You can’t see past what you’re best at.
Gibson, filmed in the backseat of a car in 2000, put it in terms of technology: people are “unaware of the extent to which they’ve been interpenetrated and co-opted” by their tools. Too close to see, he said. It’s altered our physical being. We can’t be stripped of it, so we can’t examine it.
Twenty-six years later an OpenAI model disproved a conjecture that combinatorial geometers had believed since Erdős posed the unit distance problem in 1946. Every expert assumed the square grid was essentially optimal for maximizing how many point-pairs can sit at exactly distance one. They assumed this because their geometric intuition said so. The same intuition that lets them do geometry at all.
The model had no geometric intuition. It imported algebraic number theory, found a construction that beats the grid by a polynomial factor, and produced a proof that Tim Gowers said he’d recommend to the Annals without hesitation.
Eighty years of expertise was a specific kind of blindness. The intuitions are load-bearing. You can’t practice combinatorial geometry without them. But they’re also what made the counterexample invisible. Proximity to the material let you work; it also kept you from seeing.