Astral Codex Ten

Jonathan Anomaly

7 min read

Jonathan Anomaly

Article

Jonathan Anomaly is a recurring person in the Astral Codex Ten archive, appearing 3 times across 3 issues between February 03, 2022 and July 31, 2025. The archive places it in contexts such as “[Email jonathan.anomaly@protonmail.com]”; “Jonathan Anomaly on genetic enhancement via polygenic embryo selection”; “a professor named Jonathan Anomaly got cancelled from Penn”. It most often appears alongside Aella, Israel, Ivy League.

Metadata

  • Category: People
  • Mention count: 3
  • Issue count: 3
  • First seen: February 03, 2022
  • Last seen: July 31, 2025

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

ACX Grants ++: The First Half
February 03, 2022 · Original source
#14: Survey On Embryo Selection I would like to design and conduct a survey akin to the moral machine project, but for embryo selection rather than self-driving cars. The idea is to glean the informed preferences of parents over the kinds of traits they want their children to have. This is already a challenge, and the survey would need to be carefully worded to avoid framing effects, and other psychological biases. But the bigger challenge, and my interest in the project, is that parents' preferences will largely depend on the preferences over traits that other parents have. As I wrote about in my book Creating Future People, there is a range of traits in which what is individually rational is partly a function of what other parents are expected to choose, and in which what is individually rational could diverge from what is socially optimal. Survey answers could, in principle, be fed into an AI to help refine the preferences parents exhibit over which embryos to implant. This information would be enormously valuable in for parents and fertility clinics. To the extent that the traits of future people influence the welfare of the entire world, embryo selection done well – guided by accurate information rather than guesswork – could be one of the most important forms of “effective altruism” the world has ever seen. [Email jonathan.anomaly@protonmail.com]
23
October 31, 2023 · Original source
Jonathan Anomaly on genetic enhancement via polygenic embryo selection (the prediction market connection is strained, but it’s always nice to hear a PGS talk)
Inline links: genetic enhancement via polygenic embryo selection
Suddenly, Trait-Based Embryo Selection
July 31, 2025 · Original source
…and the section of the post called “Antagonistic Pleiotropy” for more. However, this paradoxical benefit is only true for a few conditions like diabetes - for everything else, selecting on health index does better than you would naively think, but still does not decrease the risk of a given condition as much as selecting against that condition directly. 3That is, new mutations in that particular baby, as opposed to older mutations already present in the parents. 4Conflicts of interest: I have used Orchid’s and Herasight’s products on my own embryos (not the ones used to conceive my existing kids, but for a potential third child), employees of Genomic Prediction and Herasight have been extremely helpful in contributing expertise to ACX posts on genetics, and I might invest in this field at some point (though haven’t done so yet). This post started as Herasight asking me to write about their white paper, then spiraled out of control. There were some unexpected time pressures and the result is that I didn’t get a chance to run everything in Herasight’s white paper by their competitors as thoroughly as I would like. Although I talked to representatives of all four companies profiled here, I feel like this probably reflects Herasight’s perspective better than other companies’, and that this is a major flaw. If other companies have responses, I’ll publish them. Thanks to all companies involved for their assistance on this article. Finally, I am favorably disposed toward Herasight because of how I learned about them: a professor named Jonathan Anomaly got cancelled from Penn for being too gung-ho about genetic enhancement, and used his newfound freedom to join a very-early-stage Herasight, raise their ambitions, and sell everyone (including me) on the idea. I grew up on a diet of books and movies about mad scientists, and I’m a sucker for a story about a guy named Doctor Anomaly pursuing revenge against the small-minded fools who destroyed his career by creating a race of superbabies. 5The version of the tool I looked at said 5.9 points for five embryos, up to 9 points for twenty embryos. The version of the tool on their current said says 5.3 - 9, so they might have recalculated after I finalized this article. 6Used in quotation marks because these scores were fine for the predictive tasks they were applied for - they just weren’t finding genes that directly caused the outcome of interest. 7Conflict of interest notice: this table was originally unadjusted. A representative of Herasight claimed that this was unfair, because each company used slightly different reporting conventions, and offered to correct for this in a neutral way. I retraced their reasoning, confirmed that the correction did not especially benefit Herasight at the expense of other companies, and accepted the correction. The original unadjusted table is below: Herasight was insufficiently comfortable with Nucleus’ methodology to even be willing to posit a corrected value, so I left their self-reported value in gray. 8Zagorsky (2007) says an extra IQ point means $234-$616/year in higher salary. The midpoint of $425 equals $670 in today’s dollars; assuming a forty-year career, Nucleus’ +1 point estimate is worth $26,800 (vs. $9,249 Nucleus cost) and Herasight’s +6 point estimate is worth $160,800 (vs. $53,250 Herasight cost). 9As part of researching this article, I asked all four major companies about their within-family validation strategies. Here are some details: Genomic Prediction discusses their strategy in this paper. The results are complicated to interpret - the within-family numbers often have such wide error bars that they overlap with both the across-family numbers and with zero - but looking qualitatively it seems like most scores on average lose about 25% of their risk reduction ability (though averages might not be the right way to do this, and some might be much more affected than others). Their website reports unadjusted, not within-family validated numbers; GP says they say this clearly on their site (which is true), Herasight counters that they still present their numbers as applicable to embryo selection (which is also true). To get the most applicable-to-embryo-selection numbers, you might want to adjust GP’s stated numbers down somewhat; it’s hard to say exactly how much, but maybe 20 - 25%?
Inline links: 3, 4, 5, 6, 7, https://substackcdn.com/image/fetch/$s_!Vimq!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca0f1f15-268d-465a-a70f-b7f1173c6111_566x166.png, https://substackcdn.com/image/fetch/$s_!3B0A!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Faccc7a65-b142-4bf6-927d-53eb607d71ef_552x155.png, 8, 9, in this paper