AI tutors are not self-driving education. Two years of access exposed the gap between availability and actual learning: scientists offered Khan Academy’s Khanmigo generative AI tutor to students across 18 Tennessee middle schools, and most students barely used it as intended.
Access was nearly universal, but engagement was thin. Students often sent off-topic messages or tried to coerce the tutor into giving correct answers instead of using it to work through problems.
That result does not make the technology useless. Students who used Khanmigo progressed faster in math than a control group that did not use the AI, showing that the tool can help when students engage with it as a tutor rather than a vending machine for answers.
Philip Oreopoulos, a University of Toronto researcher, described the problem plainly: “If a keen, engaged student wants to use the technology as a good tutor, it’s ready to go. The issue is that most students don’t use it that way.”
His other warning is less flattering to the grand promises surrounding education technology: “The same technology that can be used as a good tutor, can also be used to make your life easier. But making your life easier equals less effort, less learning.” The shortcut is not a bug from the student’s perspective. It is the feature they found first.
Human guidance remains the missing layer
Stanford researchers concluded that AI is not a fix-all, and that its effectiveness depends on human guidance. The quality of the software matters, but so does the way teachers or parents integrate it into learning.
“The effectiveness of AI-led tutoring depends as much on integration, such as by teachers in classrooms or parents at home, as on the quality of the software itself,” the researchers said. Their strongest finding points away from the fantasy of replacing educators: “The strongest results so far come from AI tools built for the human tutors, not the students.”
That approach is visible at Cheshire Academy, where the school offered teachers and students tools including ChatGPT, Perplexity, and MagicSchool. The school trained staff on prompt crafting, the limits of AI, and the risk of incorrect or biased responses.
Teachers at Cheshire Academy use AI to prepare class materials, plan lessons, and create grading rubrics. MagicSchool can generate questions, assignments, grading rubrics, presentations, lesson plans, and administrative reports from prompts; teachers can specify the grade level, choose question types, and include documents.
Some teachers prefer general-purpose tools such as ChatGPT for administrative tasks instead of MagicSchool. The software menu is expanding, but the basic challenge remains stubbornly human: deciding when a generated answer helps and when it replaces the work students need to do.
Schools are teaching students to question the tool
Cheshire Academy has not treated AI as a permission slip for unlimited automation. The school developed assignments that ask students to reflect on AI’s role in learning, including editing homework with AI and grading AI-assisted assignments.
It also created a Student AI Council to promote reflection on healthy AI use. Assignments receive traffic light labels that show whether AI use is allowed, restricted, or prohibited—a simple system for a problem schools have struggled to explain without producing a policy document long enough to qualify as homework.
Teachers still hesitate to use AI for student feedback because of concerns about quality, personalization, accuracy, and privacy. Those concerns fit a wider problem: MIT Technology Review reports that many teachers feel confused about how to handle AI in education.
The demand is already here. Seventy percent of teens use AI for schoolwork, yet most students do not use classroom AI as intended, choosing off-topic activity or shortcuts instead. Giving students access is easy; building habits that turn access into learning is the expensive part.
There is also a basic access problem that no clever prompt can solve. UNESCO reports that around 2.6 billion people lack internet access, which limits educational opportunities before an AI tutor enters the picture.
The lesson from Tennessee and Cheshire Academy is direct: AI can support learning, but it cannot design the learning culture around itself. Students need structure, teachers need training, and schools need clear boundaries—otherwise the tutor becomes another shortcut with better branding.
Based on
- Scientists Gave 12-Year-Olds Access to an AI Tutor and You Probably Already Know What They Did With It — futurism.com
- 10 AI Updates That Put Teenagers In Charge Of Their Own Learning — forbes.com
- How to encourage smarter AI use in the classroom | MIT Technology Review — technologyreview.com
- Rethinking AI In Education: From Digital Access To Human Empowerment — forbes.com
