I used to believe that it was wrong, or at least counterproductive, to shun people for openly holding evil viewpoints. Surely that would just cause smart-but-evil people to pretend they're on the side of good, while less capable liars would be punished for honesty. And all this conformity forcing would achieve is killing the discussion around these topics, meaning that evil beliefs arising from incorrect or incoherent data couldn't get corrected. Moreover, if some of our own prosocial views are based on falsehoods, wouldn't it be a great service to strengthen our foundations.

For instance, imagine a smart principled racist, someone who actually believes that outwardly visible ethnicity is a reasonable way to predict how someone will behave. It's hard to even think about, as the halo effect tells us that you cannot be both smart and racist. I sure used to think I'm both. If there were real argument against racism, why were those not presented instead of just asserting that it's wrong? So-called "scientific racism" used to be the majority viewpoint, and I'm not surprised that reasoning from the first principles would lead there first. After all, I'm quite sure that appearances are a reasonable way to predict behavior, after all, everything is correlated, and the actual argument is more about not mistreating individuals based on statistical predictions. Which is reasonable, and I'm mostly sad that the core point itself is rarely articulated.

Anyway, the argument goes like this: That specific person is making decisions about other people, like hiring employees or selling insurance. If we disallow the visible use of racial bias, they'll continue using those exact same ideas, but not talk about them anymore. Now everybody is worse off. Some enforcement can be done to disincentivize this, and it partially works. But wouldn't it have been better if the enforcement could target the outcome and not the intention?

Other examples of evil behavior include legal profit-maximizing, meat-eating, meritocracy, human rights advocacy, supporting the wrong political party, and anything that's called prejudice. I'm sure my examples are not that representative sample of the real problems, and there must be more of these, but it's hard to see outside one's bubble.

The first step towards abandoning this line of thinking came from realizing that good and evil are not that clearly defined. I didn't mean those as value judgements even before, instead of something like majority opinions in my social circles. But it's not just about minority-majority, it's about politics, which is why these themes are often reduced to good and evil instead of having civilized utilitarian policy arguments. (Hah)

I'm sure I was somewhat meta-contrarian when arguing this, but it would be incorrect to say I didn't actually believe it, at least partially. It's such a convincing take, despite (or because) of its unintuitiveness. It feels like leaning on mistake theory, although by the time I'd learned about mistake theory I had long since given up this argument. But there's still a part of me that believes there's at least some merit to it, which is why I'm trying to deconstruct it here.

Prejudice is contagious, in a memetic sense. Maybe it's even a cognitohazard? I wonder how much of xenophobia is learned instead of innate. I still think hearing both sides of an argument and then judging is the sane way to go about things, but the defaults actually matter. Most people don't seem to think about these things much, and just having some simple answers is worth a lot.

Another counterargument is that it's hard to hold different internal and external beliefs. Maybe people will actually change their internal beliefs if forced to act as if they were different? Then again, free speech is quite nice, and the soul-corrupting effects that this causes whenever society holds some values you don't share are not good either.

And lastly, a big part of why prejudice hurts people is not the decisions made in secret, but the openly hostile behavior. Just reducing that is worth quite a bit.