What the Calculator Can Teach Us About AI

Most of what people worry about with AI in education and training is right. If a tool hands learners the answer, they skip the productive struggle, and what they don’t practise they don’t retain.

Schools have been here before, with the calculator. The answer then was not to ban it. It was to design the curriculum and the assessment around it.

When calculators reached classrooms, many feared children would stop learning arithmetic. In the US, the National Council of Teachers of Mathematics recommended in 1980 that maths programmes “take full advantage of the power of calculators and computers at all grade levels”.

In England, the Cockcroft Report of 1982 said a calculator “in no way reduces the need for mathematical understanding”.

The research backed them. A review of 79 studies found that calculators did not harm basic skills at almost every age, and pupils who used them liked maths more and felt more confident (Hembree and Dessart, 1986). A later review of 54 studies found the same broad picture (Ellington, 2003).

What made it work was assessment design. In England today, every pupil sitting GCSE maths takes one paper without a calculator alongside two with one. The calculator is part of the subject, and the exam still checks that pupils can do the arithmetic themselves. We need the same kind of thinking with AI.

AI is a harder case, because a calculator does one narrow job and AI can do the whole task. The evidence shows what happens when nobody designs around that.

In a trial with nearly 1,000 high school maths students in Turkey, those given plain ChatGPT for practice scored 48% higher while they had it. On the exam afterwards, without it, they scored 17% lower than students who never had AI. A version with pedagogical guardrails, which scaffolded with hints rather than giving answers, removed most of that harm (Bastani and colleagues, 2025).

Good design can do more than avoid harm. In a trial on a physics course at Harvard, in the US, students’ learning gains were more than double with an AI tutor built on learning science, which scaffolded each problem one step at a time, compared with an active-learning lesson in class (Kestin and colleagues, 2025). The difference came from how the tool was built.

The same pattern shows up at work. In a survey of 319 knowledge workers, those who trusted AI more reported doing less critical thinking, and those more confident in their own skills reported doing more (Lee and colleagues, 2025). The researchers call the worker’s new role “task stewardship”: overseeing what the AI produces and staying accountable for it.

That is how I built my own portfolio. I set the vision and shaped every idea. AI did most of the research, the writing and the code. I stayed the human in the loop: I designed the concepts, checked every design and source, and proofread everything.

Schools, universities and workplace training teams should design learning experiences the same way. The learner owns the design decisions, critically evaluates what the AI gives back and iterates until it is right. The AI does the labour.

There is a line to draw here. Handing the labour to AI only works for a skill a person already has. I could judge what the AI wrote because I can write, and check its sources because I know how research is done. Someone still learning a skill has to do the work themselves, which is what the trial in Turkey showed. So the AI does the labour around the skill being learned, and never the skill itself.

We learn by doing. People remember what they produce better than what they only read (Slamecka and Graf, 1978).

The calculator gives us the last piece. Wherever people learn with AI, in a classroom, a university or a training programme at work, assess the learner’s design decisions and judgement without it.

A pupil explains and defends the choices in their project. A new sales hire talks a manager through why one pitch suits a client better than another. A site supervisor fixes a flawed safety plan by hand. Each is a version of the paper every GCSE pupil still sits without a calculator.

AI is becoming another reasoning partner. Ideas take shape faster, the admin is taken away, and thinking is encouraged rather than replaced. It is going to show up in all our lives, and it is not going away.

So its use should be encouraged, in a way that keeps the design and the judgement with us. Used correctly, it makes our time more productive and efficient, scales what one person can do and leaves us freer to think. We still have to know how to do things ourselves, and the future of training, in classrooms and at work, should be built around that.