Sasha, thank you for this excellent post. It's just the right level of technical detail, IMHO.
I have a request for you. Would you please consider writing a piece about genetic risk and heritability regarding dichotomous developmental disorders for which there are no frequently encountered highly penetrant variants? There are many interesting and serious aspects of this that are rarely mentioned in public discourse, and accordingly a great deal of misunderstanding.
I would be happy to collaborate with you on this. Please DM me if interested.
The “genetics is meaningless at the individual level” argument has always struck me as a bit of an ivory-tower oversimplification. In practice, people rely on heritable traits all the time — in mate selection, in animal breeding, in risk assessment — because those traits do tell us something useful about individuals, even if probabilities are involved. This is like saying blood pressure or cholesterol scores are meaningless because they don’t predict heart attacks with 100% certainty. The claim sounds profound: polygenic scores only explain part of the variance in a trait, so they’re “useless” for predicting outcomes in a single person. In reality, this is a sleight of hand.
No one is promising perfect prediction. The relevant question isn’t “Can we see your entire future in your DNA?” but “Does knowing this genetic information shift the odds enough to matter in a decision?” If the answer is yes — and in embryo selection, animal breeding, and even mate choice it often is — then the tool has value. We make high-stakes decisions on weaker predictors every day: blood pressure readings, cholesterol scores, SAT results, investment risk models. None of these explain 100% of the outcome, yet they’re useful because they shift probability in our favour. The “individual level” dismissal is annoying because it’s:
1. Patronising — It assumes the public can’t understand probability or risk reduction, so better to tell them it’s “meaningless” than explain nuance.
(al
2. Specious — It pretends that imperfect prediction equals uselessness, which if applied consistently would abolish most of medicine, agriculture, and insurance.
3. Anti–common sense — People intuitively grasp that if you can tilt the odds toward a desired outcome, that matters. To insist otherwise sounds like saying dog breeding doesn’t work because not every puppy meets the standard.
Scott Alexander’s argument is about expected value, not fortune telling. Your article - reframing it as failed prophecy - dodges the actual decision-making context. It feels like a game of academic gotcha: technically correct in a narrow way, but misleading and condescending in the way it’s presented. Also the probability that a highly intelligent scientist like Scott Alexander may not understand probability ratios (despite his whole article referencing them throughout) is pretty much zero, making the probability that the premise of your article is correct also pretty much zero. Dismissing the whole enterprise as irrelevant just alienates people who can plainly see patterns in their own families and lives. I grew up with my dad (also a geneticist) doing this, the question I couldn't ask him but would love to ask you is why is it that geneticists prefer to patronise everyone else (and make their own field less interesting) than admit that given say the choice between a 10% probability and a 2% probability of a bad outcome, all else being equal, taking the 10% is the better option and doesn't make the person stupid). Thanks for making me think anyway
Honestly, it sounds like you’ve missed the point of the article. Your rebuttal boils down to:
1. The quoted percentages of risk reduction are predictive, though not necessarily perfect.
2. Because they are predictive, they are useful.
3. Because they are useful, the author is wrong.
However, you’ve got it wrong at 1. The point of the article is that the percentages being quoted do not mean what you think they mean, and that you should not reductively look at them as predictors.
Apologies in advance if this has misrepresented you.
I don’t think you’ve understood the post, it is not saying embryo selection is bad because you can’t guarantee a reduction in risk.
In simple terms the author is pointing out that because diseases are often categorised in discrete terms this means small reductions in disease liability can appear like large reductions in risk of disease contraction. If you then use these inflated risk reductions to do expected value then your expected value will also be inflated.
Also could I ask if you’ve used AI to heavily write or edit this comment? It reads like the AI hasnt understood the full context.
Hi thanks and very interesting response. I do notice that science-coded communicators often resist shifting the conversation towards interpretation rather than pure fact-recitation, I suspect because detachment is coded as objectivity. I have often noted that being capable of understanding the technical side, but choosing to foreground meaning and narrative seems to feel threatening, possibly because it breaks the unspoken rules of the in-group — in this world, high-IQ people prove it by speaking their dialect, not by translating it for others.
So yes, I understand the author’s technical point: when conditions are defined by a diagnostic threshold, a small reduction in underlying liability can look like a large drop in “risk” if you only look at binary outcomes, which can inflate expected value calculations. This is statistically correct in a population-genetics sense.
