Generative AI tools can limit users’ critical thinking when doing tasks. People using generative AI also think less critically when they trust the AI to do a task, such as developing an argument for a paper or presentation. The researchers behind the findings say the solution is to adapt the technology, rather than to limit its use.
at Microsoft Research and his colleagues asked 319 people to take part in a survey. The participants, who were recruited through the Prolific crowdsourcing platform, were asked to share three examples of how they used generative AI tools such as ChatGPT in their work. They were then asked whether they had engaged in critical thinking while doing each task.
If they answered yes, they were asked to provide an example as to how they did so, the results of which were analysed. The respondents were also asked to rate how much effort the activity required compared to doing it without AI, on a five-point scale, and their confidence in the resulting work.
Advertisement
For around four in 10 tasks, the participants reported using no critical thinking whatsoever. The researchers suggest that while using AI could help enhance the speed at which people learn, the ease with which generative AI tools create and present the content means they might not learn as deeply. “When studying human behaviour, seemingly opposing ideas can both be true,” said Tankelevitch in a written response to questions.
“This underlies a deep and long-term suspicion with cognitive technologies,” says at the University of Salford in the UK. “That their use will diminish human skill sets, whether it is a calculator reducing our ability to do simple mental calculations or Google Maps removing the need to navigate.”
Even as AI improves, this raises important concerns. If users have less and less critical thinking skills, they may not be able to identify whether the AI is correct, or know how to proceed without AI when they recognise it is wrong. But there may be checks and balances in place still, the authors of the study say. “Our survey-based study suggests that when people view a task as low-stakes, they may not review outputs as critically,” said Tankelevitch. “However, when the stakes are higher, people naturally engage in more critical evaluation.”
Tankelevitch defended the research and its recommendations: “All of the research underway to understand AI’s impact on cognition is essential to helping us design tools that promote critical thinking. Deep reasoning models are already supporting this by making AI’s processes more transparent – making it easier for users to review, question and learn from the insights they generate.”
Others aren’t as convinced. “Writing is about developing critical thinking, and we shouldn’t delegate that to anyone or anything, let alone machines that are designed by tech companies with clear financial interests,” says at the University of Oxford. She says this could have bigger societal implications. “Democracy is a kind of conversation. If we externalise dialogue to machines, we lose our place in the dialogue.”
Proceedings of CHI'25 conference
