Alpha School’s expansion of its AI-first education model into public schools in Texas and South Carolina is prompting renewed questions over how much of the school day should be handed to software. Experience from a tutoring platform suggests that even highly capable digital tools may struggle to keep pupils engaged without a human being responsible for their progress.
The warning comes from the person behind Tutorela, a US tutoring marketplace that combines human tuition with AI support between lessons. Its experience followed an attempt to build a comprehensive digital maths system for students using a separate tutoring marketplace in Israel.
That platform offered more than 10,000 exercises, detailed written and video solutions, progress tracking and rewards. An AI assistant was later added so students could receive help with any step at any time, free of charge.
Yet only about 10% of students used the app independently, and most of those who started did not continue. The exercises and explanations were not changed; instead, the company altered who was responsible for directing the learning.
Human tutors drove engagement with AI learning tools
Tutors began selecting work suited to each pupil, assigning it and checking on progress at the next lesson. More than half of the students who received assignments from a tutor completed them, compared with about 10% who used the app on their own.
The experience is echoed by a two-year randomised trial involving 18 middle schools in Tennessee and Khan Academy’s Khanmigo AI tutor. Nearly every pupil tried the tool, but the median student messaged it on only a third of the days they practised, and in just 17% of sessions in which they made a mistake.
Most interactions consisted of one- or two-word answers or clicks on suggested prompts. The trial found that pupils’ maths gains were similar to those achieved through Khan Academy practice without AI, while Khan Academy has said that only about 15% of students with access to Khanmigo engage with it.
“It may be that human attention is the key ingredient,” the researchers wrote.
The argument is not that AI has no place in education. Used under a tutor’s direction, it can answer questions outside lesson hours, explain a difficult step repeatedly and provide more practice than a human could mark alone.
That arrangement can leave a tutor with a better-prepared pupil while giving the student help at the moment they are stuck. But removing the human from the process risks preserving the explanations while losing the reason for a child to complete the work.
What AI-first education models need to address
The central issue for schools considering AI-heavy systems is therefore not simply whether the software works. It is who will ensure that each pupil uses it, sets appropriate goals and returns to review the results.
Without that accountability, the experience described by the tutoring platform and the Tennessee trial points to a pattern of broad access but limited engagement, with students able to stop using the tools without anyone noticing.
For parents, the same principle applies: AI tools can support learning, but they should be paired with a tutor, teacher or parent who sets the direction and follows up. The proposed model is not teachers replaced by AI, but teachers supported by it between lessons.