check in

[Ultimately_Untrue_Thought.git] / content / drafts / sexual-dimorphism-in-the-sequences-in-relation-to-my-gender-problems.md

diff --git a/content/drafts/sexual-dimorphism-in-the-sequences-in-relation-to-my-gender-problems.md b/content/drafts/sexual-dimorphism-in-the-sequences-in-relation-to-my-gender-problems.md
index a16d05f..6ad81a4 100644 (file)
--- a/content/drafts/sexual-dimorphism-in-the-sequences-in-relation-to-my-gender-problems.md
+++ b/content/drafts/sexual-dimorphism-in-the-sequences-in-relation-to-my-gender-problems.md
@@ -172,7 +172,7 @@ Notably, for _images_ of faces, we actually _do_ have transformation technology!
 
 If you let each pixel vary independently, the space of possible 1024x1024 images is 1,048,576-dimensional, but the vast hypermajority of those images aren't photorealistic human faces. Letting each pixel vary independently is the wrong way to think about it: changing the lighting or pose changes a lot of pixels in what humans would regard as images of "the same" face. So instead, our machine-learning algorithms learn a [compressed](https://www.lesswrong.com/posts/ex63DPisEjomutkCw/msg-len) representation of what makes the tiny subspace (relative to images-in-general) of _faces in particular_ similar to each other. That [latent space](https://towardsdatascience.com/understanding-latent-space-in-machine-learning-de5a7c687d8d) is a lot smaller (say, 512 dimensions), but still rich enough to embed the distinctions that humans notice: [you can find a hyperplane that separates](https://youtu.be/dCKbRCUyop8?t=1433) smiling from non-smiling faces, or glasses from no-glasses, or young from old, or different races—or female and male. Sliding along the [normal vector](https://en.wikipedia.org/wiki/Normal_(geometry)) to that [hyperplane](https://en.wikipedia.org/wiki/Hyperplane) gives the desired transformation: producing images that are "more female" (as the model has learned that concept) while keeping "everything else" the same.
 
 
 If you let each pixel vary independently, the space of possible 1024x1024 images is 1,048,576-dimensional, but the vast hypermajority of those images aren't photorealistic human faces. Letting each pixel vary independently is the wrong way to think about it: changing the lighting or pose changes a lot of pixels in what humans would regard as images of "the same" face. So instead, our machine-learning algorithms learn a [compressed](https://www.lesswrong.com/posts/ex63DPisEjomutkCw/msg-len) representation of what makes the tiny subspace (relative to images-in-general) of _faces in particular_ similar to each other. That [latent space](https://towardsdatascience.com/understanding-latent-space-in-machine-learning-de5a7c687d8d) is a lot smaller (say, 512 dimensions), but still rich enough to embed the distinctions that humans notice: [you can find a hyperplane that separates](https://youtu.be/dCKbRCUyop8?t=1433) smiling from non-smiling faces, or glasses from no-glasses, or young from old, or different races—or female and male. Sliding along the [normal vector](https://en.wikipedia.org/wiki/Normal_(geometry)) to that [hyperplane](https://en.wikipedia.org/wiki/Hyperplane) gives the desired transformation: producing images that are "more female" (as the model has learned that concept) while keeping "everything else" the same.
 

-Two-dimensional _images_ of people are _vastly_ simpler than the actual people themselves in the real physical universe. But _in theory_, a lot of the same _mathematical principles_ would apply to hypothetical future nanotechnology-wielding AI systems that could synthesize a human being from scratch (this-person-_didn't_-exist-dot-com?), or do a real-world sex transformation (PersonApp?)—and the same statistical morals apply to reasoning about sex differences in psychology and (which is to say) the brain.
+Two-dimensional _images_ of people are _vastly_ simpler than the actual people themselves in the real physical universe. But _in theory_, a lot of the same _mathematical principles_ would apply to hypothetical future nanotechnology-wielding AI systems that could, like the AI in "Failed Utopia #4-2", synthesize a human being from scratch (this-person-_didn't_-exist-dot-com?), or do a real-world sex transformation (PersonApp?)—and the same statistical morals apply to reasoning about sex differences in psychology and (which is to say) the brain.

 
 Daphna Joel _et al._ [argue](https://www.pnas.org/content/112/50/15468) [that](https://www.pnas.org/content/112/50/15468) human brains are "unique 'mosaics' of features" that cannot be categorized into distinct _female_ and _male_ classes, because it's rare for brains to be "internally consistent"—female-typical or male-typical along _every_ dimension. It's true and important that brains aren't _discretely_ sexually dimorphic the way genitals are, but as [Marco del Guidice _et al._ point out](http://cogprints.org/10046/1/Delgiudice_etal_critique_joel_2015.pdf), the "cannot be categorized into two distinct classes" claim seems false in an important sense. The lack of "internal consistency" in Joel _et al._'s sense is exactly the behavior we expect from multivariate normal-ish distributions with different-but-not-vastly-different means. (There aren't going to be many traits where the sexes are like, _four_ or whatever standard deviations apart.) It's just like how sequences of flips of a very Heads-biased and very Tails-biased coin are going to be unique "mosaics" of Heads and Tails, but pretty distinguishable with enough flips—and indeed, with the right stats methodology, [MRI scans can predict sex at 96.8% accuracy](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6374327/).
 
 
 Daphna Joel _et al._ [argue](https://www.pnas.org/content/112/50/15468) [that](https://www.pnas.org/content/112/50/15468) human brains are "unique 'mosaics' of features" that cannot be categorized into distinct _female_ and _male_ classes, because it's rare for brains to be "internally consistent"—female-typical or male-typical along _every_ dimension. It's true and important that brains aren't _discretely_ sexually dimorphic the way genitals are, but as [Marco del Guidice _et al._ point out](http://cogprints.org/10046/1/Delgiudice_etal_critique_joel_2015.pdf), the "cannot be categorized into two distinct classes" claim seems false in an important sense. The lack of "internal consistency" in Joel _et al._'s sense is exactly the behavior we expect from multivariate normal-ish distributions with different-but-not-vastly-different means. (There aren't going to be many traits where the sexes are like, _four_ or whatever standard deviations apart.) It's just like how sequences of flips of a very Heads-biased and very Tails-biased coin are going to be unique "mosaics" of Heads and Tails, but pretty distinguishable with enough flips—and indeed, with the right stats methodology, [MRI scans can predict sex at 96.8% accuracy](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6374327/).