Why I built GCSE Maths AI

About Dave

GCSE Maths AI was developed by Dave Elliman, who was Professor of Applied Computer Science at the University of Nottingham. He later moved to a high-tech start-up carrying out research in machine learning and AI, which expanded ten-fold while he worked there. He now leads Neuro-Symbolic Ltd, the company behind this app.

The application was designed in the light of his long experience as a teacher and tutor. Two episodes in particular shaped it.

Four Monday evenings

Dave attended a parents' evening when his son James was in Year 10, with GCSEs not far off. James's maths teacher said that James was very poor at maths and predicted a grade 3 out of 9, though a 4 was possible if he did his revision.

Dave was appalled. James had always seemed bright and quick to learn, and a grade like that would limit his chances of further study and his career prospects. So they made a plan. On each of the four Monday evenings leading up to the exam they would go to the gym, have a meal, and then sit down together and work through one of the last four years' exam papers.

The first evening was slow going. Each question needed explaining, and there were big gaps in James's understanding. But he listened and caught on fast, as his dad had expected. The second paper went more smoothly, and James's confidence grew. By the third he could recognise the types of question and the right approach to each one. On the fourth he needed little help and moved at a pace his old dad struggled to keep up with.

He got a top grade in the exam.

Dave was left with a clear impression: if about six hours of tuition could do more than five years at school, maths could be taught much better than it usually is.

A lesson from a hundred programmers

Some years earlier, Dave had been asked to take over a course in C++ programming for a class of about a hundred students. It ran over a semester with lectures on Monday and Wednesday and a weekly exercise, handed in by five o'clock on Friday and returned, marked, the following Friday. A two-hour lab session once a week, staffed by the lecturer and four demonstrators, was the only place to get help. The course was unpopular, and only just over half of the exercises were ever completed.

He was also told that the demonstrator support would not be available. Marking a hundred programming exercises every week would have swallowed most of his time and left little room for research. It is also a peculiarly tiresome task, one that makes you feel you have turned into a compiler.

Faced with this, he built a computer-aided teaching system called Coursemarker to handle the exercises. The key innovation was a series of unit tests for each exercise, designed to diagnose all the common faults in the programs the students wrote. Together with standard tools that checked style, grammar and correctness, this meant students got almost instant feedback the moment they submitted their work. They were allowed five attempts before the deadline, with the last one counting for marks.

The effect was dramatic. The proportion of exercises successfully completed rose from just over 50% to 99%, and student satisfaction rose with it. Dave was even criticised because the marks were too high. It felt like a highly successful innovation, and the reasons were clear:

• Students could get help and feedback the moment they hit a problem. Before, they had to wait for the lab session and join a queue.
• The feedback was well informed. Before, it had often come from demonstrators who were not C++ experts.
• It arrived while the work was still fresh. Feedback that comes a week later is next to useless as a teaching aid.

Rapid, well-informed feedback that left no student stuck and demoralised was absolutely key to teaching. That is the idea at the heart of GCSE Maths AI.

What this means for your child

A good tutor does what Dave did on those Monday evenings: finds the gaps, fills them in the right order, and stays beside the pupil while they work, giving them time to think but never leaving them stuck. GCSE Maths AI is built to do the same, for every pupil, at any hour, at a fraction of the cost.