Oh. I looked it over carefully twice (once in my RSS reader and once before commenting) to make sure IRC wasn't there since it was quite an omission, but now that I look a third time... I suppose that kind of illustrates the overall point well.

No IRC :(

To be immune to selection because it's part of intelligence would imply a strong genetic correlation. Aside from the fact that I am doubtful any such genetic correlation will ever be found (there is no noted phenotypic correlation of homosexuality & intelligence that I've heard of), this still has the issue that homosexuality ought to be decreasing noticeably over time: while intelligence has apparently been neutral or selected for over the past few millennia and so hypothetically could've slowed the selection against homosexuality, intelligence itself has been selected against for at least a century, so that would accelerate the selection now that there are fitness penalties for both homosexuality & intelligence (by a fair bit, because selection on continuous traits is much faster than selection on rare binary traits).

That theory is even worse than the inclusive fitness one because you offer no mechanism whatsoever to offset the huge fitness penalty.

Sexual imprinting is a highly successful evolved mechanism critical to reproductive fitness which does in fact succeed in the overwhelming majority of cases; in many ways, it is more important than trivial details like 'eating food' because at least an offspring which immediately starves to death doesn't drain parental resources and compete with siblings and the parents can try again! There should be a very good reason why such an important thing, found throughout the animal kingdom in far stupider & less sexually-dimorphic organisms, goes wrong in such a consistent way when other complex behaviors work at a higher rate and fail much more bluntly & chaotically. 'Random imprinting' is too weak a mechanism to thwart such a critical device, and doesn't explain why the errors do not rapidly disappear with general or sex-linked adaptations. (Even as a 5% liability-threshold binary trait, a reproductive fitness penalty of 50%, to be generous to a trait which involves active aversion to procreative sex, would imply it should be far lower now than when it first arose*.)

Further, such a random nonshared environment theory doesn't explain why dizygotic and monozygotic same-sex twins differ in concordance. (They don't differ in language, so your example is evidence against your imprinting theory.)

* https://www.researchgate.net/profile/J_Bailey2/publication/21311211_A_genetic_study_of_male_sexual_orientation/links/02e7e53c1a72a8a596000000.pdf gives a low end heritability estimate of 0.31; population prevalence among males is usually estimated ~5% giving a liability threshold of ~-1.64; homosexuality is amply documented for the past 2500 years or so, at least back to the ancient Greeks, which at a generation time of ~25 years, means 100 generations. So assuming a fitness penalty of 'just' half and that selection started only 100 generations ago (rather than much further back), we would expect the rate of homosexuality to be less than 1/5th what it is.

 threshold_select <- function(fraction_0, heritability, selection_intensity) {
library(VGAM) ## for 'probit'
fraction_probit_0 = probit(fraction_0)
## threshold for not manifesting trait:
s_0 = dnorm(fraction_probit_0) / fraction_0
## new rate after one selection:
fraction_1 = pnorm(fraction_probit_0 + heritability * s_0 * selection_intensity)
return(fraction_1)
}
threshold_select(0.95, 0.31, 0.5)
# [1] 0.9517116257
fractions <- 0.95
for (i in 2:100) { fractions[i] <- threshold_select(fractions[(i-1)], 0.31, 0.5); }
round(fractions, digits=3)
# [1] 0.950 0.952 0.953 0.955 0.956 0.958 0.959 0.960 0.961 0.963
# [11] 0.964 0.965 0.966 0.966 0.967 0.968 0.969 0.970 0.971 0.971
# [21] 0.972 0.973 0.973 0.974 0.974 0.975 0.975 0.976 0.977 0.977
# [31] 0.977 0.978 0.978 0.979 0.979 0.980 0.980 0.980 0.981 0.981
# [41] 0.981 0.982 0.982 0.982 0.983 0.983 0.983 0.983 0.984 0.984
# [51] 0.984 0.984 0.985 0.985 0.985 0.985 0.986 0.986 0.986 0.986
# [61] 0.986 0.987 0.987 0.987 0.987 0.987 0.987 0.988 0.988 0.988
# [71] 0.988 0.988 0.988 0.989 0.989 0.989 0.989 0.989 0.989 0.989
# [81] 0.989 0.990 0.990 0.990 0.990 0.990 0.990 0.990 0.990 0.990
# [91] 0.991 0.991 0.991 0.991 0.991 0.991 0.991 0.991 0.991 0.99

