I would push back on your first couple of sentences, which imply that heritability of intelligence (or anything else) is a fixed property of a phenotype that we are trying to estimate as accurately as possible. It isn’t: it’s a ratio of genetic and phenotypic variances, and its value depends on the magnitude of both of them. The heritability of a trait in a population of clones is zero; in a genetically diverse population it is much higher. This variation doesn’t represent error around some true value, because there is no principled way to establish what the “true” values of genetic and environmental variability are. These problems are especially acute in human beings. In plants and animals, the scientist has experimental control over the variances, so heritability can be estimated with respect to some fixed set of variances and covariances. Human variation is uncontrolled and uncontrollable.

You say in the response to the LLM that you don’t mention epistasis, but I do think it is important here. Human heritability is estimated by relating genetic and phenotypic similarity— on the one end comparing MZ and DZ twins, on the other comparing the SNP-similarity of more or less “unrelated” people. One reason these two estimates don’t reach the same conclusion is that MZ twins instantiate genetic similarity in ways that no other pairs of people reproduce. This is not to say that twin heritability is the correct one, it just suggests that genetics makes twins similar in ways that they don’t make most people similar. Most people, of course, don’t have twins to predict from. This paper by Eftedal et al is interesting in that regard: pnas.org/doi/abs/10.107….