even buying all the arguments of the race/IQ chaps hook, line, and sinker (which, to reiterate, neither I nor any rationalist I’ve ever spoken to or read actually does)
In public. Absent some kind of epic infosec fail (like a disgrunted acquaintance leaking private emails), you can’t know what beliefs people might have that they’re incentivized not to talk about. To this you might reply, “Absence of evidence is evidence of absence: without a reason to think that people are holding back, you could just as baselessly speculate that maybe rationalists are secretly Satanists or flat-earthers.”
Is there any such reason? I don’t know. There is an interesting Scott Alexander post, “Kolmogorov Complicity And The Parable Of Lightning”, about the pragmatic necessity for intellectuals to avoid contradicting powerful orthodoxies. But doesn’t that seem like a weirdly specific abstract topic for Scott to write about, unless he had some concrete contemporary example in mind? What do you think he was thinking of?
the differences are tiny and absolutely dwarfed by within-race variance anyway, so just wouldn’t make much difference to anything.
Sort of. Statistical group differences can be large enough to have social consequences in at least some contexts, while also being small enough such that the correct Bayesian update about an individual based on group membership is small (not zero, but small). For cognitive ability measures, the current black–white gap in the U.S. is about 0.85 standard deviations (Cohen’s d), and the white–Asian gap is at about d ≈ 0.3. Is that “tiny”? I don’t know; depends on how you want to use words. For comparison, the male–female height gap is about d ≈ 1.5. Physical sex differences like the fact that men are taller than women on average seem like the kind of background common knowledge that is sometimes decision-relevant (e.g., to product designers) even if tall women exist and you can notice someone’s height independently of their sex (um, almost; empirically, your brain does do a Bayesian group-membership base rate adjustment without your conscious awareness).
If intelligence is a Gaussian distribution then small differences in the mean result in massive differences in relative representation at the tails. A small difference in average performance can result in overwhelming differences within elite groups while simultaneously being unimportant to most people.
I think both of these are reasonable responses. I brushed over a lot of nuance in this section because I just didn’t want this part of the discussion to dominate. I realise that the raw population differences (in the US) are quite chunky (though d=0.83 is higher than I have heard), but what I glossed over completely is that the only not-obviously-silly investigations of the issue I’ve seen do admit that clearly there are environmental confounders, and then claim that those confounders only account for some of that difference, and that some genetic component remains. These estimations of genetic component (which, to my vague memory, were in the d=0.1-0.3 range) are what I was calling ‘tiny’.
However I now realise I was being a bit inconsistent in also saying that I’ve never seen rationalists endorse this, because actually I have seen rationalists endorse that tiny effect, but never the absurd-on-its-face idea that the entire observed gap is genetic. So I’m using one version of the hypothesis when refuting one step in the argument, and another version in refuting another step. This was wrong of me.
Perhaps even calling the smaller estimate ‘tiny’ is still a bit harsh because, as Isusr says, this is still enough for the far ends of the distribution to look very different. So I think the best thing is that I drop this entire size-related part of my argument.
Zack is also right that there is a bit of hard-to-observe going on in my ‘no rationalists actually believe this’ argument as well. I think I’m on solid ground in saying that I don’t think more than a negligible number of rationalists by the hardcore all-the-observed-differences-are-genetic interpretation, but perhaps many are privately convinced there’s some genetic component—I wouldn’t know. I don’t think Kolmogorov Complicity is about that, I think it’s about gender issues in CS. But I think the whole point is one can’t really tell, so this is also a fair point.
So, on reflection, I drop the ‘no rationalists buy these claims’ and ‘the differences are tiny anyway’ parts of my argument as they’re somewhat based on different assumptions about what the claims are, and both have their own problems. I will rest my entire ‘the race bit is silly’ position upon the much more solid grounds of ‘but intelligence isn’t rationality’ and ‘even if it were, it wouldn’t imply what you imply it implies about who should have political power’, which are both problems I have independently with the rest of the piece.
I think a very long, drawn-out argument could still rescue something from my other two points and show that this part of the piece is even weaker than would be implied from just those two errors, but I don’t really want to bother because it would be complicated and difficult, the failure of the argument is overdetermined anyway, and talking about it makes me irritable and upset and poses non-negligible risk to the entire community so I just don’t see it as worth it in this case.
