Women are better at verbal ability and social cognition, whereas [men are better at visuospatial skills](http://zackmdavis.net/blog/2016/12/alpha-gamma-phi/). The sexes achieve similar levels of overall performance via somewhat different mental "toolkits." Murray devotes a section to a 2007 result of Johnson and Bouchard, who report that general intelligence ["masks the dimensions on which [sex differences in mental abilities] lie"](/papers/johnson-bouchard-sex_differences_in_mental_abilities_g_masks_the_dimensions.pdf): people's overall skill in using tools from the metaphorical mental toolbox leads to underestimates of differences in toolkits (that is, nonmetaphorically, the effect sizes of sex differences in specific mental abilities), which you want to statistically correct for. This result in particular is _super gratifying_ to me personally, because [I independently had a very similar idea a few months back](/2019/Sep/does-general-intelligence-deflate-standardized-effect-sizes-of-cognitive-sex-differences/)—it's _super validating_ as an amateur to find that the pros have been thinking along the same track!
-The second part of the book is about some ways in which people with different ancestries are different from each other! Obviously, there are no "distinct" "races" (that would be dumb), but it turns out (as found by endeavors such as [Li _et al._ 2008](/papers/li_et_al-worldwide_human_relationships_inferred.pdf)) that when you throw clustering and [dimensionality-reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction) algorithms at SNP data (<em>s</em>ingle <em>n</em>ucleotide <em>p</em>olymorphisms, places in the genome where more than one allele has non-negligible frequency), you get groupings that are a pretty good match to classical or self-identified "races".
+<a id="ancestries"></a>The second part of the book is about some ways in which people with different ancestries are different from each other! Obviously, there are no "distinct" "races" (that would be dumb), but it turns out (as found by endeavors such as [Li _et al._ 2008](/papers/li_et_al-worldwide_human_relationships_inferred.pdf)) that when you throw clustering and [dimensionality-reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction) algorithms at SNP data (<em>s</em>ingle <em>n</em>ucleotide <em>p</em>olymorphisms, places in the genome where more than one allele has non-negligible frequency), you get groupings that are a pretty good match to classical or self-identified "races".
Ask the computer to assume that an individual's ancestry came from _K_ fictive ancestral populations where _K_ := 2, and it'll infer that sub-Saharan Africans are descended entirely from one, East Asians and some native Americans are descended entirely from the other, and everyone else is an admixture. But if you set _K_ := 3, populations from Europe and the near East (which were construed as admixtures in the _K_ := 2 model) split off as a new inferred population cluster. And so on.
And sufficient suspicion makes communication nearly impossible. (If you _know_ someone is lying, their words mean nothing, [not even as the opposite of the truth](https://www.lesswrong.com/posts/qNZM3EGoE5ZeMdCRt/reversed-stupidity-is-not-intelligence).) As far as many of Murray's detractors are concerned, it almost doesn't matter what the text of _Human Diversity_ says, how meticulously researched of a psychology/neuroscience/genetics lit review it is. From their perspective, Murray is "hiding the ball": they're not mad about _this_ book; they're mad about specifically chapters 13 and 14 of a book Murray coauthored twenty-five years ago. (I don't think I'm claiming to be a mind-reader here; the first 20% of [_The New York Times_'s review of _Human Diversity_](https://archive.is/b4xKB) is pretty explicit and representative.)
-In 1994's _The Bell Curve: Intelligence and Class Structure in American Life_, Murray and coauthor Richard J. Herrnstein argued that a lot of variation in life outcomes is explained by variation in intelligence. Some people think that folk concepts of "intelligence" or being "smart" are ill-defined and therefore not a proper object of scientific study. But that hasn't stopped some psychologists from trying to construct tests purporting to measure an "intelligence quotient" (or _IQ_ for short). It turns out that if you give people a bunch of different mental tests, the results all positively correlate with each other: people who are good at one mental task, like listening to a list of numbers and repeating them backwards ("reverse digit span"), are also good at others, like knowing what words mean ("vocabulary"). There's a lot of fancy linear algebra involved, but basically, you can visualize people's test results as a hyper[ellipsoid](https://en.wikipedia.org/wiki/Ellipsoid) in some high-dimensional space where the dimensions are the different tests. (I rely on this ["configuration space"](https://www.lesswrong.com/posts/WBw8dDkAWohFjWQSk/the-cluster-structure-of-thingspace) visual metaphor _so much_ for _so many_ things that when I started [my secret ("secret") gender blog](/), it felt right to put it under a `.space` [TLD](https://en.wikipedia.org/wiki/Top-level_domain).) The longest axis of the hyperellipsoid corresponds to the "_g_ factor" of "general" intelligence—the choice of axis that cuts through the most variance in mental abilities.
+In 1994's _The Bell Curve: Intelligence and Class Structure in American Life_, Murray and coauthor Richard J. Herrnstein argued that a lot of variation in life outcomes is explained by variation in intelligence. Some people think that folk concepts of "intelligence" or being "smart" are ill-defined and therefore not a proper object of scientific study. But that hasn't stopped some psychologists from trying to construct tests purporting to measure an "intelligence quotient" (or _IQ_ for short). It turns out that if you give people a bunch of different mental tests, the results all positively correlate with each other: people who are good at one mental task, like listening to a list of numbers and repeating them backwards ("reverse digit span"), are also good at others, like knowing what words mean ("vocabulary"). There's a lot of fancy linear algebra involved, but basically, you can visualize people's test results as a hyper[ellipsoid](https://en.wikipedia.org/wiki/Ellipsoid) in some high-dimensional space where the dimensions are the different tests. (I rely on this ["configuration space"](https://www.lesswrong.com/posts/WBw8dDkAWohFjWQSk/the-cluster-structure-of-thingspace) visual metaphor _so much_ for _so many_ things that when I started [my secret ("secret") gender blog](/), it felt right to put it under a `.space` [TLD](https://en.wikipedia.org/wiki/Top-level_domain).) <a id="the-length-of-a-hyperellipsoid"></a>The longest axis of the hyperellipsoid corresponds to the "_g_ factor" of "general" intelligence—the choice of axis that cuts through the most variance in mental abilities.
