The underlying statistical phenomenon is just regression to the mean: if people aren’t perfect about determining how good something is, then the one who does the thing is likely to have overestimated how good it is.
I agree that people should take this kind of statistical reasoning into account when deciding whether to do things, but it’s not at all clear to me that the “Unilateralist’s Curse” catchphrase is a good summary of the policy you would get if you applied this reasoning evenhandedly: if people aren’t perfect about determining how bad something is, then the one who vetoes the thing is likely to have overestimated how bad it is.
In order for the “Unilateralist’s Curse” effect to be more important than the “Unilateralist’s Blessing” effect, I think you need additional modeling assumptions to the effect that the payoff function is such that more variance is bad. I don’t think this holds for the reference class of “blog posts criticizing institutions”? In a world with more variance in blog posts criticizing institutions, we get more good criticisms and more bad criticisms, which sounds like a good deal to me!
I think you should read Bostrom’s actual paper for why this is a more compelling argument specifically when dealing with large risks. And it is worth noting that the reference class isn’t “blog posts criticizing institutions”—which I’m in favor of—it’s “blog posts attacking the credibility of the only institution that can feasibly respond to an incipient epidemic just as the epidemic is taking off and the public is unsure what to do about it.”
I would support a policy where, if an LW post starts to go viral, then original authors or mods are encouraged to add disclaimers to the top of posts that they wouldn’t otherwise need to add when writing for the LW audience. As SSC sometimes does.
I would not support a policy where LW authors always preemptively write for a general audience.
Here we face the tragedy of “reference class tennis”. When you don’t know how much to trust your own reasoning vs. someone else’s, you might hope to defer the historical record for some suitable reference class of analogous disputes. But if you and your interlocutor disagree on which reference class is appropriate, then you just have the same kind of problem again.
I really don’t think this is a reference class tennis problem, given that I’m criticizing a specific post for specific reasons, not making an argument that we should judge this on the basis of a specific reference class.
And given that, I’m still seeing amazingly little engagement of the object level question of whether the criticisms I noted are valid.
The underlying statistical phenomenon is just regression to the mean: if people aren’t perfect about determining how good something is, then the one who does the thing is likely to have overestimated how good it is.
I agree that people should take this kind of statistical reasoning into account when deciding whether to do things, but it’s not at all clear to me that the “Unilateralist’s Curse” catchphrase is a good summary of the policy you would get if you applied this reasoning evenhandedly: if people aren’t perfect about determining how bad something is, then the one who vetoes the thing is likely to have overestimated how bad it is.
In order for the “Unilateralist’s Curse” effect to be more important than the “Unilateralist’s Blessing” effect, I think you need additional modeling assumptions to the effect that the payoff function is such that more variance is bad. I don’t think this holds for the reference class of “blog posts criticizing institutions”? In a world with more variance in blog posts criticizing institutions, we get more good criticisms and more bad criticisms, which sounds like a good deal to me!
I think you should read Bostrom’s actual paper for why this is a more compelling argument specifically when dealing with large risks. And it is worth noting that the reference class isn’t “blog posts criticizing institutions”—which I’m in favor of—it’s “blog posts attacking the credibility of the only institution that can feasibly respond to an incipient epidemic just as the epidemic is taking off and the public is unsure what to do about it.”
Is it your impression that the general public reads LessWrong?
What’s the model where an LW blogpost in any way undermines the CDC’s credibility with the general public?
It’s my impression that posts on lesswrong occasionally go viral, as has happened a couple times lately.
Thanks for the answer, that’s a fair point.
I would support a policy where, if an LW post starts to go viral, then original authors or mods are encouraged to add disclaimers to the top of posts that they wouldn’t otherwise need to add when writing for the LW audience. As SSC sometimes does.
I would not support a policy where LW authors always preemptively write for a general audience.
I’ve been away for some time. Any idea what posts he’s talking about here?
We end up on the frontpage of Reddit or HN from time to time. The last post that got a lot of clicks (15k+) was this one: https://www.lesswrong.com/posts/W5PhyEQqEWTcpRpqn/dunbar-s-function
Is there anything more recent? That post was 11 years ago.
It was on reddit like two weeks ago. LW posts have a long shelf life apparently.
Thanks for the suggestion! I just re-skimmed the Bostrom et al. paper (it’s been a while) and wrote up my thoughts in a top-level post.
Here we face the tragedy of “reference class tennis”. When you don’t know how much to trust your own reasoning vs. someone else’s, you might hope to defer the historical record for some suitable reference class of analogous disputes. But if you and your interlocutor disagree on which reference class is appropriate, then you just have the same kind of problem again.
I really don’t think this is a reference class tennis problem, given that I’m criticizing a specific post for specific reasons, not making an argument that we should judge this on the basis of a specific reference class.
And given that, I’m still seeing amazingly little engagement of the object level question of whether the criticisms I noted are valid.