There is a small genre of parental bewilderment that goes: the hardest exercise got done, the easiest one did not. Reading at the top difficulty setting, finished. Maths, finished. A spelling rule that the child demonstrably knows, abandoned halfway, the one task of the week that quietly ate the largest share of the time. The dashboard has an opinion about this, and the opinion is “needs more practice on the spelling rule”. I would like to argue that the dashboard is answering a question nobody asked it.

There is a quiet fear running through clinical teaching right now, and it is rarely said out loud because it sounds faintly ridiculous. It goes: the student in front of me has a tool on their phone that will produce a better differential than I will, faster, and without the coffee. On what grounds, exactly, am I supervising them?

Our study on virtual patients came out in Perspectives on Medical Education last month, and the finding I keep returning to is not in the title. It is a small embarrassment buried in a survey table. Of the five things we asked twenty-one medical students to rate, the one they liked least was the one the rest of the evidence points at. I would like to say we predicted this. We did not.

There is a way to answer reading comprehension questions that requires no reading. You scan the passage, you scan the options, and you pick the option whose words you have just seen. It is fast, it is cheap, and for the first few years of school it is nearly correct. I find this genuinely funny, in the way that things are funny when they are also slightly horrifying: we spend enormous effort designing instruments to measure understanding, and a child can walk straight past the instrument using a trick that has nothing to do with understanding at all, and be rewarded for it, repeatedly, for years.

Everyone who builds learning software eventually discovers the same humiliating fact: the product works beautifully in the demo and is never opened at home. The usual explanations are all about the artefact. The content is too easy, or too hard. The feedback loop is too slow. The interface is not delightful enough. Somewhere a roadmap gets rewritten, a difficulty curve gets retuned, and six months later the numbers look exactly the same.

I went to the same conference twice, two years apart, and the second time it had swapped its bones. In 2024 the talk was about teaching practice: how to bring AI into the classroom, which tools were worth the licence fee, whether the students would cheat. This year, in Vienna, almost everything of consequence was about assessment. Not about whether AI can pass our exams — that argument is over, and it won — but about what our exams were measuring in the first place, and whether we would want to defend the answer out loud.

I signed my child up for the summer camp for the least noble reason available. The holiday here runs six weeks, six weeks is a long time to fill, and I wanted one of those weeks to belong to somebody who was not me. My expectations were correspondingly modest: a field, some supervision, a child returned in the evening in roughly the condition she departed in. This is, I think, the standard parental transaction, and I entered into it without ceremony.

There is a particular fantasy that arrives with every new technology, and generative AI has inherited the most seductive version of it: the fantasy of the vending machine. You insert a prompt, you receive a finished thing, and somewhere in between the messy human labor of thinking has been quietly abolished. People talk about AI this way constantly — as a machine that does the work — and I have come to believe this is precisely the wrong way to think about it, in roughly the way that believing your stomach digests food for you is wrong. It does, technically. But you still have to decide what to eat, and that turns out to be the entire game.

There’s a book in my bathroom. Not in the bathroom in the sense that someone reads it there regularly — I put it there on purpose, adjusted its position three times over several weeks, and waited. My daughter is seven. She hadn’t shown much interest in this particular book. I’d mentioned it once, which is to say I made the mistake of recommending it, which of course meant she’d now rather read anything else.

There’s a peculiar torture humans have inflicted on machines lately: show an AI a stick figure of a running person drawn by a four-year-old, and ask it why those three wobbly lines mean “running.” The machine, which can identify a golden retriever from 2,000 pixels with 99.7% confidence, suddenly stammers.

Xiao Ran recently started using AI tools to assist with homework. Whether it’s writing reflections for a language arts class or researching historical events, she turns to ChatGPT for instant answers. If the results aren’t satisfactory, she switches to DeepSeek for additional materials from different angles. At first, she thought these tools were “learning miracles”—no more flipping through books or browsing endless web pages, and answers appeared to be at her fingertips.

A step-by-step beginner’s guide to creating a personal website and blog using Jekyll and hosting it for free using GitHub Pages.