Tag: DfE

  • A Person on the Other Side of the Screen

    Dr. Lokman Khan
    16 November 2026

    I have taught chemistry in England since 2006. In the evenings I also tutor, and every lesson I give is online. A camera, a shared whiteboard and a student stuck on moles or bond enthalpies. It is unglamorous work. I am now convinced it is also one of the few things in English education that reliably works. That is why the government’s new enthusiasm for AI tutors worries me more than it excites me.

    Start with the evidence. The DfE cited EEF research showing that one-to-one tuition can accelerate learning by around five months on average. Few interventions come close. Parents understand this instinctively. The Sutton Trust’s Private Tutoring 2026 report found that 29% of secondary school students in England and Wales have had private tutoring at some point, up from 18 per cent in 2005. In London 45% of pupils had received private tutoring, compared to 27% in the rest of England.

    Online delivery is what turned tutoring from a London habit into something closer to a national service. A pupil in rural Lincolnshire can now learn A level chemistry from a specialist in Sheffield. In a Save My Exams survey of more than 2,000 UK students, 71% of tutored students receive at least some of their tutoring online. Specialist knowledge no longer depends on postcode. That is a real gain, and it is worth defending.

    But access still depends on money. Sutton Trust data show that 30% of pupils in the best-off households have accessed private tuition, compared to 23% in the worst-off households. The gap looks modest until you ask who needs help most. The DfE itself says just one in four disadvantaged children achieve a pass in English and maths at GCSE at grade 5 or above, compared to over half of their peers. Their families are the least able to pay £40-£100 per hour for a private UK tutor.

    The state once tried to close this gap with people. The National Tutoring Programme, launched in 2020, saw over £1 billion made available and more than 6.1 million tutoring courses delivered. It was clumsy. It was bureaucratic. Many schools resented the paperwork. Yet it put trained adults in front of struggling children. Then the subsidy was cut year by year, and the programme has now ended. Ministers said tutoring would become a “permanent feature” of the system. The Sutton Trust later found that three in five state schools (58 per cent) had reduced their tutoring offer from the previous year.

    Now comes the replacement. In January the DfE announced plans to develop AI tutoring tools for up to 450,000 disadvantaged Year 9–11 pupils, with AI tutoring trials starting autumn 2026 and rollout by end of 2027. Whatever the final bill, it will be a fraction of what human tutoring costs. That is the attraction. The direction of travel is clear. The rich keep their tutors. The poor get software.

    I am not hostile to AI. I use it to prepare resources, generate practice questions and test my own explanations. It saves me hours every week. But a tool that helps a tutor is not a tutor. The research here is sobering. A field experiment with nearly a thousand high school maths students, published in PNAS in 2025, gave some pupils a standard GPT-4 chat interface. Their performance on assisted practice problems improved by 48%. When access was taken away for an unassisted exam, they performed 17% worse than the control group who never had access. Without guardrails, students used GPT-4 as a “crutch” during practice and then performed worse on their own. A version with safeguards, which offered hints instead of answers, largely avoided the damage.

    The lesson is not that AI is useless. It is that AI is only as good as the structure around it. Who provides that structure for a fourteen-year-old on free school meals, working alone at the kitchen table at nine o’clock at night? Not the chatbot. A motivated pupil with a watchful parent will use an AI tutor well. A disengaged pupil will use it to fetch answers. The children this scheme targets are, by definition, more likely to be the second kind.

    What a good tutor does is mostly invisible. I notice when a student says “yeah, I get it” in a voice that means the opposite. I notice that the real problem with an equilibrium question is weak algebra left over from Year 9. I notice the student who stopped trying because someone once made him feel slow. An algorithm can mark an answer. It cannot yet read a face. Diagnosis, trust and accountability are the work. A pupil turns up to a session because a person is waiting. Nobody feels guilty about letting down an app.

    The caution is not mine alone. In July, experts told an MPs’ committee that some tools can leave teachers ‘bogged down’ with ‘low-level’ work. Schools Week reported experts warning that evidence on AI provision is ‘in its infancy’. Google is no longer helping the government develop AI tutoring tools for schools. None of this describes a mature product ready to stand in for human support for nearly half a million children.

    So here is my position, plainly. Online human tutoring is the most effective equaliser England has, and AI should work behind it, not in place of it. Use AI to cut the cost of each human hour. Let it mark practice work, build materials, track gaps and free tutors to teach more pupils. Then fund a permanent, ring-fenced tutoring programme for disadvantaged pupils, delivered online by qualified subject specialists. Many of them are already teachers, working in the very schools that need them.

    The alternative is a two-tier system dressed up as progress. Wealthy families will buy both, the chatbot for homework and the human for everything that matters. Poorer families will be told the chatbot is the same thing. It is not. If we believe tutoring works, and the evidence says it does, then the children who need it most deserve a person on the other side of the screen.