Aside from the inclusive fitness claim, Cochran's gay germ hypothesis is also consistent with the continued existence of homosexuality: the pathogen co-evolves and so while the genes do get selected against, which genes keeps changing. Unfortunately, his theory still remains something of a 'germ of the gaps' theory - no one's come up with a remotely plausible theory or found decent evidence that homosexuality spikes the fertility of relatives so much as to compensate for the sterility of homosexuals (remember, inclusive fitness decreases fast: if a homosexual has 1.05 rather than 2.1 children, then their siblings have to have 2.1 additional children, their cousins 4.2 additional children, and so on), so a theory which merely isn't contradicted by any evidence looks pretty good by comparison.

One thing I thought of which would be direct evidence for the infection theory: polygenic scores for homosexuality. It's somewhat heritable, so given a large sample size like UK Biobank, it should be possible to explain a few % of variance and construct a PGS based on a fairly narrow age cohort like 1 or 2 decades. Then the PGS can be applied longitudinally outside the sample. If it's pathogenic co-evolution and the relevant genes keep changing, then the homosexuality PGS should show highest predictive validity in the original age bracket, but then decrease steadily as one moves away from the age bracket into the past or toward the present, showing a clear inverted V shape. While polygenic scores can increase or decrease steadily or show sudden shocks for various reasons just like heritabilities can increase/decrease over time (eg education PGS decrease due to dysgenics, height PGS increase and so on), they don't typically show a distinct V shape, so finding one for homosexuality would be very striking.

Also a book excerpt. Books tend to be better.

https://www.gwern.net/newsletter/2017/02

I think they got in the single-digit thousands, perhaps 5-10,000, but I don't really recall.

There must have been power estimates done internally, but if there was one ever made public explaining how much power they expected from enrichment, I didn't hear about it. I won't pretend I know the details of what they were thinking sufficient to do my own power analysis, but I didn't think it was a terrible idea at the time; it was worth trying, and the results could always (I assumed) be meta-analyzed with later bigger results.

One interesting aspect of my analysis I would like to highlight is the part on multiple selection and genetic correlations. The immediate implication is that estimates of the value of embryo selection for IQ will be considerable underestimates if they ignore the many other traits that this selection will improve, and also that it is both feasible & desirable to make selection choices based on a weighted average of many polygenic scores. But this has had much broader implications for how I conceptualize the genetics of intelligence. (The following is based on too many papers to easily list at the moment, but if you read through my genetics bibliography compilation you'll find cites for a lot of these.)

I used to think that IQ variants were relatively neutral and specific to IQ, and variance in the population was maintained by selective neutrality (ie pro-IQ variants being too metabolically expensive or developmentally fragile to be selected for) and so arguments like in OP that 'we should describe IQ boosting as instead reducing stupidity or reducing the risk of intellectual disability' were, more or less, dishonest rhetorical tricks. (The ID claim is particularly questionable; most ID is from single mutations of large effect, stuff like embryo selection isn't going to override that.) Cochran had discussed the possibility of genetic load and 'grit in the gears' from rare variants, but the GCTAs indicated that most of the additive variance was explained by rather common genetic variants (common being >1% of the population having it) and whole-genome studies looking into de novo mutations and counting rare mutation load and finding it not hugely predictive eliminated that as an explanation. So it looked to me like it was more the case that the glass was half-full and there were 'genes for IQ' rather than 'lack of genes against IQ', and the highly general benefits across health & longevity were due to downstream effects like Gottfredson argued, in being able to take care of yourself, having a long-term perspective etc.

