Patri Friedman

(@pboothe says that these folks trained their AI on data where criminal photos were mugshots and non-criminal ones were corporate photos.)

Perhaps this paper, despite bravely publishing heresy, is wrong. But I bet most condemning it do so b/c it's heresy, not b/c it's wrong.

It is not actually a coincidence that "condemning the heresy" turned out to be right. Many orthodoxies encode valuable heuristics.

It is *reasonable* to demand greater care and stronger evidence for claims that match patterns known to have often led to mistakes before.

Sure. But the general pattern also supports the idea that standards required for claims related to recent disasters tend to be unreasonable

Well, those standards would have helped these folks dodge a collection of humiliating and harmful blunders in this case.

And what makes systemic racism & similar bias a "recent disaster" rather than a "consistent pattern of cognitive bias throughout history"?

Recent disasters like the Holocaust, the eugenics movements, and slavery distort rational evaluation of any vaguely associated topic.