In public. Absent some kind of epic infosec fail (like a disgrunted acquaintance leaking private emails), you can’t know what beliefs people might have that they’re incentivized not to talk about. To this you might reply, “Absence of evidence is evidence of absence: without a reason to think that people are holding back, you could just as baselessly speculate that maybe rationalists are secretly Satanists or flat-earthers.”
Is there any such reason? I don’t know. There is an interesting Scott Alexander post, “Kolmogorov Complicity And The Parable Of Lightning”, about the pragmatic necessity for intellectuals to avoid contradicting powerful orthodoxies. But doesn’t that seem like a weirdly specific abstract topic for Scott to write about, unless he had some concrete contemporary example in mind? What do you think he was thinking of?
Sort of. Statistical group differences can be large enough to have social consequences in at least some contexts, while also being small enough such that the correct Bayesian update about an individual based on group membership is small (not zero, but small). For cognitive ability measures, the current black–white gap in the U.S. is about 0.85 standard deviations (Cohen’s d), and the white–Asian gap is at about d ≈ 0.3. Is that “tiny”? I don’t know; depends on how you want to use words. For comparison, the male–female height gap is about d ≈ 1.5. Physical sex differences like the fact that men are taller than women on average seem like the kind of background common knowledge that is sometimes decision-relevant (e.g., to product designers) even if tall women exist and you can notice someone’s height independently of their sex (um, almost; empirically, your brain does do a Bayesian group-membership base rate adjustment without your conscious awareness).
If intelligence is a Gaussian distribution then small differences in the mean result in massive differences in relative representation at the tails. A small difference in average performance can result in overwhelming differences within elite groups while simultaneously being unimportant to most people.
I think both of these are reasonable responses. I brushed over a lot of nuance in this section because I just didn’t want this part of the discussion to dominate. I realise that the raw population differences (in the US) are quite chunky (though d=0.83 is higher than I have heard), but what I glossed over completely is that the only not-obviously-silly investigations of the issue I’ve seen do admit that clearly there are environmental confounders, and then claim that those confounders only account for some of that difference, and that some genetic component remains. These estimations of genetic component (which, to my vague memory, were in the d=0.1-0.3 range) are what I was calling ‘tiny’.
However I now realise I was being a bit inconsistent in also saying that I’ve never seen rationalists endorse this, because actually I have seen rationalists endorse that tiny effect, but never the absurd-on-its-face idea that the entire observed gap is genetic. So I’m using one version of the hypothesis when refuting one step in the argument, and another version in refuting another step. This was wrong of me.
Perhaps even calling the smaller estimate ‘tiny’ is still a bit harsh because, as Isusr says, this is still enough for the far ends of the distribution to look very different. So I think the best thing is that I drop this entire size-related part of my argument.
Zack is also right that there is a bit of hard-to-observe going on in my ‘no rationalists actually believe this’ argument as well. I think I’m on solid ground in saying that I don’t think more than a negligible number of rationalists by the hardcore all-the-observed-differences-are-genetic interpretation, but perhaps many are privately convinced there’s some genetic component—I wouldn’t know. I don’t think Kolmogorov Complicity is about that, I think it’s about gender issues in CS. But I think the whole point is one can’t really tell, so this is also a fair point.
So, on reflection, I drop the ‘no rationalists buy these claims’ and ‘the differences are tiny anyway’ parts of my argument as they’re somewhat based on different assumptions about what the claims are, and both have their own problems. I will rest my entire ‘the race bit is silly’ position upon the much more solid grounds of ‘but intelligence isn’t rationality’ and ‘even if it were, it wouldn’t imply what you imply it implies about who should have political power’, which are both problems I have independently with the rest of the piece.
I think a very long, drawn-out argument could still rescue something from my other two points and show that this part of the piece is even weaker than would be implied from just those two errors, but I don’t really want to bother because it would be complicated and difficult, the failure of the argument is overdetermined anyway, and talking about it makes me irritable and upset and poses non-negligible risk to the entire community so I just don’t see it as worth it in this case.