When Paddy Rodgers, director of the Royal Museums Greenwich, spoke about the dangers of AI, he stood before an exhibition about astronomical discoveries — things humanity had gained over centuries through mistakes, doubts, and repeated observations. His thesis: it is this process of searching, not the result, that shapes the ability to think.
It's not the answer that's harmful — it's the answer's immediacy
Rodgers compared modern chatbots to Wikipedia — but not in favor of the former. Wikipedia, despite all criticism, provided links to primary sources: readers could verify where the information came from and how much to trust it. An AI response doesn't show this chain. It simply exists — convincingly formatted and devoid of a trace.
«Excessive dependence on AI can reduce the culture of questioning, evaluation, and curiosity that fuels innovation and expertise»
Paddy Rodgers, director of the Royal Museums Greenwich
This culture of questioning is not an abstraction. A study published in the journal Societies in January 2025 involving 666 people of different ages and education levels documented a statistically significant negative correlation between the frequency of using AI tools and critical thinking indicators. The mediator between them proved to be so-called cognitive offloading — the habit of delegating mental effort to an external tool.
Younger people are more dependent
Researchers separately noted that younger participants demonstrated higher dependence on AI and lower critical thinking scores compared to older participants. This doesn't mean older people are smarter — they simply formed thinking skills before the age of chatbots and now use AI as a tool rather than a substitute for thought.
A separate study published in Frontiers in Psychology in 2025 showed that prolonged contact with AI during learning leads to memory decline, even if students had been tested at the beginning. The effect doesn't appear immediately — it accumulates.
Where the line between tool and prosthetic lies
An Oxford Brookes University lecturer formulated this precisely: AI is useful when a person directs it toward more important aspects of learning. The problem begins when a person simply «outsources their thinking» — and then the limitations of the technology become the limitations of the person themselves.
- AI doesn't show unexpected discoveries that emerge during manual searching
- An AI response disconnects you from primary sources that can be verified
- Cognitive offloading reduces not only effort — but also the capacity for effort
Rodgers reminded us of something obvious but important: no scientific discovery in history was made solely thanks to technology. Behind each one stood a person who didn't know the answer — and that's precisely why they searched for it.
If cognitive offloading does indeed accumulate with age of AI use, then the first generation that learned alongside chatbots from the start of school will show the real scale of the effect in approximately 10 years — and then the question will not be whether AI is harmful, but whether anyone will remain in the education system capable of evaluating it without its help.
