This essay asks a single question: when a moral or policy argument sounds universal, who actually gets to fill in the content of "good" — and does the process leave room for the people affected to push back? Eugenics turns out to be one of the clearest historical cases where that question was never asked. This piece works through why.
David Friedman wrote a piece this past June (2026) called "The Right Kind of Eugenics." His argument is simple and has real force: eugenics gets demonized because people equate it with state-mandated breeding programs. But if parents choose what genes they want their own children to have, what exactly is wrong with that? Central planning is worse than market choice — that's a lesson the twentieth century taught us repeatedly. By the same logic, letting parents decide for themselves is the "right kind" of eugenics.
There's an internal logic to this argument, and I'm not going to say Friedman is simply wrong. But there's a question he never answers, and I want to name it.
Start with the least controversial ground. No parent wants their child born with a preventable genetic disease. Prenatal screening, genetic counseling, and preventive interventions such as folic acid supplementation are all, in some sense, interventions on reproductive outcomes — and none of us think they raise a moral problem. Wanting a "good birth" is a real, understandable human desire.
You can't build a critique of eugenics on the premise that the goal itself is wrong. The moment you try, someone will point out: are you against genetic disease screening? Are you against parents giving their kids a better start? That's a strong rebuttal, and there's no ready answer to it.
One distinction needs to be made explicit here. When a doctor recommends embryo testing for a couple who both carry the Huntington's gene, or a sperm bank screens out donors carrying serious genetic diseases, the intervention is directly aimed at avoiding serious disease in the prospective child, "bad" is anchored to observable suffering from disease, and there is reasonably stable medical consensus on what counts. This essay isn't about that.
What this essay is about is a different kind of intervention: when "a good birth" expands from preventing specific diseases into optimizing IQ, height, "cognitive potential," or other composite traits. At that point, "good" is no longer anchored to observable suffering — it's filled in by somebody's judgment. Whose judgment? By what standard? That's the question this essay is actually asking.
So the critique can't start with "the goal is wrong." But it has to start somewhere.
"A good birth" is an abstraction. On paper it contains no specific content. The problem is: the moment you try to turn it into policy — whether state policy or a market product — you have to answer one question:
Good, according to what?
There's no fully neutral, context-free answer to that question.
Francis Galton coined the word "eugenics" in 1883. What he defined as "good" was what his own Victorian English upper class considered important: intelligence, physique, so-called "genius." He spent years tabulating the descendants of elite families and concluded that excellence was heritable. He believed he was doing science. He believed his goal was to benefit humanity as a whole.
Forty years later, the American eugenics movement operationalized the same "improvement of the race" goal into forced sterilization of the poor, of immigrants, of the disabled. Virginia passed its sterilization law in 1924; in 1927 the U.S. Supreme Court ruled 8–1 in Buck v. Bell that forced sterilization was constitutional. Oliver Wendell Holmes — at the time considered one of the finest legal minds in America — wrote the line that would become famous: "Three generations of imbeciles are enough." More than 60,000 Americans were sterilized afterward, many without full informed consent.
Less than twenty years later, Nazi Germany carried out mass genocide. In the Nuremberg proceedings, Nazi defense counsel cited Buck v. Bell as a precedent in their defense.
Now David Friedman offers a fourth version: "good" is defined autonomously by parents, mediated through a market for reproductive technology.
Notice that these four versions share the same argumentative structure. Each claims to be advancing "a better birth." Each believes itself to represent reason and progress. Each has some internal logic that makes sense on its own terms. What differs is only this: whose preferences get to count as the definition of "good."
Nobody here is a villain, and nobody here corrupted a once-pure idea. Galton wasn't hijacked. He defined "excellence" using the standards of his own Victorian elite class — he just never noticed that's what he was doing. American eugenicists didn't misuse Galton; they applied the same logic of definitional authority to a different political situation. The Nazis didn't misuse the American model; they cited it, literally, as legal precedent.
Every time, the word "good" gets filled in by whoever happens to occupy a specific position of power. And every time, that act of filling-in gets packaged as neutral — scientific, medical, technical, or market-based — rather than what it actually is: an exercise of power.
There's a mechanism buried in this pattern that's worth naming directly: the reason this filling-in process is so hard to see isn't that the people doing it are lying. It's that they aren't.
Friedman's argument contains a hidden assumption: if there's no central definer — if every parent is choosing autonomously — then the question of "who defines good" gets dissolved into a million individual preferences, and the problem disappears along with the central authority.
That's a clever rebuttal. But it skips two steps.
Step one: individual preferences don't form in a vacuum. What parents judge to be "good genes" comes from the culture they're embedded in, market incentives, insurance pricing, medical advice, and which traits make life easier in the society they live in. There's a critical split hiding here: parents have the power to choose, but the experts, institutions, and regulatory frameworks that design the menu of options hold something more substantial — the power to define what counts as excellent in the first place.
