When It Feels Like AI Checked Your Work |
AI can make work feel "reviewed" before anyone independent has actually checked it.
A real check on AI output has to come from outside the loop that produced the answer.
When AI collaboration feels too smooth, that's the cue to add friction—not to trust the result.
I sent a change to an AI model — a dataset update for the calculations in a paper I have been writing on megainfrastructure. The model came back with the revised numbers, cleanly formatted, in the shape I had asked for. I read the top line, scanned the rest, and pasted it into the working draft. The whole exchange took under five minutes. It was only later, looking at the file again, that I noticed I had not actually read what the model returned. I had reviewed the feel of it—the formatting, the plausibility, the fact that no error had been raised—and treated that as review. My first thought was not, "The model might be wrong." It was, "If it was wrong, I should have caught it." That second thought is the one to sit with. It assumes a check happened.
The Illusion of Answer-Making
We often mistake answer-making for intelligence because we forget the second half: checking whether the answer deserves to be believed. Not the answer alone, but the answer plus the act that ratified it. Sometimes that was a colleague who disagreed.........