Then a twin study suggested that the health benefits were actually genetic; the 'generalist genes' hypothesis kept popping up in psychological traits like IQ with latent factors like overall brain size not fractionating into lots of more specific traits; the high-IQ D-F and GWAS studies failed to turn up any rare positive variants which ought to exist under neutrality; the GCTA estimates kept increasing when done using better measurements & better GCTA algorithms even for diseases that could not possibly be selectively neutral or beneficial in any way and must always be selected against; studies of recent human evolution over the past few thousand years demonstrate that common variants are constantly being selected for and against, implying considerable mutation load even on common variants, and even the harmful variants from the Neanderthals still haven't been purged in Europeans; rare variants are almost always harmful but surprisingly even common SNPs up to 50% frequency tend to be harmful too*; dysgenics has been confirmed; historical human population sizes imply poor purging of bad variants; and later I began poring over the intelligence & education genetic correlations that began pouring in thanks to GCTA & LD score regression. The correlations are almost all good (except for autism) regardless of behavior or organ or disease, to the extent it's very difficult to tell plausible stories about how higher income/education/intelligence could affect all of these simultaneously so much, and the intercorrelations go well beyond as correlations between diseases are everywhere as well. (I even tried some factor analysis to see if I could pull out a single grand factor.) In short, the 'bodily integrity' hypothesis appears to be explaining the big picture.

* for example, if you take the IQ and education polygenic scores from Benyamin et al 2014 and Okbay et al 2016 and simply sum the effects from each majority SNP variant (negating as necessary), corresponding to a hypothetical modal person, both scores are >2SDs, even though you would expect ~0 since it's far from obvious that the 25% frequency version of a random SNP can be a priori expected to be harmful to education or intelligence; and a linear regression also shows that higher frequencies predict better effects. I don't know how general this is but I'm definitely curious now.

So under this scenario, what we see is not good genes 'for' intelligence so much as a high level of genetic load from lots of unexpectedly common broken genes floating around the population which haven't been able to be purged due to small individual bad effects, small effective human population sizes, fast-changing environments, introgression from other hominids like the Neanderthals, which by degrading 'upstream' biological systems like mitochondria or key proteins then have global downstream effects across the whole body & mind (regardless of conditioning on measured IQ), producing these broad genetic and phenotypic correlations between intelligence & everything under the sun. Intelligence, which so far has only been reified by neuroscience & neuroimaging down to very global brain traits like overall speed & connectivity & white-matter integrity, pops up everywhere because it is the most fragile trait, affected by coordination between the most bodily systems, more easily pushed off-kilter and degraded than traits like eye color or height. Perhaps Cochran was right all along that someone with the modal human would be much smarter, healthier, saner, and happier than the rest of us - he was just wrong in thinking the mutation load would be in rare and de novo mutations, when most of it has been lurking in common variants all along, and we're all far more screwed up than anyone guessed.

From this perspective, the fact that doing embryo selection against schizophrenia will usually also be embryo selection for intelligence, and be safe to do without nasty unintended effects, is not a surprise at all. Of course it would tend to reduce the chance of schizophrenia; it would also reduce other mental illnesses like depression or anorexia or bipolar, as well as behavioral problems like BMI, or cardiovascular problems, or...

It's not about playing God or 'being better than well', it's just realizing that no one was 'well' to begin with (anymore than people centuries ago were 'well' rather than all, even the elites, stunted from malnutrition and parasites and pandemic childhood disease and hard labor) and everyone is carrying a considerable burden of many thousands of broken variants, and some people by chance & inheritance have fewer bad variants than others and 'healthy' people merely are very similarly sick due to lots of small variants adding up to a tight normal distribution and the variance written off as simple normal variation which no one can do anything about and shouldn't worry anyone. It would be perfectly honest to describe this as trying to cure 'micro-disabilities' or 'micro-illnesses' (hey, if we can have 'micromorts' or 'microaggressions', why not?) rather than 'enhancement' or adopt a slogan like 'leveling the genetic playing field'.

If everyone is suffering these micro-sicknesses, the precautionary argument drops away, as do the fears of some suitably ironic cosmic punishment for tampering with the genome, the worries about 'selecting only for one thing', and maybe some of the Puritan objections to 'enhancement' or 'cheating'. It also provides a powerful informative prior for selection and synthesis: if in doubt, choose the most common variant.