ChatGPT helped teachers and data scientists this week to write school behaviour policies, produce newsletters for parents – and even explore how it might tailor revision based on a pupil’s answer to mock exam questions.
But it wasn’t for real. The scenarios were part of a government-sponsored “hackathon” in London.
Tech experts and school leaders discussed how large-language AI models could use education, health and care plan data to generate worksheets based on a pupil’s needs, and act as a personal assistant for language students.
Schools Week joined education secretary Gillian Keegan and academies minister Baroness Barran at the event, held at the London headquarters of Faculty AI, to hear about potential solutions to teacher workload.
AI: A Christmas miracle?
Keegan believes models such as ChatGPT could save teachers from the pain of mock exam-marking and tailor revision work for individual or groups of pupils.
She said teachers had complained that the timing and volume of autumn mock exams left them with “virtually no time” to mark and then set revision for the Christmas holidays.
The hackathon explored whether through AI “we could get these marks and we could personalise for each child…what they still need to brush up on”. Pupils could then use their holiday to “really focus their learning on the bits that are going to help them get a better grade”.
“That was just a very practical [solution] that nobody came here thinking about, but the fact you’ve got teachers, the fact you’ve got computer scientists, you’ve got researchers and experts all together, and they’ve actually created examples of it right here that could be utilised. It’s quite astonishing.”
In one breakout session, school leaders worked with data scientists to discuss whether AI could run question-level analysis of mock exams, identify which pupils or groups of pupils struggled with specific topics, and then suggest interventions.
Enass Al-Ani, the executive principal of Small Heath Leadership Academy in Birmingham, told Schools Week it had “huge potential”.
“Take an exam paper. Mathematics for example. It may have 22 questions, and as a teacher, you want to know which of the questions students were able to answer well, but which questions they couldn’t answer very well.
“We would hope AI would help us to then analyse that for the teachers and create resources … [for] the learners who couldn’t answer those questions.”
ChatGPT on its best behaviour
Schools have to publish swathes of written policies on their websites, ranging from their approach to behaviour and uniform to curriculum and how they support SEND pupils.
At the hackathon, leaders and experts explored how AI could draft and review those policies.
“What the AI was able to do was to be fed examples of existing best practice policies to learn from, and then tailor to the school context,” said Tom Nixon, head of government practice at Faculty.
“And that potentially shortens the time it takes school leadership to create policies, but also reviews, checks quality, checks for adherence to best practice or legislation. So I think that’s really exciting.”
Models such as ChatGPT could also help schools keep staff up-to-date on their policies, according to Melanie Renowden, chief executive of the National Institute of Teaching.
She said it did a “pretty good job” of writing the policies themselves. Delegates then asked it to create “low-stakes quizzing” on the policies to aid with staff training, and presented scenarios to ask how policies should be applied.
“It starts to become a sort of policy digital assistant.”
Parent communication is another possible admin task that could be slimmed down for teachers.
Renowden said after “some refined instructions” to ChatGPT it quickly produced “some really beautiful newsletters” that teachers felt pretty confident about using (after a quick quality-assurance).
Nixon said he saw examples outside education of generative AI “being very powerful for creating marketing content, for creating communications content.
“And the ability to use that and translate some of that, those ideas, means that schools can more quickly, more easily communicate with parents, [and] share what’s happening in the school.”
During the hackathon, pupils from Eden Girls’ School Waltham Forest and Harris Academy Wimbledon discussed with data scientists how AI could help language learning, while leaders worked on developing lesson plans.
AI needs help with lesson planning
John Roberts, the product and engineering director at the Oak National Academy, said it had found “you can only go so far with the underlying models”, which needed to be trained to ensure content is in keeping with the national curriculum and exam specifications.
“What we’ve been looking at is, how do you bring the content that Oak has, and the open licence content that we have, and then feed that in to make sure that the outputs are really aligned to that curriculum. That’s going to be really beneficial I think for others in the sector.”
Jonathan O’Donnell, a computing consultant at the Harris Federation, said he had been working with data scientists to train AI to help with lesson planning.
ChatGPT initially “wasn’t producing the outputs that we were hoping for”. For example, it generated plans based on A-level resources when they wanted it to focus on GCSE.
“So we’re feeding it even more information. But what that really highlighted for us was the fact that we need these models to be trained in subject disciplines for each individual use case scenario for each individual academy.
“That could have potentially a great amount of impact on teacher planning time, adapting lessons for each individual student.”
Better support for SEND
Keegan wants the development of AI in education to help close the attainment gap, and the hackathon explored how this could be achieved for pupils with special educational needs and disabilities.
O’Donnell said Harris had worked on uploading transcripts of lessons to create “personalised home worksheets” for pupils “based upon their needs”.
It was also investigating how it could feed models such as ChatGPT non-identifying elements from pupils’ EHCPs to personalise work.
But schools needed to be “very, very careful” not to feed models “sensitive information” about pupils.
“The teacher will know the students and the needs of their students, so they don’t need to put in [identifying information] for that person.”