Tukuk-OSConfirmation bias is the default setting of human thinking: we gather evidence that flatters what we already believe, interpret ambiguity in its favour, and remember its hits while forgetting its misses. The classic lab demonstration: given a number sequence obeying a hidden rule, people overwhelmingly test by proposing sequences that confirm their guess instead of sequences that could falsify it — and stay wrong confidently. The same machinery runs at reading scale: the article you would believe without checking is the one you already agree with, and the one that makes you fact-check hardest is the one that challenges you — which produces the perverse result that believers in a claim end up with more exposure to reasons for it than reasons against it.
Social media turned a private weakness into infrastructure: feeds optimise for engagement, engagement rewards agreement and outrage, and both push you toward content that confirms — the filter-bubble effect is real in measurement but weaker than either pole claims (people’s networks do leak across lines; the stronger effect is self-selection: you click what flatters you). The more useful framing is that algorithms do not need to censor anything to bias you: they only need to sort your attention by what feels right.
The counter-moves are boring and they work. Steelman: restate the opposing position until someone who holds it says “yes, that is it” — before answering. Premortem: assume you are already wrong and ask what evidence would have shown it — then go find exactly that. Ask the asymmetry question: would I demand this standard of proof from a claim I liked? And notice the feeling: comfort is not evidence — the sentence “this is obviously true” deserves more suspicion when it feels good than when it grates.