My roommate was cooking dinner the other night and had ChatGPT open on voice mode, talking to it like a cooking partner. "What should I add?" "How long should I cook this fajita for?" It was completely natural to him. I watched and thought: that is a person who has stopped developing cooking intuition. Every question he used to answer by trial and error — by tasting, adjusting, learning — is now outsourced instantly. He will always be a beginner.
I am not much different. The specific skill I have been losing is not cooking intuition — it is the ability to structure and articulate my own thinking.
Here is what the degradation actually looks like. I will be working through a problem and I can feel the shape of the argument in my head. I know what I think is true and roughly why. But when I try to convert that into words — to organize it into something coherent that another person could follow — I struggle. The connection between the thought and the articulation has weakened. So I open ChatGPT, describe what I am trying to say in fragments, and it assembles the structure for me. What comes back is cleaner and more organized than anything I produced, and I use it.
The result is that I am operating increasingly like a black box. Input goes in, correct output comes out, but I cannot fully describe the process in between. I can identify the right answer faster than I can explain why it is right or construct an argument for it from scratch. The surface looks fine. Underneath, something is atrophying.
This is different from using a calculator instead of doing arithmetic. Calculators handle a discrete task with a clear boundary — you know you are outsourcing the calculation, and everything else remains yours. LLMs are blurring a much more fundamental boundary: the boundary between your thinking and someone else's. When I use ChatGPT to structure an argument, I am not just outsourcing a mechanical task. I am outsourcing the cognitive work of turning raw thought into communicable form. And that cognitive work, it turns out, is not separable from the quality of the thought itself. The act of struggling to articulate something is part of how the idea gets refined. When I skip that struggle, I get a cleaner sentence but a shallower idea.
There is a common counterargument that tools always change cognition and that this is not inherently bad. Writing replaced the need to memorize everything; printing reduced the need to copy texts by hand; calculators made arithmetic less central to mathematical education. Each shift was met with concern and each mostly worked out. Maybe LLMs are just the next version of this — freeing up cognitive bandwidth for higher-order thinking.
I find this less convincing than it sounds. The previous tools each replaced a task that was clearly separable from the underlying thinking. Writing did not replace thinking — it extended it. Calculators did not replace mathematical reasoning — they automated a mechanical step within it. LLMs are different because they operate at the level of reasoning and language themselves. The "task" they are replacing is not a discrete mechanical step — it is the process of converting thought into argument, which is inseparable from the process of developing the thought in the first place.
The question I keep coming back to is whether this is reversible. Skills that are not used do decline, but they can usually be rebuilt with practice. The more concerning scenario is if the habit of offloading becomes so default — so early, for so many people — that the skill of structured articulation never properly develops in the first place. For students who have grown up with these tools, that may already be the situation.
My own answer has been to write more without AI, to sit with the discomfort of an unpolished draft rather than immediately handing it off to be cleaned up. It is slower and the output is initially worse, but the thinking behind it is more genuinely mine. That trade feels worth making. An idea you can fully articulate and defend is more valuable than a polished paragraph you cannot reconstruct from scratch.