So I’ve been watching youtubers interview Steve Hsu. The first interview I saw was done by some Israeli, the second one I watched was by a guy of South Asian descent, and the third was some white guy who seemed to think he was Buddha.

Steve made a lot of important points, some of which I’ve discussed before.

When Steve first entered this field he feared that traits like height and IQ would be too non-additive to decode and too pleiotropic to edit. Pleiotropy is when a genomic variant affects more than one seemingly unrelated phenotype, for example one theory is that high IQ kids wear glasses because the genes for IQ also cause myopia.

Luckily, genetic architecture is overwhelmingly additive and with over 3 billion base-pairs in the genome, pleiotropy is not that bad.

The additive nature of the genome has been long understood by animal breeders and was formally explained in the famous Fisher Theorem in the 1930s. Put simply, phenotypes that are caused by additive genes are favored by natural selection because they’re easier to pass on. That’s because we get a random sample of mom and dad’s genes, so if a particular trait requires an interaction of several genes, it’s unlikely we’ll get all of them so what good are any of them? It’s thus much better to have every gene (genomic variant) having at least a small effect, independent of other genes in its nexus.

At least for white people living in the West, Steve can predict your height from your DNA with a correlation of r = 0.64. That’s actually quite incredible considering he’s limited to only common genetic variants (who knows how much additional variance there is in rare variants and non-additive ones). How high will the correlation get when the whole genome becomes cheap enough to sequence in huge numbers?

Unfortunately the correlation drops when he tries to predict height in South Asia. Let’s say you have a gene that causes you to like milk and milk makes you grow tall. This gene will help predict your height in the West but perhaps not in India where milk is scarce so maybe stuff like that is why the correlation declines.

Of course it could also be that races differ in genetic architecture but Steve assumes they are the same (not sure why this should be the default assumption since we know, for example, whites and East Asians have different genes for white skin)

If they are the same, then Steve needs some international samples to force the machine learning to find truly causal equations that transcend culture and I would like to try these equations on ancient DNA to find out whether the decline in height (and brain size) during the Holocene was genetic or environmental.

Right now Steve can only predict IQ (within countries) with an accuracy of around 0.4 (he says) but that’s only because in the age of wokeism, it’s virtually impossible to sequence large samples of people who have taken quality IQ tests. He can see the accuracy trend-line is still rising as sample size increases, unlike his height predictor which already has such large samples that it has plateaued.

Once IQ predictors become as good as height, we’ll see a massive increase in average IQ and height as rich couples will use surrogate mothers to produced 100 fertilized eggs and only the top 1% from each couples’ eggs will be chosen.

I feel bad for my nieces and nephews (and RR’s baby) because they’re about to become part of a genetic underclass. Already 10% of Denmark babies are born through in vitro insemination. Within the next 30 years, those who are not will find themselves six inches shorter and 30 IQ points dumber than the youngest adults. And on top of that they might also be more ugly and less healthy. It’s even conceivable that life span will increase to 300 years.

Already the pace of technological progress has been rapid over the last hundred years, despite the fact that genetic IQ has been static or declining. Now just imagine how fast technology will progress when cultural evolution is combined with artificial high speed biological evolution.