. A growing body of data, from school studies to worker surveys, suggests that thoughtless use of AI may cost more than it seems. Not in money, but in cognitive skills: the ability to think, try, make mistakes, and keep going.
The practical takeaway from all these studies is this: use AI as a tool for learning, not as a replacement for thinking. Ask for hints, not ready-made answers. And if you notice that without a neural network you can’t even begin to think about a problem, perhaps the water is already warmer than it seems.
