Some teenagers are already using artificial intelligence for homework, revision and coursework. Others avoid it entirely. At Mearns Castle High School in East Renfrewshire, staff decided that pretending it does not exist was not an option.
The school is believed to be the first in the UK to appoint a dedicated principal teacher of AI literacy. The post went to Jamie Laycock, formerly principal teacher of biology. The message behind the role, as he puts it: "We can't just hope AI goes away."
What makes this worth attention is not that a school has embraced AI. It is that the job is the opposite. Laycock's brief is to teach pupils to question it, challenge it and use it responsibly — scepticism as a subject.
The four things on the syllabus
In an early lesson, pupils were shown a mix of real and AI-generated images and asked to spot the fakes. It was harder than they expected, and Laycock says it gets harder by the day.
Hallucination. He dislikes the word. "I think hallucination is actually a really bad word, but it basically makes stuff up." His point about why this hits young people hardest is the sharpest in the piece: without what he calls foundational knowledge, a pupil has no way to tell an invention from a fact. The tool is least safe for the person who knows least — which is the person most likely to be using it for homework.
Anthropomorphisation. Treating a chatbot as a trusted friend or expert because it sounds like one. "It has human-like qualities but actually this is just a machine. And for a young person, in particular, that is actually a really difficult thing to keep in perspective all the time."
Sycophancy. "It's sycophantic. It always tells you you're amazing, that your idea is great." A model that agrees with you is pleasant and useless as a check on your thinking, and he warns it makes pupils over-reliant.
Bias. Models carry the norms of the material they were trained on, which is mostly not written by or about people in the room.
Lessons are tailored to different ages and stages.
Why this list is better than a tools course
Most "AI in schools" programmes teach prompting. Prompting is a skill with a shelf life measured in months; every one of the four items above is a durable property of how these systems work, and none of them is going away with the next model release.
There is also a quiet admission in the syllabus. Three of the four — hallucination, sycophancy, bias — are not misuse. They are the normal behaviour of a working system. Teaching them as hazards rather than bugs is more honest than most corporate training on the subject.
The version Bangladesh needs is not a copy of this one
Bangladeshi classrooms have the same problem arriving faster, with fewer staff and no post anyone is funding. But importing this syllabus unchanged would miss two things specific to here.
The first is language. These models are weakest in Bangla — more prone to invention, thinner on local facts, and worse at saying they do not know. A pupil checking a Bangla answer has less to check it against, so the hallucination lesson matters more here, not less.
The second is what the output is used for. The flattery problem in a UK classroom produces a weak essay. In a country where a very large number of students use these tools to draft job applications, admission essays and freelance proposals in a second language, a model that tells everyone their work is excellent is actively setting them up to be rejected without knowing why.
The practical form this could take does not need a new post or a budget line. It needs one lesson, repeated: give pupils a question they already know the answer to, let the model answer it, and have them mark the model. That single exercise teaches hallucination, sycophancy and the limits of fluency at once, and any teacher can run it with a phone.
On how far the conversation has got here, see the report saying schools are not ready, and the hundred thousand who sat an exam for 1,600 AI training places.
