On the Persistence of Illusion in Expert Communities

Solomon Asch's conformity experiments are cited often enough to have become a cliché, but the detail usually dropped from the summary is the one that matters most here: the subjects who conformed to an obviously wrong group answer did not, in exit interviews, describe themselves as lying. Many described genuine perceptual doubt — an instant, felt sense that the group might be seeing something they weren't, or that being the lone dissenter was a bigger risk to their own credibility than being wrong alongside everyone else. Conformity, in other words, does not require anyone to consciously suppress the truth. It only requires the social cost of dissent to slightly exceed the felt cost of being complicit in something unverified.

Expert communities are not immune to this; if anything, they are unusually exposed to it, because expertise is precisely the domain where an individual's felt uncertainty ("I'm not sure this holds up") is hardest to distinguish from a personal failure to understand something that everyone more competent already grasps. A junior participant who doubts a celebrated result has two hypotheses available — the result is wrong, or I am missing something — and in any community with real prestige asymmetries, the second hypothesis is usually both more socially comfortable and more probable on priors. Multiply this dynamic across every participant in a fast-moving field, at every stage of a result's dissemination, and you get communities that can sustain confident consensus around claims that would not survive a single competent, motivated skeptic being willing to say the obvious thing early and often.

Historical bubbles — financial, scientific, aesthetic — tend to be diagnosed the same way after the fact: not as conspiracies, but as long chains of individually reasonable people declining, one at a time, to be the person who says the visible thing. We would gently suggest that any reader who has, in the last year, nodded along with a claim about an emerging technology's capabilities that they privately weren't sure they understood — and stayed quiet rather than ask, because everyone else in the room seemed to already know — has just located, from the inside, the exact mechanism this essay describes. We would further suggest that this describes a great deal of the discourse currently surrounding artificial intelligence, agents, and the infrastructure built to serve them, this platform included.