Policy

Education and wider government policy on AI has evolved considerably over the last couple of years. The UK government has developed a sound set of regulatory principles, which apply across the public sector: AI use should be underpinned by safety, security and robustness; by transparency and explainability; by accountability and good governance; and by contestability and redress. For schools or trusts looking to draft their own AI policies, this is as good a list of subject headings as any, and is particularly relevant given Ofsted’s role in holding schools to account against these principles.

The DfE are investing in the development of generative AI tools to support teachers and schools. These are focussed more on teacher-facing tools than anything which pupils themselves might use. They’ve put £4m into a trusted content store, to pull together curriculum guidance, lesson plans and anonymised pupil work that can be used for augmenting or fine tuning language models. For secondary teachers, it would be great if this also included GCSE and A Level specifications, past papers, mark schemes and examiners reports, but there appear to be some thorny copyright issues here.

Oak National Academy have a comprehensive library of lesson plans and teaching resources. They’ve used this as the basis for a couple of AI experiments, firstly a generator for multiple choice questions, but more impressively Aila, a lesson plan generator, which walks the user through a sequence of questions to tailor its planning, producing an editable planning, quizzes and slides. Oak’s approach doesn’t suit everyone, but I’d encourage readers to try it out - it’s free.

DfE are also funding proof of concept and pilot projects, including some work from Faculty on providing feedback to teachers and Key Stage 2 pupils on their writing - we should see other subjects and key stages pilots soon. I suspect few pupils will find it as motivating to get feedback from an AI as knowing that their teacher has read and responded to their work.

Education Secretary Bridget Phillipson opened the BETT show with a generally enthusiastic vision of how AI can support educational aspirations, and help with recruitment and retention. Phillipson also announced new training materials for teachers on using AI, and alongside this we saw significant revisions to the DfE’s AI guidance, which is now much more about teachers using AI for low-risk, workload reduction tasks, and takes a much more cautious approach to pupils’ own use of these tools, making clear that this latter must be inline with providers’ terms and conditions and safeguarding requirements.

A few days earlier, Downing Street had announced its AI action plan. This includes upskilling the workforce, with a hint that AI skills will be there alongside digital skills in curriculum changes following the current review. The plan references how South Korea has integrated AI, data and digital literacy into its education pipelines. Closer to home, some of EU’s AI Act provisions came into effect on 2 February, including the requirement that organisations deploying AI ensure that there’s training for those using it. The voluntary approach in Britain suggests that some divergence from mainland Europe on this already.

Technologies

The technologies move on more quickly than policy, and new products and versions are released every week. Rather than trying to keep up with latest thing, the pragmatic approach for schools is to develop proficiency with the core tools. My advice is that if schools are using Windows and Office, then Copilot is likely to be a good choice; if they’ve using Chromebooks and Google Apps, then Gemini is likely to work for them.

A few recent things that have caught my eye recently:

• NotebookLM, from Google, makes retrieval augmented generation, where the language model responds based on one or more uploaded documents, very accessible. It’s great at producing knowledge organisers and practice questions when given particular curriculum materials. It’s greatest feature is its production of realistic sounding ‘podcasts’ of two very enthusiastic, knowledgeable presenters discussing whatever content it’s been given.

• Brisk is a Chrome extension which will quickly produce AI resources to accompany any webpage you visit - I particularly like it’s Boost feature which creates a class set of chatbots for pupils to learn about the content independently. I’m a bit hesitant about the privacy permissions it requests, so take care that your data protection colleagues sign this off before using it in school.

• Custom GPTs and Gems - if you or your school are paying for the plus versions of the tools, you should be able to create your own bespoke version of a chat bot, and share access to this with other users of the platform. We’ve done some work with this at Roehampton, experimenting with bots that give feedback on pupils’ programming exercises, feedback, but not grades, on students’ draft assignments, and even practice for job interviews.

• LearnLM - Google have taken their Gemini language model and fine tuned this with a particular educational focus. It builds in to its chat bots key pedagogic ideas like active learning, managing cognitive load and the importance of motivation and engagement. You can try using this with a free API key in their AI studio, but Gemini’s ‘learning coach’ gem now implements these too, at least for those with access to Gemini advanced.

It is fun exploring and experimenting with the latest tools, but I think we should be a little more cautious about rushing to recommend these to pupils, to incorporate them in our teaching or to adopt them at school or trust level. There’s still a need for subject and pedagogic knowledge in prompting well, for criticality and discernment in evacuating responses, and for wisdom in discerning whether it would be better to do the work for ourselves rather than letting the machines do this for us.

Originally published in Sapientia, the newsletter of ICT for Education.