It's important not to overinterpret the _g_ factor as some unitary essence of intelligence rather than the length of a hyperellipsoid. It seems likely that [if you gave people a bunch of _physical_ tests, they would positively correlate with each other](https://www.talyarkoni.org/blog/2010/03/07/what-the-general-factor-of-intelligence-is-and-isnt-or-why-intuitive-unitarianism-is-a-lousy-guide-to-the-neurobiology-of-higher-cognitive-ability/), such that you could extract a ["general factor of athleticism"](https://isteve.blogspot.com/2007/09/g-factor-of-sports.html). (It would be really interesting if anyone's actually done this using the same methodology used to construct IQ tests!) But _athleticism_ is going to be an _very_ "coarse" construct for which [the tails come apart](https://www.lesswrong.com/posts/dC7mP5nSwvpL65Qu5/why-the-tails-come-apart): for example, world champion 100-meter sprinter Usain Bolt's best time in the _800_ meters is [reportedly only around 2:10](https://www.newyorker.com/sports/sporting-scene/how-fast-would-usain-bolt-run-the-mile) [or 2:07](https://archive.is/T988h)! (For comparison, _I_ ran a 2:08.3 in high school once!)
I used to be a naïve egalitarian. I was very passionate about it. I was eighteen years old. I am—again—still fond of the moral sentiment, and eager to renormalize it into something that makes sense. (Some egalitarian anxieties do translate perfectly well into the Bayesian setting, as I'll explain in a moment.) But the abject horror I felt at eighteen at the mere suggestion of _making generalizations_ about _people_ just—doesn't make sense. It's not even that it _shouldn't_ be practiced (it's not that my heart wasn't in the right place), but that it _can't_ be practiced—that the people who think they're practicing it are just confused about how their own minds work.
-Give people photographs of various women and men and ask them to judge how tall the people in the photos are, as [Nelson _et al._ 1990 did](/papers/nelson_et_al-everyday_base_rates_sex_stereotypes_potent_and_resilient.pdf), and people's guesses reflect both the photo-subjects' actual heights, but also (to a lesser degree) their sex. Unless you expect people to be perfect at assessing height from photographs (when they don't know how far away the cameraperson was standing, aren't ["trigonometrically omniscient"](https://plato.stanford.edu/entries/logic-epistemic/#LogiOmni), _&c._), this behavior is just _correct_: men really are taller than women on average, so P(true-height|apparent-height, sex) ≠ P(height|apparent-height) [because of](https://humanvarieties.org/2017/07/01/measurement-error-regression-to-the-mean-and-group-differences/) [regression to the mean](https://en.wikipedia.org/wiki/Regression_toward_the_mean) (and women and men regress to different means). But [this all happens subconsciously](/2020/Apr/peering-through-reverent-fingers/): in the same study, when the authors tried height-matching the photographs (for every photo of a woman of a given height, there was another photo in the set of a man of the same height) _and telling_ the participants about the height-matching _and_ offering a cash reward to the best height-judge, more than half of the stereotyping effect remained. It would seem that people can't consciously readjust their learned priors in reaction to verbal instructions pertaining to an artificial context.
+<a id="everyday-base-rates"></a>Give people photographs of various women and men and ask them to judge how tall the people in the photos are, as [Nelson _et al._ 1990 did](/papers/nelson_et_al-everyday_base_rates_sex_stereotypes_potent_and_resilient.pdf), and people's guesses reflect both the photo-subjects' actual heights, but also (to a lesser degree) their sex. Unless you expect people to be perfect at assessing height from photographs (when they don't know how far away the cameraperson was standing, aren't ["trigonometrically omniscient"](https://plato.stanford.edu/entries/logic-epistemic/#LogiOmni), _&c._), this behavior is just _correct_: men really are taller than women on average, so P(true-height|apparent-height, sex) ≠ P(true-height|apparent-height) [because of](https://humanvarieties.org/2017/07/01/measurement-error-regression-to-the-mean-and-group-differences/) [regression to the mean](https://en.wikipedia.org/wiki/Regression_toward_the_mean) (and women and men regress to different means). But [this all happens subconsciously](/2020/Apr/peering-through-reverent-fingers/): in the same study, when the authors tried height-matching the photographs (for every photo of a woman of a given height, there was another photo in the set of a man of the same height) _and telling_ the participants about the height-matching _and_ offering a cash reward to the best height-judge, more than half of the stereotyping effect remained. It would seem that people can't consciously readjust their learned priors in reaction to verbal instructions pertaining to an artificial context.
Once you understand at a _technical_ level that probabilistic reasoning about demographic features is both epistemically justified, _and_ implicitly implemented as part of the way your brain processes information _anyway_, then a moral theory that forbids this starts to look less compelling? Of course, statistical discrimination on demographic features is only epistemically justified to exactly the extent that it helps _get the right answer_. Renormalized-egalitarians can still be properly outraged about the monstrous tragedies where I have moral property P but I _can't prove it to you_, so you instead guess _incorrectly_ that I don't just because other people who look like me mostly don't, and you don't have any better information to go on—or tragedies in which a feedback loop between predictions and social norms creates or amplifies group differences that wouldn't exist under some other social equilibrium.
projects / Ultimately_Untrue_Thought.git / blobdiff