Yesterday, I came across the following tweet:

Admittedly, I was slightly fuming at first, but then I realised that Victor Kumar makes a really neat point. Tying reading and cognitive offloading together in this way, he brings out (and debunks, at least implicitly) two misconceptions at once: (1) Cognitive offloading is often unduly restricted to the use of AI devices and then simply contrasted with analog skills like reading or writing. (2) At least in philosophy, reading is often unduly restricted to the passive intake of someone else’s thought and then simply contrasted with thinking for oneself. Schopenhauer famously put this as follows: “When we read, someone else is thinking for us: we merely repeat his mental process …” (More on this here.)

Neat as this point may be, there is a worry about deskilling I’d like to address in the following. People often debunk arguments about deskilling through AI use in similar ways by pointing out that we seem to have handled cognitive offloading through various means quite well in the past, suggesting that AI, and especially LLMs, might just turn out to be the next form of refined tool use, allowing us to offload torturous tasks to the LLM, while using the freed time and capacity for higher cognitive endeavours.

What this kind of argument seems to miss is a tipping point at which cognitive offloading, despite our best intentions, might well turn into deskilling. Compare cooking: Of course, there was a time during which we learned to prepare food in such a way that we prevented poisoning ourselves in the process. Gradually becoming shared knowledge, this skill became widespread and implemented thanks to a lot of offloading and division of labour. Certainly, this process can be seen as a great improvement. But now we’ve reached a level of offloading in food production where we increasingly allow for outright self-harm by stuffing ourselves with fast food and make do with industrial cruelty to non-human animals, resulting in harm and diseases on an evergrowing scale.

Returning to AI use, then, we shouldn’t miss the tipping point at which cognitive offloading might turn into all kinds of self-harm.