When Should Students Use AI?
By Dan Fitzpatrick
MIT and RAND research suggests when students use AI shapes what they learn. Teaching them to think before prompting protects critical thinking as classroom AI use climbs.
By December 2025, 62% of middle, high school and college students were using AI to help with homework. That's up from 48% just seven months earlier. The same students are uneasy about what that habit is doing to them. Two-thirds said using AI for schoolwork was harming their critical thinking, a figure that had climbed from just over half earlier in the year.That tension of answers on tap, but confidence draining away, is the real story of AI in education right now. And a piece published in The Indian Express on May 9 captures it more sharply than most policy documents. Ashish Dhawan and Pramath Raj Sinha, both founding chairpersons of the board at Ashoka University, argue that the future will not belong to students who can use AI tools, but to those who can think beyond them. When answers are abundant, they write, the differentiator is no longer access to knowledge. It is the ability to ask the right question, evaluate the reply, and apply judgment.It is a compelling argument. It is also, until recently, one that rested mostly on intuition. What has changed in the past year is that the research has started to catch up and it is more specific, and more useful to teachers, than the slogan suggests.The evidence behind "the quality of asking"A 2025 study by Michael Gerlich at SBS Swiss Business School, surveyed and interviewed 666 people across a range of ages and education levels. It found a significant negative correlation between frequent AI use and critical thinking performance and identified the mechanism: cognitive offloading, the habit of handing mental effort to an external tool. Younger participants showed the highest dependence on AI and the lowest critical thinking scores.A separate study from the MIT Media Lab, Your Brain on ChatGPT, went further and put electrodes on the question. Researchers led by Nataliya Kosmyna split 54 participants into three groups, one writing essays with ChatGPT, one with a search engine, one with no tools at all, and measured...