But there’s also a difference in language code at play here. In statistical–technical discourse, meaning is conveyed in terms of risk distributions, thresholds, and probabilities. In lived-experience–driven discourse (which is how most patients non-specialists (and the best science communicators think), meaning is conveyed in terms of how much better or worse life would feel if symptoms were milder, delayed, or less frequent. These are different registers — and each group often perceives the other as “missing the point”: the statistical code can sound cold or patronising, while the meaning-driven code can sound naïve or imprecise.
From a meaning-driven perspective, even a small shift along the liability spectrum is worth something if it delays onset, reduces severity, or changes life trajectory. Prospective parents might rationally decide to pay for a tool that increases the odds of that shift, even knowing the limitations in an individual case.
Also, there’s a rhetorical layer here: the post reads as “only someone who doesn’t understand risk stratification would make this choice.” That frames disagreement as evidence of ignorance. Your response uses the same frame.
One question for you. If you wanted to argue instead that “a fully informed person could rationally make this choice under certain conditions,” what would that look like in your language code? Or do you believe there is no scenario in which an informed, non-naïve person would ever choose embryo selection for disease risk reduction?
Bruh. The point of the article is that the "reductions" possible with polygenic selection are months of delays in onset or slight reductions in severity that may lead to worse outcomes because of clinician inattention to patients at lower risk. The problem is that the selection is being presented at a CURATIVE. This isn't actually difficult to understand.
RE: "Also, there’s a rhetorical layer here: the post reads as “only someone who doesn’t understand risk stratification would make this choice.” That frames disagreement as evidence of ignorance. Your response uses the same frame."
The post doesn't suggest this in the least. This is what the post says: "I’m emphasizing all of this because the intuition that embryo selection is a (probabilistic) cure has already taken hold." The correct reading (read: direct reading) is that the author is positing a widespread fundamental misunderstanding of risk stratification, and that the companies offering polygenic embryo selection are exploitatively perpetuating this misunderstanding.
RE: "One question for you. If you wanted to argue instead that “a fully informed person could rationally make this choice under certain conditions,” what would that look like in your language code? Or do you believe there is no scenario in which an informed, non-naïve person would ever choose embryo selection for disease risk reduction?"
This is orthogonal to the point of the article. It's almost as if you think the post is saying "polygenic embryo selection is quackery and you shouldn't opt for it". I have news for you: it is not saying that.
Intentional reference to the great Raymond Carver? Embryo selection is indeed a perverse love...great post.
This is an incredible post. Very well written!
Thanks, I've really enjoyed your neuroscience posting as well!
Great post
Sasha, thank you for this excellent post. It's just the right level of technical detail, IMHO.
I have a request for you. Would you please consider writing a piece about genetic risk and heritability regarding dichotomous developmental disorders for which there are no frequently encountered highly penetrant variants? There are many interesting and serious aspects of this that are rarely mentioned in public discourse, and accordingly a great deal of misunderstanding.
I would be happy to collaborate with you on this. Please DM me if interested.
Must read
The “genetics is meaningless at the individual level” argument has always struck me as a bit of an ivory-tower oversimplification. In practice, people rely on heritable traits all the time — in mate selection, in animal breeding, in risk assessment — because those traits do tell us something useful about individuals, even if probabilities are involved. This is like saying blood pressure or cholesterol scores are meaningless because they don’t predict heart attacks with 100% certainty. The claim sounds profound: polygenic scores only explain part of the variance in a trait, so they’re “useless” for predicting outcomes in a single person. In reality, this is a sleight of hand.
No one is promising perfect prediction. The relevant question isn’t “Can we see your entire future in your DNA?” but “Does knowing this genetic information shift the odds enough to matter in a decision?” If the answer is yes — and in embryo selection, animal breeding, and even mate choice it often is — then the tool has value. We make high-stakes decisions on weaker predictors every day: blood pressure readings, cholesterol scores, SAT results, investment risk models. None of these explain 100% of the outcome, yet they’re useful because they shift probability in our favour. The “individual level” dismissal is annoying because it’s:
1. Patronising — It assumes the public can’t understand probability or risk reduction, so better to tell them it’s “meaningless” than explain nuance.
(al
2. Specious — It pretends that imperfect prediction equals uselessness, which if applied consistently would abolish most of medicine, agriculture, and insurance.
3. Anti–common sense — People intuitively grasp that if you can tilt the odds toward a desired outcome, that matters. To insist otherwise sounds like saying dog breeding doesn’t work because not every puppy meets the standard.
