AI and thinking

A child is stuck — on a maths problem, on a paragraph that will not start, on a question about why the moon changes shape. That stuck feeling used to last a while. Now it lasts about four seconds, because there is a box on the screen that will answer. The answer arrives, it is good, and everybody moves on.

What makes this a real objection rather than a grumble is that effort is not a side effect of learning. It is closer to the mechanism. Working at something, getting it wrong, holding a half-formed idea in your head long enough to test it — that is the part that does the building. A tool that removes the effort is not obviously separable from a tool that removes the learning, and you do not have to dislike technology to notice that.

The worry also has a long and serious history. Writing was going to weaken memory. Printing was going to make people shallow. Calculators were going to end arithmetic, and search engines were going to end knowing things. That history usually gets deployed to close the argument, and it should not be. People worrying before and being wrong about the details is not evidence that they are wrong now — each of those tools did change what people were good at, and those changes were real. The honest position is that the question is open, not that precedent has already settled it.

Producing an answer yourself makes it stick better than reading the same answer. Learning research has called this the generation effect for a long time, and teachers recognise it without needing the term: the work of retrieving or attempting something is a large part of what fixes it. A worked solution handed over feels more efficient and is frequently less useful.

Skills that go unpractised fade. This is unremarkable and it applies to everything — mental arithmetic, spelling, finding your way across a city without a map, holding an argument together for three paragraphs. Handing a task to a tool permanently is not neutral; the capacity behind it quietly declines. The general habit has a name in the research, cognitive offloading, and it is neither new nor sinister — a shopping list is offloading. It still has a cost, and the cost is paid in the skill you stopped using.

Convenience wins over effort when both are on offer. Not always and not for everyone, but reliably enough that it is the safe assumption when you are setting a rule at home. People do not usually take the harder route because it is better for them later; they take it when something in the situation makes the easier route unavailable.

A tool that produces a finished, fluent answer in two seconds sits on the wrong side of all three. It removes the generating, it absorbs the practice, and it is the most convenient thing on the screen. You do not need any claim about brains being rewired to find that concerning. The ordinary mechanics of learning are enough.

What none of this supports is a number. The research can give you a direction — effort helps, disuse costs — but it cannot tell you how much a household loses when a fourteen-year-old uses a chat assistant for history homework twice a week. Anyone who offers you that figure is estimating. The direction is solid; the size of the effect, for this technology, is not measured.

Pick something — a weekly piece of writing, the first attempt at every maths problem, reading a chapter and saying out loud what it was about — and keep the tool out of it completely. The point is not moral. It is that you cannot see what is changing if every piece of work has help inside it. An unassisted sample is the only way the difference stays visible, and it is what you will want if you ever have to make a decision about any of this.

None of this requires you to allow the tools at all. If the answer in your house is not yet, the habits are still worth having, because your child will meet these tools at school, at a friend’s house, and on a phone that is not yours — and the habit of arguing with an answer travels.

Whether a generation raised with these tools ends up with fewer capacities or different ones is not knowable from here. Both have happened before with other technologies, and which one this turns out to be will be argued about long after the current round of confident predictions is forgotten. Anyone telling you now is guessing, in whichever direction they are guessing.

The long-run effect on writing and reasoning is genuinely unmeasured. Writing is not only a way to transmit a thought; for most people it is where the thought gets finished. What happens to that when a competent draft is always one sentence away is a real question, and a few years of use cannot settle it.

There is also the awkward possibility that none of this becomes measurable until after it has happened. Changes in how people think are slow, tangled up with everything else in a life, and mostly visible in hindsight. That is not a reason to panic and it is not a reason to relax. It does mean that a parent who looks at this uncertainty and decides to delay these tools, or to keep them narrow, has drawn a legitimate conclusion from it rather than avoiding the question. Waiting is a defensible reading of an unfinished picture, and it is yours to make.