King's College London

King’s is one of 16 UK start-ups and universities chosen by the Department for Education to develop cutting-edge AI tools for the classroom. Professor Yulan He will develop a platform for teachers capable of delivering personalised assessment and feedback to their students studying A Level science subjects.

The project is funded through the Department for Education’s £1 million investment in AI, which is part of the government’s recently announced plan to roll-out AI across the public sector to boost growth and deliver services more efficiently.

The five-month project will see Professor He working with AQA, building on a tool she previously created for them that uses AI to mark exam scripts. The new tool not only aims to reduce teacher workloads by automating marking and feedback, but also to benefit students by offering personalised feedback, aligned to the KS4 science curriculum.

A Professor of Computer Science in the Department for Informatics, Yulan He said, "we want to go beyond simply sharing the correct answer, to providing highly personalised feedback, tailored to the learner’s knowledge level.

"It has been widely recognised that personalised feedback has a positive impact on learner performance, and the tool will support learners to better understand their misconceptions and errors. This will be particularly helpful for students with diverse learning abilities."

The AI will be able to spot patterns across learner performances, indicating to a student and their teacher where the student is making persistent errors. The teacher will then be able to extrapolate this data across the whole class and make inferences on where their class is losing marks and where to make interventions.

Teachers will also be able to interact with the AI via a chat interface – like ChatGPT, asking the tool to make modifications, and even asking questions about their students, with the AI offering feedback on the class.

In order to mitigate the problem of hallucinations – the AI providing false or inaccurate information, the tool will be developed using vigorous verification methods, and will be thoroughly tested by teachers and learners. Part of the mitigations include a truthfulness scoring system, where the AI indicates to teachers, as a percentage, how confident it feels about the accuracy of the answer it provided, with teachers only needing to cross-check occurrences where the score is low.

The platform will be based on a tool Professor He developed with AQA thanks to funding from the EPSRC Impact Acceleration Account, which provides explainable pupil answer scoring. In the AERA Chat platform, teachers can upload questions, key answer elements, and marking rubrics. It then marks pupil answers in bulk or individually, generating both marking decisions and rationales.

Professor He said, “This project directly addresses the challenge of improving formative assessment and feedback in secondary school science education using AI. The personalised feedback tailored to specific to students’ learning goals and knowledge levels will promote a deeper understanding in science subjects, and free up time-poor teachers by automating a resource intensive process.”