Downside risks of genomic selection

There are downside risks to selecting genomes of future human children for traits like health, intelligence, lifespan, mental health, and so on. This essay maps out some of these, starting with these intuitions:

  • Unnaturalness.
  • Objectification.
  • Transgression.
  • Misalignment.

Intuitions like these are almost always pointing at real things, rooted in real concerns. Often we can't make all those things explicit. This essay tries to make somewhat explicit as many of them as possible. If you have more concerns about genomic selection not yet covered here, please comment or send me an email (my gmail is tsvibtcontact)! I've added a few points to this list since first posting it, so there are probably some still missing.

Please note that this list is trying to be a fairly complete list of downside risks, not a prioritized lists of costs and benefits. So some items on this list are unlikely, implausible, fixable, and/or not worse than the status quo; and overall the benefits, which are not listed, may outweigh the downside risks. (Some of the possible risks listed here are probably the opposite of the truth; for example, parents who want to spend the resources to use GS are probably especially caring, kind, invested, sane, safe, and competent, rather than the reverse.)

Table of Contents:

1. Terms

This essay uses the term "genomic selection" (GS), meaning, creating organisms (e.g., humans) that are normal except that they have genomes that have been artificially selected in some way. GS is distinct from "eugenics", which is aimed at influencing a population's gene pool and reproductive patterns; GS just means, technological methods to create individual organisms whose genomes have been selected for some properties. So GS excludes coercive eugenics such as sterilizing, deporting, or murdering people, or punishing people for reproducing or not reproducing (such as by banning abortion), as well as less transparently coercive actions aimed at influencing the gene pool from outside of the wishes of parents such as paying people to reproduce or not reproduce. GS is more narrow than "genetic engineering", which includes things like designing novel viruses or genetically modifying adult organisms.

2. Process note: making intuitions explicit

I'd appreciate help making more of these intuitions explicit, because explicit reasons give more adaptable advice: it's hard to engineer to avoid a gut feeling, and it's hard to compare a gut feeling to other costs and benefits to decide whether something is good to do. Engineering and comparing are much easier with a specific explicit worry.

Also, these intuitions can be slippery, where arguments are used as soldiers that can ignore all counterevidence by a defeat-in-detail strategy of rehearsing all arguments against each counterargument, rather than collecting all arguments and counterarguments in memory and understanding how the truth is consistent with all of them. For example, Bob might say that intelligence increasing isn't necessarily good because intelligence isn't everything, and Alice might point out that the Flynn effect is generally considered good which shows that we value intelligence, and Bob might respond that genome selection is different from whatever natural factors cause the Flynn effect. Alice might then make explicit that Bob is pointing at Goodharting, and address ways in which a genome selection strategy might be Goodharting; to which Bob might respond by dismissing those counterarguments as irrelevant because maybe intelligence itself isn't actually good to increase.

This pattern isn't always just a mistake of reasoning on Bob's part. Often it is cover for an ungrounded status quo bias, conflictual anti-reason, and other non-reasoning processes that don't help with anything that any of us actually care about. In that case, we'd want to get clear on the facts of the matter, so that we can help make them clear for whatever reasoning processes there are, and avoid being taken in by anti-reason processes.

On the other hand, a superficially similar pattern of bouncing around between different propositions shows up when someone is trying to express intuitions they haven't yet made clear to themselves, trying to get ahold of what's underlying them, while another person is arguing against partial or straw-man expressions of the intuitions. In that case, again, clarity helps things, but can be premature; clarifying things can be mistaken as adversarially pretending to be comprehensive, and can verbally overshadow inexplicit but real knowledge. So describing or speaking from intuitions can be helpful even without bringing out clarity, because it gets us appropriately on the way to clarity.

3. Unnaturalness

An attempt to speak from the center of this intuition:

Influencing the genetics of our children is unnatural. For a few billion years, the genes have fallen where they may, and breaking from that is a big change. The ways that we and the world are adapted to that state of affairs is complicated and we don't understand or know about a lot of it.



There's a sanctity to bearing children. It's in some sense the root and heart of life, so it structures everything about the world and us. So messing it with the process of bearing children is messing with something at the core of life. That's not something that people can do in a way that accounts for what matters, so we should keep it as a region that's protected from manipulation.

What follows is more specific, explicit concerns that are part of what's pointed at by this intuition. Two general types of concern are first, then more specific ones.

Chesterton's fence

A general reason to suspect risk from messing with genes, is Chesterton's Fence. The idea is that if something was put there by someone, there's probably a reason, and if you mess with it without understanding the reason, you're liable to cause harm.

