My daughter is almost two. Right now, her hardest problem is getting a shape into the right hole in a plastic bucket, and she will not let me help. She pushes my hand away, turns the piece over herself, tries again, and again, until it drops in, and then she looks up at me as if she has just solved something enormous. She has, in her own way. Nobody handed her the answer.

I think about her when I read the research on generative AI and thinking. In fifteen years, she will be sitting in a classroom not unlike the ones at Wilderness, and whatever grown-up version of that bucket and shape she meets there, the tools around her will be able to make it disappear in half a second if she lets them. I want her to have the chance to push the hand away. That is what has been sitting underneath the policy work we have been doing this term on AI and academic integrity, and it is what this piece, following on from my last one on the see-saw of challenge and resource, is really about.

Cognitive offloading is one of the most ordinary things a mind does. Researchers define it as “the use of physical action to alter the information processing requirements of a task” (Risko & Gilbert, 2016): in plainer words, letting something outside your own head carry part of the mental work, so that a task demands less of you. We do it constantly, and mostly it is a good thing. A shopping list, a calculator and a diary all reduce the load so the mind is free for something else. When we let a search engine hold facts for us, we tend to remember where to find them rather than the facts themselves (Sparrow et al., 2011). The purpose of offloading is precisely to make a task easier. The question is never whether we offload, but which difficulty we are removing.

That question is the whole of it, and it is why a tool as capable as generative AI deserves careful thought. Some difficulty is merely friction, and worth removing: the long arithmetic that stands between a student and the physics she is really trying to learn. But some difficulty is the learning itself. When the effort of thinking can be handed over completely, the challenge that would have built the mind can be cleared away alongside the challenge that was only ever in the way. In my last piece I described learning as a see-saw, balancing the challenge a girl meets against the resources she can draw on, with growth living just past the edge of what she can already do (Dodge et al., 2012). Offloading, by its nature, lightens the challenge side. Used well, that frees her to reach further. Used without noticing, it can flatten the see-saw altogether, so that the “work is done” and yet nothing has been asked and nothing has grown.

This is what learning scientists have long understood: in learning, the difficulty is not in the way; the difficulty is the mechanism. Elizabeth and Robert Bjork call them desirable difficulties, the effortful, slower, uncomfortable conditions that feel like failure in the moment and yet leave understanding deeper and more durable (Bjork & Bjork, 2011). Retrieving an idea only half-remembered, ordering an argument that resists you, sitting with a problem that will not yet yield. These are not obstacles to a mind being formed. They are the sensation of it being formed. As the Bjorks put it, “optimising learning and instruction often requires going against one’s intuitions” (Bjork & Bjork, 2011). The uncomfortable truth is that the experiences which build a mind are exactly the ones any of us would most like to skip, and a tool that lets us skip them is one to use with care.

The costs of removing that effort are beginning to be measured. In 2025, researchers at the MIT Media Lab recorded the brain activity of people writing essays. Those who used ChatGPT showed markedly weaker neural connectivity than those who wrote unaided; many could not quote a single sentence from the essay they had just produced, and felt little ownership of it. When they later wrote without the tool, the weaker patterns lingered. The researchers called this accumulation “cognitive debt” (Kosmyna et al., 2025). It is a small, early study, not yet peer-reviewed, and it should be read with care. But its direction echoes a wider literature: a survey of knowledge workers linked heavier use of generative AI to less critical thinking and reduced mental effort (Lee et al., 2025), and a study of over 600 people found that heavier reliance on AI tools was associated with lower critical thinking scores, with cognitive offloading the apparent link between the two (Gerlich, 2025).

None of this is an argument against the technology. Our girls will grow up alongside these tools and will need to be fluent, confident and discerning in using them; a school that pretended them away would be preparing students for a world that no longer exists. The task is discernment: learning to tell the difference between offloading that frees the mind for harder work, and offloading that quietly removes the very challenge the mind needed. There is a time to let a tool carry a load and there is a time when the struggle is the entire point, when the rough first draft, thought and written by hand, is the very thing forming the mind. At those times, the tool is best left closed. Holding that line is the new frontier of the rigour I wrote about last time. It is simply keeping the challenge where it belongs.

In practice, this means we will often ask our girls to think first and reach for the tool second: to make the rough, effortful, imperfect draft before any machine polishes it; to use AI to test their thinking rather than to supply it. It means our teachers are designing assessments that requires challenging cognition. It means talking with your daughters, at home and at school, about the quiet value of the hard way, so that reaching for the difficulty becomes something they choose, and in time something they prize. The desirable difficulty.

What I want for our girls is simple to say and precious to hold: that they leave Wilderness able to stay inside a hard problem long enough for their own minds to do the work, having felt the difficulty and chosen it. That capacity is among the most valuable things an education can give, and it is exactly what an easier path can quietly erode, in all of us, if we let it. It is worth protecting, gently and deliberately.

It is, in the end, another way of being Always True.
 

References
Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher, R. W. Pew, L. M. Hough, & J. R. Pomerantz (Eds.), Psychology and the real world: Essays illustrating fundamental contributions to society (pp. 56–64). Worth Publishers.

Dodge, R., Daly, A. P., Huyton, J. L., & Sanders, L. (2012). The challenge of defining wellbeing. International Journal of Wellbeing, 2(3), 222–235.

Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6. https://doi.org/10.3390/soc15010006

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task (arXiv:2506.08872). arXiv. https://doi.org/10.48550/arXiv.2506.08872

Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (pp. 1–22). Association for Computing Machinery.

Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002

Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776–778. https://doi.org/10.1126/science.1207745