So I had read Kevin Mitchell’s Innate and then followed up by looking at some of the papers back in 2023 or so.
It seems that some of these things are turning out to be false/not replicating? E.g the mutational load hypothesis so I think it’s more about going to the extremes, it might be hard to get 200 IQ geniuses all over the place. I think increasing the general populations IQ by 30-50 points remains quite feasible and that it would be great boon.
My certainty here is not that high fwiw.
I did a chat with fable about this and here’s the sources it looked at:
Sources on the “mutational load / asymmetry” view of intelligence genetics (the claim in Kevin Mitchell’s *Innate*), plus the newer work that bounds it:
**Theory**
- [Keller & Miller 2006, *Behavioral and Brain Sciences*](https://pubmed.ncbi.nlm.nih.gov/17094843/) — Why harmful, heritable mental disorders persist: new mutations arrive faster than selection removes them (polygenic mutation-selection balance). The theoretical backbone of the load argument.
**Evidence that intelligence was selected upward and that random variation skews negative**
- [Joshi et al. 2015, *Nature*](https://pubmed.ncbi.nlm.nih.gov/26131930/) — 354k people: more homozygous DNA (children of related parents) → lower cognition. Directional dominance is the signature of a trait under past directional selection. - [Ganna et al. 2016, *Nature Neuroscience*](https://www.nature.com/articles/nn.4404) — In the general population, each ultra-rare disruptive mutation ≈ 3 months less education; ≈ 6.5 months if it hits a constrained, brain-expressed gene. - [Kingdom et al. 2024, *Nature Genetics*](https://www.nature.com/articles/s41588-024-01710-0) — UK Biobank: rare damaging variants in developmental-disorder genes additively lower cognitive and socioeconomic outcomes; partially offset by a higher EA polygenic score.
**Evidence that bounds the strong version (“most heritable variance is load”)**
- [Common and rare variant associations with cognitive performance across development, *Nature Human Behaviour*](https://www.nature.com/articles/s41562-026-02491-8) — Common variants explain ~20% of variance; all rare coding variants exome-wide only ~1%. - [Distinct genetic architecture in the tails of complex traits, *Nature* 2026](https://www.nature.com/articles/s41586-026-10516-5) — Middle of the distribution (~±2 SD) is common-variant, linear-PGS territory; the tails are dominated by rare large-effect variants. Reconciles the two views.
I think it’s more about going to the extremes, it might be hard to get 200 IQ geniuses all over the place.
I think of my target as being something like: make it be the case that, for parents who want to, they can safely give their kid an average expected IQ of 175 or so. My guess is that this is both feasible and safe (given a bunch of work), but I’m currently investigating how likely we should expect this to be feasible for different numbers (e.g. what’s the genetic architecture of intelligence at the high tail).
(The following is garbled / maybe wrong because it’s unfinished thinking.) It’s also plausible that IQ is just not a good description of things at the high tail. We just mean something like “can do thinky stuff at the level of 1-in-a-million geniuses” (which corresponds to a bit less than 5 SDs, i.e. a bit less than 175 IQ). Such people definitionally exist, so we don’t need to worry about “IQ”. One question is then something like “For 1-in-a-million geniuses, is their genius explained by non-direct-genetic effects to an unusually large extent compared to average?”. If not, then we’re implicitly saying that direct genetic effects aren’t hitting diminishing returns faster than non-direct-genetic effects. That in turn suggests we can push the direct genetic effects a bit further than that. If yes on the other hand, then we’re saying that genetic effects strongly co-saturate with each other right around the 2 or 3 SD mark, but not at all before (because we can observe them not doing that in the normal range), and somehow don’t cosaturate with environment / other non-genetic things. Which seems weird.
So I had read Kevin Mitchell’s Innate and then followed up by looking at some of the papers back in 2023 or so.
It seems that some of these things are turning out to be false/not replicating? E.g the mutational load hypothesis so I think it’s more about going to the extremes, it might be hard to get 200 IQ geniuses all over the place. I think increasing the general populations IQ by 30-50 points remains quite feasible and that it would be great boon.
My certainty here is not that high fwiw.
I did a chat with fable about this and here’s the sources it looked at:
Sources on the “mutational load / asymmetry” view of intelligence genetics (the claim in Kevin Mitchell’s *Innate*), plus the newer work that bounds it:
**Theory**
- [Keller & Miller 2006, *Behavioral and Brain Sciences*](https://pubmed.ncbi.nlm.nih.gov/17094843/) — Why harmful, heritable mental disorders persist: new mutations arrive faster than selection removes them (polygenic mutation-selection balance). The theoretical backbone of the load argument.
**Evidence that intelligence was selected upward and that random variation skews negative**
- [Joshi et al. 2015, *Nature*](https://pubmed.ncbi.nlm.nih.gov/26131930/) — 354k people: more homozygous DNA (children of related parents) → lower cognition. Directional dominance is the signature of a trait under past directional selection.
- [Ganna et al. 2016, *Nature Neuroscience*](https://www.nature.com/articles/nn.4404) — In the general population, each ultra-rare disruptive mutation ≈ 3 months less education; ≈ 6.5 months if it hits a constrained, brain-expressed gene.
- [Kingdom et al. 2024, *Nature Genetics*](https://www.nature.com/articles/s41588-024-01710-0) — UK Biobank: rare damaging variants in developmental-disorder genes additively lower cognitive and socioeconomic outcomes; partially offset by a higher EA polygenic score.
**Evidence that bounds the strong version (“most heritable variance is load”)**
- [Common and rare variant associations with cognitive performance across development, *Nature Human Behaviour*](https://www.nature.com/articles/s41562-026-02491-8) — Common variants explain ~20% of variance; all rare coding variants exome-wide only ~1%.
- [Distinct genetic architecture in the tails of complex traits, *Nature* 2026](https://www.nature.com/articles/s41586-026-10516-5) — Middle of the distribution (~±2 SD) is common-variant, linear-PGS territory; the tails are dominated by rare large-effect variants. Reconciles the two views.
**Background / provenance of the idea**
- [West Hunter, “The genetics of stupidity” (2012)](https://westhunt.wordpress.com/2012/11/08/the-genetics-of-stupidity/) — Cochran’s version of the same load argument.
- [Yeo et al. 2011](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3027642/) — Early small study (n≈200) linking rare CNV deletions to IQ; the effect size didn’t hold up in larger samples.
I think of my target as being something like: make it be the case that, for parents who want to, they can safely give their kid an average expected IQ of 175 or so. My guess is that this is both feasible and safe (given a bunch of work), but I’m currently investigating how likely we should expect this to be feasible for different numbers (e.g. what’s the genetic architecture of intelligence at the high tail).
(The following is garbled / maybe wrong because it’s unfinished thinking.) It’s also plausible that IQ is just not a good description of things at the high tail. We just mean something like “can do thinky stuff at the level of 1-in-a-million geniuses” (which corresponds to a bit less than 5 SDs, i.e. a bit less than 175 IQ). Such people definitionally exist, so we don’t need to worry about “IQ”. One question is then something like “For 1-in-a-million geniuses, is their genius explained by non-direct-genetic effects to an unusually large extent compared to average?”. If not, then we’re implicitly saying that direct genetic effects aren’t hitting diminishing returns faster than non-direct-genetic effects. That in turn suggests we can push the direct genetic effects a bit further than that. If yes on the other hand, then we’re saying that genetic effects strongly co-saturate with each other right around the 2 or 3 SD mark, but not at all before (because we can observe them not doing that in the normal range), and somehow don’t cosaturate with environment / other non-genetic things. Which seems weird.