Some AI tools are carefully designed for education and can help children follow their curiosity, fill in learning holes and help with learning differences. Khan Academy’s AI tutor, Khanmigo, uses vetted education content to coach students on mastering new material without quickly giving them the answer. OpenAI claims study mode in ChatGPT works similarly.
But these tools stop being helpful when they start doing the thinking for children, which is how many young people are using AI. When students put their essay prompts or problem sets into regular ChatGPT and it spits out perfect work, they are shortcutting their learning. If the training wheels on a child’s bike kept the rider upright and pedalled and steered automatically, the child would not likely learn to ride. When students use Gemini or DeepSeek to do their history homework for them, that’s what’s happening.
This is what researchers at MIT recently found when they tested how AI affected writing skills. They split university students ages 18 to 39 into three groups: one wrote with ChatGPT from the start; the second wrote on their own but could use Google search; and the third group was not allowed to use any tools. Later, all of the students revised their writing using ChatGPT to help.
Those who wrote with ChatGPT from the beginning exhibited the worst writing quality and motivation; and as shown from brain activity measurements, parts of their brain associated with learning were less active. They struggled to revise their writing because it was never theirs to begin with. Participants who drafted their work unaided performed best. Given that even well-educated university students are at risk, we should be even more, worried about children who have yet to fully develop their thinking skills.
Learning is hard. Seventh-grade English teachers don’t instruct their students in the art of essay writing in hopes that they’ll create high art. Rather, students learn to organise their thoughts, evaluate evidence, form an argument and articulate a thesis by writing the essay. When AI helps students short-circuit this process, critical thinking skills may fail to develop. In the words of one student participating in research for the Brookings Global Task Force on AI in Education, “If you are letting someone else do the work for you, you are not learning.” Other research not yet peer reviewed suggests that frequent cognitive offloading to digital devices may account for the recent decline in student IQ levels.
Schools are scrambling to figure out how to manage teaching and learning in the new AI landscape. Some are restricting or banning the use of AI, while others have begun to incorporate it more into the lessons. Overall, however, less than 20% of the nation’s teachers reported their school had a formal AI policy, according to a June survey.
But school policy won’t be able to cover everything. On Snapchat, for instance, over 150 million people have used the My AI tool, which can write an essay and do math problems. As one high school senior in Washington, DC, told Brookings: “A lot of schools, including mine, they blocked ChatGPT, but people will go on their phones and use Snapchat AI or Meta AI with like Instagram.”
As we’ve written in our book, parents are already in the dark about their children’s engagement in school, despite their best efforts to stay apprised. (Grades, for instance, give only a partial picture of one’s learning.) AI is making this worse. Researchers found that 22 to 26% of parents of students in secondary school believe that their children use generative AI for education-related purposes, yet other studies suggest the use among secondary school students may be closer to 70%. As one education nonprofit has learned in its surveys, young people admit to using AI a lot, but they refrain from talking to adults about it because they sense their fear, and worry they will be judged.
Rectifying this parental awareness gap is paramount, since our research also shows that families are as influential as teachers and peers in helping young children and teens engage deeply in learning. But parents cannot foster a nation of engaged learners alone. When rolling out AI literacy programmes, which should be vetted by education experts, schools should target parents as well as students. There are materials, such as the AILit Framework, that educators can use to help teach students how to best use AI for their own learning and avoid its negative trappings. Students’ families could be included in these programmes, too.
Parents should demand that commercial AI companies implement and enforce age-verification strategies, similar to what Nebraska and Britain are requiring from social media companies and some other websites. OpenAI’s own policies require parental consent for children 13 to 18 to use ChatGPT, which, if enforced, would increase parents’ awareness of their children’s AI use. But as of now, children easily get around this requirement by entering a false birthday, or use a different chatbot like My AI on Snapchat.
If we don’t help our children to use AI wisely – to elevate ideas, gain skills and build new knowledge – we risk a whole new level of learning loss: a nation of compliant and unmotivated young people who have not developed the muscles to struggle productively, think, work and contribute to our communities. This is the opposite of the human creativity and problem-solving required to navigate the opportunities and pitfalls of our new AI age.
Jenny Anderson and Rebecca Winthrop are the authors of The Disengaged Teen: Helping Kids Learn Better, Feel Better, and Live Better.
This article originally appeared in The New York Times.
Written by: Jenny Anderson and Rebecca Winthrop
©2025 THE NEW YORK TIMES