Parents can order off the menu. But who decides what's on it? Who decides which traits get listed as "optional enhancements" and which get listed as "risks to avoid"? These are not neutral technical decisions, even though they're routinely presented in technical language. A voluntary market doesn't eliminate this problem — it just relocates the power to a place that's harder to see.
Step two, and the deeper one: Galton's sincerity isn't a defense of his project. It's the precondition that made the whole mechanism work at all.
If Galton had known he was dressing up his own class's aesthetic preferences as universal scientific standards, that would be hypocrisy. Hypocrites are self-aware; some even carry a certain cynical distance from what they're doing. But Galton had none of that awareness. He genuinely believed he was doing science. He genuinely believed "genius" was objectively measurable. He genuinely believed spreading these traits would benefit all of humanity. It's precisely this sincerity that made the translation of class interest into neutral scientific language nearly invisible — invisible to outsiders, and invisible to Galton himself.
This is a translation process that runs at the level of the unconscious: the traits held by people in one specific social position get mapped onto the goal of "human improvement." The translation completes itself, but the act of translating disappears, leaving behind only what looks like a universal scientific conclusion.
This is also why Holmes's line deserves a closer look. "Three generations of imbeciles are enough" — he meant it. He believed the sentence was protecting society. And it's precisely because he meant it that the judgment could appear, with such confidence, in the formal ruling of the highest court in the United States, carrying the full weight of judicial authority. A judge who knew he was lying wouldn't have written that sentence.
This leap from personal sincerity to institutional violence didn't happen through some conspiracy of evil actors. It happened through the mediation of professionalism. Galton's personal intuition, translated into medical standards and dressed in the language of actuarial science, became scientific "objective fact." That fact then entered the legal pipeline, where Holmes converted it into a ruling with the force of law. At every stage of this process, the traces of power get filtered out, until all that remains in the legal text is a cold, neutral, unquestionable technical authority.
Sincerity doesn't grant immunity. Not because we get to retroactively condemn these people, but because this is the mechanism's most dangerous feature: it doesn't require malice to run. Anyone occupying a position powerful enough to define "good," without noticing that's what they're doing, is repeating this same process. That includes us, right now.
This argumentative structure never actually went away. It was just waiting for new technological clothing.
In late 2024, a startup called Heliospect Genomics was the subject of an undercover investigation by Hope Not Hate, followed up by reporting in The Guardian. The company offers a service to wealthy couples undergoing IVF: an algorithm that scores and ranks embryos by predicted IQ and other traits, helping clients select the "optimal" embryo for implantation. The underlying training data comes from the UK Biobank — a publicly supported genetic research resource containing data from half a million volunteers. But the weighting model that defines "superior genes" was set entirely by this private company, operating in a regulatory gap. This is menu design in its purest form: publicly supported data turned into a private commercial product, offered as an exclusive selection tool for the elite.
This is a contemporary variant of eugenic selection. The angle here is different from the standard bioethics "expressivist objection" — see Asch & Parens, 2000 — about the cultural message screening sends to people currently living with disabilities. What's at issue instead is how definitional power itself gets captured by a specific position of power, even when every participant is acting voluntarily. It no longer relies on state coercion. It relies on purchasing power. As this kind of selection normalizes in the private market, society gradually absorbs a belief: that inequality no longer originates in social structure, but in biological superiority and inferiority.
Someone will object: didn't the UK already ban IQ-based embryo selection? Isn't the problem solved?
In September 2025, the UK's Human Fertilisation and Embryology Authority formally confirmed that PGT-P is not lawful in Britain. In January 2026, the American Society for Reproductive Medicine concluded that PGT-P is a nascent, unproven technology that should not be used clinically at this time. These are not empty statements — they were made in earnest. But Heliospect is registered in Wyoming and has operated in the United States, where commercial embryo-screening services of this kind are offered. The ban exists. Definitional power hasn't disappeared. It has simply relocated to whichever jurisdiction offers the least regulatory friction.
Regulation itself has become a battlefield in the contest over definitional power. A national ban may constrain institutions and clinics operating within that jurisdiction, but it does not necessarily prevent the underlying service from being offered elsewhere. An algorithm can be trained anywhere it's permitted to operate; a service can be sold anywhere it's permitted to be sold. Definitional power doesn't vanish because of a ban — it reconfigures itself along the path of least regulatory resistance.
Technocratic "optimization" decisions require someone to first decide what counts as a good outcome in the first place — a decision that is not technical but normative, even though it's routinely presented as if it were technical. Current debates in AI alignment face a structurally identical question: whose preferences should count as the relevant expression of "human values," and through what legitimate process?
