An AI Tutor for Every Child

  • 26 February 2024
  • Business, Education, Technology
  • Mr Ranulph Tees

Ranulph Tees is co-founder of AI-based Qual Research platform, Survey Mind. Having worked in Artificial Intelligence for many years and building on MSc research with the Centre for Educational Assessment at Oxford, Ranulph has experience building AI-based assessment systems for primary-age Mathematics education. He previously built Cavendish Milton, a tutoring agency focused on assessment-for-learning techniques.

Meet James

James struggled with Algebra. The combination of Covid, a Year 9 with a series of substitutes, and a Year 10 with a ‘Student-Teacher Personality Mismatch’ meant that James entered Year 11 in the bottom set for Maths, far behind where he should be. This was despite a natural intelligence and core numeracy skills. With GCSEs around the corner, and a passing grade in Maths essential for a good sixth-form, things were looking bleak for James.

As the owner-manager of a Tutoring agency, you might expect me to say that James found a tutor, he turned a corner, and that everything was rosy. But that was not the case. James’ family was not in the minority of families that could afford regular, private tuition. His family was in the majority, for whom a personal tutor is an expense too far.

James could not make a 4 (passing grade) in his Maths GCSE, and with that failure went James’ chances of a good sixth-form. He made it into a respectable FE college, but he was neither technically nor vocationally minded. He was in the wrong place for someone with his skills and interests to thrive. At just 16, James’ life prospects were considerably worse than they could have been.

Bold Predictions

We will return to James and the millions of kids like him shortly. For now, let’s talk about the future of AI. Predictions are a fool’s game, but one can make the game interesting by restricting oneself to predictions that are bold and clear, measurable and falsifiable. The following should tick those boxes:

I predict that by 2035, every school-age student in the UK with an internet connection will have their own Personal Tutor who will:

  • Deliver unlimited one-to-one time
  • Know the syllabus like a world-expert
  • Work to evidence-based pedagogical principles
  • Understand the student’s educational history in full
  • And be infinitely patient, unfailingly polite, and responsive to feedback
  • Plus, they’ll always turn up on time, and never miss a session

The caveat, of course, is that this personal tutor will be entirely artificial.

It will be a complex system of inter-linked AI models. At its core, an enormous large-language model that makes GPT-4 look retro. On its surface layer, a naturalistic text-to-speech output model and synthetic face, with an input speech-to-text model listening to the student, with a video/audio analytics engine monitoring the student’s learning in real-time.

Is this just hype? Yet more bilge spewed up for a start-up’s latest press release as they seek more funding?  Not in the least.  I speak as an old-fashioned ‘personal’ tutor, and I foresee the death of this model almost entirely. Here is my reasoning:

Firstly, I am an Educational Researcher by training. My MSc thesis at Oxford’s Centre for Educational Assessment (OUCEA) focused on how intelligent computerized assessment in Maths (pre- and proto-AI) was the single best thing, short of one-to-one tuition, to boost subject attainment. AI-enabled learning and assessment will accelerate this exponentially.

Secondly, I work in an AI start-up (albeit not one focused on education, but on healthcare research), and every day, I see how powerful these AI models are. ChatGPT hit the public consciousness, but it is to AI what Netscape Navigator was to the 90’s internet. The ‘iceberg’ metaphor is overdone; more apt is viewing ChatGPT as a bright red, attention-grabbing toad stool. Buried several feet beneath it, stretching out for hundreds of square miles, is a fascinatingly complex mycelium web, sharing and processing information in ways we simply don’t yet understand.

And I temper my prediction to the year 2035, not because the technology is not yet powerful enough; it already is. But because we don’t yet know how to make full or efficient use of it, and we’re struggling as a society to adapt to it. It is also fearsomely expensive right now, although marginal costs will fall precipitously over the coming decade.

Regulation? No – Public Provision!

Much of the civic discussion around AI and how to adapt best to it focuses on regulation. Doomsters foresee chaos, and want draconian enforcement of narrow rules now. Accelerationists want free rein with open-source builds, training and deployments. My instinctive position is closer to the Accelerationists, but both extremes miss the point. AI will do incredible things, incredibly good and incredibly evil, whether it is regulated or not. But AI's capacity to do unrivalled good for all of humanity will come from making it a public good.

Which is why I go beyond simply making my prediction; I make the following active recommendation:  AI-enabled personal tuition should be provided free to all students at the point of delivery, paid for by the state as part of the Education budget.

In such a world, James would face school and exams on something approaching a level playing field. This intervention would drastically reduce the disparity between James and his privately educated (and privately tutored) peers. And as a society we could take seriously that old cliché, and genuinely build an education system that sees every child reach their full potential.

A bold, futurist UK Government would take the lead on this. They owe it to James to try.