-An illustrative example: like many gender-dysphoric males, I cosplay female characters at conventions sometimes. And, unfortunately, like many gender-dysphoric males, I'm *not very good at it*. I think someone looking at [my cosplay photos](https://www.facebook.com/zmdavis/media_set?set=a.10155131901020199&type=3) and trying to describe their content in clear language—not trying to be nice to anyone or make a point, but just trying to use language as a map that reflects the territory—would say something like, "This is a photo of a man and he's wearing a dress." The word *man* in that sentence is expressing *cognitive work*: it's a summary of the [lawful cause-and-effect evidential entanglement](https://www.lesswrong.com/posts/6s3xABaXKPdFwA3FS/what-is-evidence) whereby the photons reflecting off the photograph are correlated with photons reflecting off my body at the time the photo was taken, which are correlated with my externally-observable secondary sex characteristics (facial structure, beard shadow, *&c.*), from which evidence an agent using an [efficient naïve-Bayes-like model](http://lesswrong.com/lw/o8/conditional_independence_and_naive_bayes/) can assign me to its "man" category and thereby make probabilistic predictions about some of my traits that aren't directly observable from the photo, and achieve a better [score on those predictions](http://yudkowsky.net/rational/technical/) than if the agent had assigned me to its "woman" category, where by "traits" I mean not *just* chromosomes ([as you suggested on Twitter](https://twitter.com/ESYudkowsky/status/1067291243728650243)), but the *conjunction* of chromosomes *and* reproductive organs _and_ muscle mass (sex difference effect size of [Cohen's *d*](https://en.wikipedia.org/wiki/Effect_size#Cohen's_d)≈2.6) *and* Big Five Agreeableness (*d*≈0.5) *and* Big Five Neuroticism (*d*≈0.4) *and* short-term memory (*d*≈0.2, favoring women) *and* white-to-gray-matter ratios in the brain *and* probable socialization history *and* [lots of other things](https://en.wikipedia.org/wiki/Sex_differences_in_human_physiology)—including differences we might not necessarily currently know about, but have prior reasons to suspect exist: no one _knew_ about sex chromosomes before 1905, but given all the other systematic differences between women and men, it would have been a reasonable guess (that turned out to be correct!) to suspect the existence of some sort of molecular mechanism of sex determination.
-
-Making someone say "trans woman" instead of "man" in that sentence depending on my verbally self-reported self-identity may not be forcing them to *lie*. But it *is* forcing them to obfuscate the probabilistic inference they were trying to communicate with the original sentence (about modeling the person in the photograph as being sampled from the "men" [cluster in configuration space](https://www.lesswrong.com/posts/WBw8dDkAWohFjWQSk/the-cluster-structure-of-thingspace)), and instead use language that suggests a different cluster-structure ("trans women", two words, are presumably a subcluster within the "women" cluster). This encoding might not confuse a well-designed AI into making any bad predictions, but [as you explained very clearly, it probably will confuse humans](https://www.lesswrong.com/posts/veN86cBhoe7mBxXLk/categorizing-has-consequences):
-
-> You can see this in terms of similarity clusters: once you draw a boundary around a group, the mind starts trying to harvest similarities from the group. And unfortunately the human pattern-detectors seem to operate in such overdrive that we see patterns whether they're there or not; a weakly negative correlation can be mistaken for a strong positive one with a bit of selective memory.
-
-(I _want_ to confidently predict that everything I've just said is completely obvious to you, because I learned it all specifically from you! A 130 IQ _nobody_ like me shouldn't have to say _any_ of this to the _author_ of "A Human's Guide to Words"! But then I don't know how to reconcile that with your recent public statement about [not seeing "how there's scientific truth at stake"](https://twitter.com/ESYudkowsky/status/1067482047126495232). Hence this desperate and [_confused_](https://www.lesswrong.com/posts/5JDkW4MYXit2CquLs/your-strength-as-a-rationalist) email plea.)
-
-In your email of 29 November, you wrote, "I hope I would have noticed if I had tweeted anything asserting [Zack's] factual statements to be factually false, since that would imply knowledge I don't claim to have," and in your Twitter [reply link](https://twitter.com/ESYudkowsky/status/1068071036732694529) to my post (thanks!!), you wrote, "[w]ithout yet judging its empirical content." However, as Michael emphasized ("That's not what Zach is talking about at all and not what the debate is about and you know this"), the _main_ point I'm trying to make is a philosophical one, not an empirical one: that category boundaries and associated language are not arbitrary (if you care about human intelligence being useful), and that sex (or "gender") is no exception.
-
-I _also_ made some empirical claims favoring Blanchard's two-type homosexual/autogynephilic model of MtF (as we discussed in 2016). But as I _tried_ to make clear in the post (I fear I am not as good of a writer as you, and perhaps should have put in more effort to make it clearer, but see [footnote 10](http://unremediatedgender.space/2018/Feb/the-categories-were-made-for-man-to-make-predictions/#note-10)), I don't think you _need_ the full autogynephilia theory to show that the categories-are-not-arbitrary point has implications for discourse on transgender issues, and I think you _already_ have enough evidence—if [used efficiently](https://www.lesswrong.com/posts/MwQRucYo6BZZwjKE7/einstein-s-arrogance)—to see that the distribution of actual trans people we know is such that the categories-are-not-abritrary point is relevant in practice.
-
-Consider again the 6.7:1 (!!) cis-woman-to-trans-woman ratio among 2018 _Slate Star Codex_ survey respondents. The ratio in the general population is going to be more like 86:1 (estimate derived from dividing 50% (female share of population according to [Fisher's principle](https://en.wikipedia.org/wiki/Fisher%27s_principle)) by 0.58% (trans share of U.S. population according to a [2016 report](http://williamsinstitute.law.ucla.edu/wp-content/uploads/How-Many-Adults-Identify-as-Transgender-in-the-United-States.pdf))).