The moral magnitudes here differ enormously, and nothing above equates these projects with Nazi eugenics. What they share is the same argumentative structure: a goal that sounds universal (better population quality, optimal social decisions, the realization of human values) obscures the fact that "who gets to define this goal" is a question of power.
Embryo selection may only be the beginning. As the starting point of biological life, the process of living itself, and even future generations become objects that can be designed, purchased, and optimized, the question of who holds definitional power will only grow more important. A companion essay looks at why a very different biological technology — organoid computing — sails through legal screening that treats embryo selection and germline editing as close to untouchable, for reasons that turn out to have little to do with whether anyone has actually resolved what these systems experience.
Friedman says letting parents choose for themselves is the "right kind" of eugenics, because it avoids state coercion.
My response isn't: you're wrong, voluntary eugenics is unacceptable too. My response is: you still haven't answered the question.
When parents select their children's traits in a market, who designed the menu? Who decided which traits get listed as "optimizable" and which get listed as "avoidable risks"? Who makes these decisions, by what standard, under what set of incentives — the voluntary/coercive distinction, on its own, cannot answer any of this.
Galton thought he was doing science. American judges thought they were protecting society. The Nazis thought they were saving civilization. Friedman believes he's defending individual liberty.
These positions aren't equivalent, and everyone here being sincere doesn't make right and wrong undecidable. What's on offer instead is a different diagnostic tool.
Call it the Position Invariance Test: take an "improvement" proposal's argumentative structure and flip it entirely. If the group currently on the receiving end of the definition — those judged "unfit to reproduce," or those who will be the ones selected out under a given scheme — were the ones designing this proposal, would they arrive at the same conclusion? If not, the proposal hasn't passed the test.
This test doesn't presuppose that "postnatal intervention is fine, prenatal selection isn't." That distinction would drag us back into essentialism, which doesn't hold up here. Cosmetic surgery and embryo selection are both active interventions; neither is more "natural" than the other.
Nor can the line be drawn as simply "whoever decides bears the responsibility." Parents already make countless decisions on their children's behalf — what school to attend, what to eat, what medical care to receive. If that were the standard, embryo selection wouldn't be special at all.
The real dividing line sits somewhere else: whether the person bearing the consequences will ever have a genuine chance, in the future, to revise, reject, or exit that definition.
Cosmetic surgery can be redone in adulthood; aesthetic preferences can change; the original decision can be rejected outright. Even in cases where the decision was originally made by someone else on your behalf — a parent arranging medical correction, say — as long as you can overturn or reassess it once you're an adult, that exit channel stays open. Embryo selection has no such channel, at least not with current technology. The decision is completed, irreversibly, before the person who will bear its consequences even exists. With the tools that exist today, this goes beyond merely "hard to reject" — there is no possibility of rejection at all. Should reversible or correctable germline intervention ever become real, this specific point would need revisiting, and the structural argument would simply move to whatever, if anything, remained irreversible. None of this indicts parental motives. It's a structural fact, for now: there is no exit here.
This also lets the test answer a harder question. Suppose someone was excluded, as a child, for not matching whatever their society currently treats as more desirable — appearance, height, or whatever composite of traits the local standard happens to reward. As a parent, they now have the option to select embryos toward that same standard for their own child. The choice is understandable, even protective: no parent wants their child to live through what they lived through.
But notice what hasn't changed. The parent escaped exclusion personally. But the standard that produced that exclusion remains intact. It is now being passed forward to the next generation through the very choices made to protect the child from the same harm. A trait that could once have been argued over, contested, or simply outgrown by a later generation is instead selected before the person who will carry it exists. Cultural standards can be challenged by the people living under them; a child can grow up and reject what their parents accepted. But if enough parents make the same selection, for the same reasons, the standard stops depending on persuasion and imitation to reproduce itself. It gets a biological delivery mechanism.
Selection doesn't inevitably produce uniformity — that would be a stronger empirical claim than anything argued here. The narrower point is that selection can convert a socially contestable preference into an intergenerational filter, and the more widely a filter like that gets adopted, the less room future generations may have to challenge that standard through lived experience.
That's the sharper version of the Position Invariance Test: if the people once excluded by a standard become its most committed reproducers, have they escaped the standard — or made it a little harder for the next generation to escape it?
