When Not to Use AI as a Cognitive Escalator

There is this concept of AI-driven cognitive atrophy going around. It may happen, there is ample evidence that our ability to perform some tasks degrades when we stop performing them.

But the key here is to know what to use AI for and when not to use it. If our brains can be engaged in more valuable thinking at higher levels of abstraction, then we should. Which functions is it ok to let atrophy to make space for new, higher-level functions?

Note that this is all very fuzzy, and that is the point: not taking an escalator to avoid walking up 12 steps to the gym is a simple determination; it is not that easy when it comes to AI. The touchpoints between humans and AI are not clear and discrete, they are nebulous, continuous and constant. So it's not easy to switch back and forth and decide what to do.

The interface has not been designed yet, it is completely ad hoc (though surprisingly effective).

Do I lose my ability to code when I use Claude Code? Probably. But then it is also the case that I can create more value by focusing on judging the output (code + what it does) than by painstakingly writing line after line of code. As we move up the abstraction layer, we lose our skills for the lower levels: that is both the benefit (we can manipulate more complex objects) and the risk (I can program in Assembly even though that's how I got started in the early 1980s, believe it or not).

This applies to other realms of life where the boundary and the interface are often no more clearly delineated.

I am thinking almost literally about a brain-computer interface or a cognition-ai interface: what should it look like if we want to get the best out of both AND maintain human agency AND scaffold in a controllable way toward higher abstraction levels so that we don't decouple for the lower levels entirely? Today the dominant input interface is the chat box and the dominant output modality is text, sometimes image.