Scott Alexander’s argument is about expected value, not fortune telling. Your article - reframing it as failed prophecy - dodges the actual decision-making context. It feels like a game of academic gotcha: technically correct in a narrow way, but misleading and condescending in the way it’s presented. Also the probability that a highly intelligent scientist like Scott Alexander may not understand probability ratios (despite his whole article referencing them throughout) is pretty much zero, making the probability that the premise of your article is correct also pretty much zero. Dismissing the whole enterprise as irrelevant just alienates people who can plainly see patterns in their own families and lives. I grew up with my dad (also a geneticist) doing this, the question I couldn't ask him but would love to ask you is why is it that geneticists prefer to patronise everyone else (and make their own field less interesting) than admit that given say the choice between a 10% probability and a 2% probability of a bad outcome, all else being equal, taking the 10% is the better option and doesn't make the person stupid). Thanks for making me think anyway
Honestly, it sounds like you’ve missed the point of the article. Your rebuttal boils down to:
1. The quoted percentages of risk reduction are predictive, though not necessarily perfect.
2. Because they are predictive, they are useful.
3. Because they are useful, the author is wrong.
However, you’ve got it wrong at 1. The point of the article is that the percentages being quoted do not mean what you think they mean, and that you should not reductively look at them as predictors.
Apologies in advance if this has misrepresented you.
I don’t think you’ve understood the post, it is not saying embryo selection is bad because you can’t guarantee a reduction in risk.
In simple terms the author is pointing out that because diseases are often categorised in discrete terms this means small reductions in disease liability can appear like large reductions in risk of disease contraction. If you then use these inflated risk reductions to do expected value then your expected value will also be inflated.
Also could I ask if you’ve used AI to heavily write or edit this comment? It reads like the AI hasnt understood the full context.
Hi thanks and very interesting response. I do notice that science-coded communicators often resist shifting the conversation towards interpretation rather than pure fact-recitation, I suspect because detachment is coded as objectivity. I have often noted that being capable of understanding the technical side, but choosing to foreground meaning and narrative seems to feel threatening, possibly because it breaks the unspoken rules of the in-group — in this world, high-IQ people prove it by speaking their dialect, not by translating it for others.
So yes, I understand the author’s technical point: when conditions are defined by a diagnostic threshold, a small reduction in underlying liability can look like a large drop in “risk” if you only look at binary outcomes, which can inflate expected value calculations. This is statistically correct in a population-genetics sense.
But there’s also a difference in language code at play here. In statistical–technical discourse, meaning is conveyed in terms of risk distributions, thresholds, and probabilities. In lived-experience–driven discourse (which is how most patients non-specialists (and the best science communicators think), meaning is conveyed in terms of how much better or worse life would feel if symptoms were milder, delayed, or less frequent. These are different registers — and each group often perceives the other as “missing the point”: the statistical code can sound cold or patronising, while the meaning-driven code can sound naïve or imprecise.
From a meaning-driven perspective, even a small shift along the liability spectrum is worth something if it delays onset, reduces severity, or changes life trajectory. Prospective parents might rationally decide to pay for a tool that increases the odds of that shift, even knowing the limitations in an individual case.
Also, there’s a rhetorical layer here: the post reads as “only someone who doesn’t understand risk stratification would make this choice.” That frames disagreement as evidence of ignorance. Your response uses the same frame.
One question for you. If you wanted to argue instead that “a fully informed person could rationally make this choice under certain conditions,” what would that look like in your language code? Or do you believe there is no scenario in which an informed, non-naïve person would ever choose embryo selection for disease risk reduction?
Bruh. The point of the article is that the "reductions" possible with polygenic selection are months of delays in onset or slight reductions in severity that may lead to worse outcomes because of clinician inattention to patients at lower risk. The problem is that the selection is being presented at a CURATIVE. This isn't actually difficult to understand.
RE: "Also, there’s a rhetorical layer here: the post reads as “only someone who doesn’t understand risk stratification would make this choice.” That frames disagreement as evidence of ignorance. Your response uses the same frame."
The post doesn't suggest this in the least. This is what the post says: "I’m emphasizing all of this because the intuition that embryo selection is a (probabilistic) cure has already taken hold." The correct reading (read: direct reading) is that the author is positing a widespread fundamental misunderstanding of risk stratification, and that the companies offering polygenic embryo selection are exploitatively perpetuating this misunderstanding.
RE: "One question for you. If you wanted to argue instead that “a fully informed person could rationally make this choice under certain conditions,” what would that look like in your language code? Or do you believe there is no scenario in which an informed, non-naïve person would ever choose embryo selection for disease risk reduction?"
This is orthogonal to the point of the article. It's almost as if you think the post is saying "polygenic embryo selection is quackery and you shouldn't opt for it". I have news for you: it is not saying that.
I think we got ourselves a new copypasta here folks.