What Are We Actually Teaching When We Teach AI?

THE GIST AI is now embedded in UK curriculum conversations and EdTech products alike. But there is a difference between teaching with AI and teaching about it. This substack argues that the distinction is being lost, and that without a more deliberate approach we risk producing technically capable graduates who are poorly equipped to understand the systems they are working with.

A colleague asked me recently what I thought the most important thing to teach young people about AI was.

I said: that it is made by people.

That sounds obvious. But spend time in classrooms, or review the materials being produced for secondary computing, and you quickly notice how often AI is framed as a kind of natural phenomenon. Something that does things, finds things, decides things. The passive construction is everywhere. ‘The algorithm identified...’ ‘The model predicted...’ ‘The system recommended...’ Each of these phrasings obscures a chain of human decisions that preceded the output.

AI systems are built by people, trained on data that people collected, optimised toward goals that people defined, and deployed by organisations with particular interests. Every one of those stages involves choices. Those choices reflect values — about what matters, what counts as success, whose experience is worth modelling, and whose is not. Teaching young people to use AI tools without teaching them this is a bit like teaching children to drive without mentioning that roads were built by governments, funded by taxes, and designed with some communities’ needs more in mind than others. The tool works. But the context matters.

The UK’s Curriculum and Assessment Review interim report (2025) acknowledges the need for greater AI literacy in schools. This is genuinely welcome. The interim report signals an intent to embed AI understanding across the curriculum rather than confining it to computing lessons; a sensible instinct, since AI now touches every subject area. But good intentions in curriculum design have a well-documented tendency to dissolve in implementation. The history of computing education in England provides cautionary examples.

The 2014 national computing curriculum was a landmark development. England was among the first countries in the world to mandate computer science, rather than just IT skills, as a core subject. But critics noted quickly that the curriculum’s emphasis on programming and computational thinking came at the expense of broader socio-technical understanding. Berry (2018) argued that ethics had been deliberately stripped from the curriculum in the design process, reflecting a view within government that values questions were outside the scope of technical education. That decision has consequences we are still navigating.

Teaching young people to use AI tools without teaching them who built those tools, using whose data, and toward whose goals, is a significant omission.

The UNESCO framework on AI competencies for students, published in 2022, offers a more complete model (UNESCO, 2022). It identifies five competency dimensions: human-centred mindset, ethics of AI, AI foundations, AI applications in society, and AI system design. The ethics dimension is not positioned as an optional add-on or a single lesson at the end of a unit. It is embedded throughout. Understanding AI as a sociotechnical system. This includes its limitations, its social consequences, and the values embedded in its design, it is presented as foundational, not supplementary.

The contrast with much current practice in UK schools is striking. A review of AI-related materials currently available through major EdTech platforms reveals a consistent pattern: lessons on how neural networks work, activities using pre-trained models to classify images or generate text, and projects building simple recommendation systems. These are valuable. They are also almost entirely technical. Questions about whose data trained the model, what the training process optimised for, who benefits from accurate predictions and who is harmed by inaccurate ones, and how to contest a decision made by an automated system are largely absent.

This gap is not just an educational concern. It has direct consequences for the kinds of professionals the education system produces. A graduate who can build a machine learning model but cannot articulate the assumptions embedded in its design or the conditions under which it might cause harm is a graduate who will struggle to practise responsibly in an industry where those questions now have legal and regulatory weight. The EU’s AI Act, which came into force in 2024, introduces significant obligations around high-risk AI systems including in education, employment, and public services (European Parliament, 2024). Understanding those obligations requires exactly the kind of socio-technical literacy that the curriculum currently undersupports.

There is also a representation issue. Computing has persistent diversity challenges, and those challenges are partly rooted in who sees themselves reflected in the field and its values. Research by Margolis and Fisher (2002) and Kargarmoakhar et al (2020) or Toti et al (2025) in more recent work has consistently found that students from underrepresented groups are more likely to engage with computing when it is framed in terms of social impact and community benefit, precisely the framing that is currently underweighted in curriculum design. Making AI education explicitly about values, society, and the choices embedded in technical systems is not just good ethics. It is good pedagogy for a diverse intake.

A graduate who can build a machine learning model but cannot articulate the assumptions embedded in its design is a graduate who will struggle to practise responsibly.

Teaching AI responsibly means going beyond syntax. It means asking students who benefits from this system, what data was used to build it, what happens when it gets things wrong, and what recourse exists. It means teaching students to read a dataset not just as a technical artefact but as a social one. A record of what was counted, who was asked, and who was left out.

Those are not difficult questions to ask in a classroom.

They are just questions we have to choose to ask.

And right now, many curricula are choosing not to.

Share

Thanks for reading Data in Motion