AI cannot teach you, build for you, or take from you.

I have recently given presentations on AI and the issues it has created for learning to higher education students at colleges, where I am chairman, in Taskent, Yangon and Singapore. Here are my thoughts on what I was attempting to convey to students.

The image I used to open a talk about artificial intelligence was made by Gemini AI, a small irony I rather enjoyed. My argument is that fear of AI undermining a student's learning and understanding is the wrong response, and that the useful question for a student to ask is what AI cannot teach you, build for you, or take from you.

So this is the address I gave to the students: Consider

first the humble pocket calculator. When scientific calculators reached classrooms in the 1970s, there were serious voices arguing that they would destroy arithmetic and with it the habit of mathematical thought. Some schools banned them. The ban failed, as bans on useful tools usually do, and it failed for a revealing reason. The calculator never replaced the need to understand why a method works, to choose the right method for the right problem, or to recognise at a glance that an answer is obviously wrong. A student who types a figure into a machine and accepts whatever appears has learned nothing. A student who knows that the result should be roughly forty, and sees four hundred, has learned something no machine can hand over. Banning AI is a losing strategy for the same reason. It treats the tool as the problem, when the real issue is what the student brings to it.

This is why I think the old question of education, what do you know? is dead. For generations we asked whether a student could produce an accurate, well-organised answer, and we built coursework around that capacity. AI can now produce such an answer in seconds, so the capacity is no longer scarce and no longer a reliable sign of learning. The questions that matter now are harder. What can you do with what you do not yet know? Can you reach a judgement that you are willing to put your name to? A fluent paragraph costs nothing to generate. A considered position, defended by someone prepared to be wrong in public, still has to be earned.

Universities are beginning to respond, and the better responses are worth describing. The most promising shift is away from grading the finished output and towards grading the process that produced it. Students may be asked to show how they arrived at a conclusion, to verify every one of the AI’s outputs against reliable sources, and to defend their work in a personalised oral examination, where there is nowhere to hide behind a polished text. An oral examination is difficult to fake and difficult to outsource, and it rewards the student who actually understands the material. I would add that assessment redesigned in this way also tends to be better assessment, because it tests thinking rather than the production of text. I have seen no sign that this is being done uniformly or quickly, and institutions differ greatly in how far they have got, but the direction of travel seems clear to me.

There is a persistent myth that the so-called soft skills are passive, decorative extras laid over the real and hard business of technical knowledge. I would reverse the relationship. Communication, empathy and teamwork are not optional add-ons. They are durable skills, and so are the cognitive capabilities that sit alongside them: critical thinking, analytical reasoning and digital literacy. Human judgement and decision-making are among the most resilient capabilities we have, precisely because they are the ones a system trained on past data is least equipped to exercise on our behalf. A graduate who can reason carefully, work with other people and explain a decision to those affected by it will not easily be replaced by software.

Then there is curiosity, which I regard as the clearest case of something AI cannot teach you. An AI system will answer any question you put to it, and often answer it well. What it will not do is wonder about something for no reason at all and decide, on its own initiative, to go and find out. The student who asks a question and then follows the answer somewhere nobody sent them is doing something categorically different from the student who simply retrieves it. Every historian, anthropologist and archaeologist I have known began as that second kind of person, the one who wandered off down a side road because something nagged at them. If AI makes answers cheap, then the ability to ask a good question, and the appetite to keep asking, becomes more valuable and not less.

So what should a student actually do with these tools? The test I suggest is a simple one. Is the AI doing the learning for me, or is it helping me learn? If it is the former, something has gone wrong. Never accept its facts at face value. Check citations, page numbers and historical data against textbooks and peer-reviewed journals, because these systems can state invented detail with complete confidence, and a reference that looks plausible may not exist at all. Use it to break down dense jargon, to rephrase a difficult theory, or to explain a concept in simpler terms, much as you might ask a patient tutor. Do not use it to cheat, and disclose your use of it honestly. Above all, use it to supercharge your curiosity rather than to replace it.

AI should, I believe, be a positive aid to learning. But you will still be judged on your knowledge, on your grasp of process, and on your ability. The machine can sit beside you while you work. It cannot sit the examination for you, and it cannot take from you the understanding you have built for yourself.