Readers here won't stop at Friedman's libertarian defense, either. A stronger position already circulates in this same discourse — not confined to any one platform, and not something to attribute to the community as a whole, but visible enough to require an answer: not that parents may select for higher predicted cognitive ability, but that they arguably ought to. On this view, declining to select is itself a parental failure, not a neutral choice. The same reasoning sometimes extends toward future germline editing rather than embryo selection — a different technology with its own separate risk profile that this essay doesn't take up here, though the definitional-authority problem below would transfer to it largely unchanged. In at least one visible instance, skepticism about the underlying science gets treated less as a disagreement to engage than a position to diagnose — labeled "denialism." That framing performs a move related to the one Holmes's opinion performed, at a far smaller institutional scale and with none of its coercive force: it treats a contested definition of "better" as settled enough that disagreement no longer needs answering, only diagnosing. For the established philosophical form of this stronger claim, see Julian Savulescu's "procreative beneficence" argument and its critics — Robert Sparrow, Michael Parker — which this essay treats as the strongest existing counterposition to what follows, not as a target in its own right.
This obligation-language is a stronger claim than Friedman's, and it deserves a stronger answer than "that's a different topic." Turning selection from permitted into obligatory doesn't dissolve the question this essay keeps returning to — it raises the stakes on it. The Position Invariance Test was never a test of prior consent; no formulation of it depends on a future child being consulted before existing, since that's metaphysically impossible and no version of this argument requires it. What it actually tests is narrower: first, who had the authority to fix the standard of "better" now being enforced; second, whether the person who will live inside that standard retains any later, practical channel to revise or reject it, once the decision that applied it is irreversible. An obligation doesn't open that channel. It removes one of the few soft checks a merely-permitted market still allowed — a parent's individual reluctance to select — and recasts that reluctance itself as the thing in need of correction. That's not a smaller version of the menu-design problem. It's a more consequential one: the definition is no longer something anyone, parent or child, retains a live option to opt out of.
Nor is "higher cognitive ability" as self-defining as it first sounds. It isn't one single, uniform trait — researchers distinguish general ability from a range of broader and narrower abilities that don't always move together. Even a technology that could reliably predict and select among them would still leave someone to decide which of those abilities count as improvements, how competing ones should be weighted against each other, and why that weighting should govern the child who receives it. Being measurable doesn't settle any of that. The measurement problem isn't the whole normative problem — but it's one more place where "we can quantify it" quietly stands in for "we know what's better," and those aren't the same claim. A much more mundane version of the same risk shows up in a college major or a technical skill that looks like the safest bet in the room, only to lose most of its market value within a decade once the conditions that made it valuable shift. Betting a person's genome on today's judgment of what's worth optimizing for carries the same risk, at a scale nobody gets to walk back.
There's a second, more physical version of the same gap. Even when a target trait is measurable, the genome doesn't work like a panel of independent switches. Traits are often wired together — a change that nudges one predicted outcome up can carry effects on other traits or risks that nobody was trying to touch. So there are really two separate unresolved questions sitting on top of each other here, not one. Who gets to decide what counts as better is the first — call it the authority problem. Whether anyone selecting for a target trait can actually get only that trait, without quietly moving others nobody chose, is the second, and it's a biological one, not a normative one. Solving the first doesn't solve the second, and a confident answer to the second still leaves the first completely open.
The Position Invariance Test isn't opposed to improvement as such — it's a legitimacy test. It doesn't presuppose the answer; it requires that any proposal claiming to be "universally beneficial" submit to the invariance check — who had the power to fix that standard, and were they ever answerable to the people who'd have to live under it? Once that decision is made and can't be undone, does the person bearing the consequences get any real, later chance to revise it, reject it, or walk away from it? These aren't the only possible criteria — they're this essay's specific operationalization of a single underlying test. Any proposal that can't answer them hasn't passed. Passing it is a floor, not a ceiling: it rules out one specific failure — a standard imposed without accountability to the people who'll live under it — without certifying that the standard itself is otherwise sound. A standard someone applies to their own child and would have accepted for themselves can still be a bad standard; invariance checks how a standard gets enforced, not whether it deserves to be held at all.
None of the places this essay has pushed back — the menu-design point, the obligation point, the measurement point, the shared-wiring point — is an argument against selection itself. Each names a condition, not a verdict. A selection process built on a standard that the people who'd have to live under it could contest and, in principle, revise; applied to a target that measurement could actually isolate, at a resolution where "better" meant something more specific than a proxy; without smuggling in unintended effects on traits nobody chose to touch — would not be an example this essay is warning about. It would be the thing this essay has been describing the absence of.
This is the real lesson eugenics leaves us with — not "don't do bad things," but "how do you recognize a distortion of definitional power operating inside good intentions." And the next time you hear some scheme claim it's advancing "the improvement of humanity" — whether that's optimizing genes, optimizing decisions, or aligning values — ask the question that never got answered:
Who designed the menu?
This essay is based on a working paper archived on Zenodo (DOI: 10.5281/zenodo.21429939). AI assistance was used in formalizing the arguments, verifying citations, and adapting the working paper into this essay; all analysis and conclusions are the author's own.