Evolution selected for reproductive fitness, and so also selected for many other things we consider good, like health and health of our children. So messing with reproduction might mess with mechanisms that evolution put there, so to speak, that have important good effects if not messed with.

Chesterton's fence is nearby some other reasons not to mess with old things. Old things are well-tested, and so have a fairly bounded downside risk (though also a fairly bounded upside risk). Also, we understand more about the risks of old things than the risks of new things, and how to deal with those risks and take advantage of opportunities; more generally, things in us and our world may be adapted to the old state of affairs, and adaptedness is in general good.

In general, using external calculation to direct a complex natural system is fraught because the director is cut off from information flows that the system itself has access to. See Seeing Like A State and the Hayek knowledge problem.

Goodharting

Selecting a genome according to some polygenic score (PGS) is an instance of Goodharting, i.e. trying to get a high score on a test in order to get something the test is supposed to measure.

For one thing, the PGS may not perfectly measure the trait you have in mind. In that case, when you push far out of the normal distribution, the PGS is no longer a good measure for that trait. Basically what you've been doing, is running a search for methods to easily increase the score. If you haven't been checking whether that method also increases the trait, not just the score, then the trait might not be increased (and other stuff might get messed with).

For another thing, the trait might be a good predictor of overall value and worthwhileness of normal humans, but might stop being a predictor of that when optimized too hard.

For a third thing, you might break other things by pushing far outside the normal. This is a general consideration which has a number of examples in the other items listed in this essay.

Non-normality of the child's life

Since a child with a strongly selected genome would be far outside the normal human distribution, ze might be genetically or environmentally caused to suffer or otherwise not live a life worth living.

§ Technical failures (how GS might produce suffering children)

  • Antagonistic pleiotropy. Polygenic scores for traits are composed by finding alleles that positively correlate with the trait. But it may be that many such alleles will also correlate negatively with (and indeed causally decrease) other desirable phenotypes. So pushing towards a very abnormally high score on some PGS may drastically decrease other desirable traits, which could cause suffering.
  • Epigenetics. Epigenetic modifications to DNA such as methylations, as well as off-DNA material such as transcription factors, control how DNA is expressed. These controls, and in particular sex-linked epigenetic imprinting, are crucial for healthy embryonic development. Depending on the genomic selection method used, the epigenetic state of the embryo might be abnormal, leading to unhealthy development.
  • DNA damage. Some GS methods involved manipulating DNA, e.g. mechanically moving chromosomes, inserting or deleting segments with CRISPR-Cas9, going through many steps of meiosis or mitosis in vitro, etc. These could cause broken or mutated DNA. (Example)
  • Breaking linkage disequilibrium. Some GS methods break the correlations between alleles at loci that are nearby on the same chromosome. In some cases this may be adaptive, so that the two haplotypes that are more common than usual are both more fit than the other two haplotypes. Depending on the strength of the linkage and the relative frequencies and fitnesses of the haplotypes, these effects may not show up in a GWAS (genome wide association study) using a linear model (or even one that allows for second-order epistasis). So a selected genome may tend to manifest less of the background linkage, which may be detrimental.
  • Increasing homozygosity. Without special care, some GS methods, such as iterated selection methods, would drastically increase homozygosity by assembling genomes with many segments duplicated, leading to many detrimental double-recessive genotypes. More subtly, it could be that segments of DNA that are especially high-scoring within given parents's genomes are more likely to be descendant copies of the same ancestral DNA segment, and therefore selecting for high-scoring segments might select for homozygosity. (Both of these effects should be easily predictable, though, just by sequencing and seeing if a genome is especially homozygous.)
  • Selecting for homozygosity. AFAIK, GWASes generally don't model dominance effects. So if there are many rare trait-positive variants that are homozygous detrimental, and these are selected for with sufficient power, the resulting genome will have many more than usual homozygous-rare (detrimental) genotypes.
  • Removing heterozygous advantage. It's sometimes said that the modal genome would be highly fit. It seems to me that this depends on how much balanced selection there is. If there are many loci undergoing balanced selection, then if we restrict to just those loci, the genome with each genotype the modal genotype will have much fewer optimal genotypes than a normal genome. If the balanced selection is due to, say, variation in the environment, then the modal genome could conceivably be overwhelmingly likely to encounter some change in the environment that it is worse equipped to handle, compared to a normal genome. (This is all speculative and I think unlikely, but I'm not sure.)
  • Epistasis. I.e. locus-locus interactions. GWASes generally assume additivity, i.e. model the phenotype as a weighted sum of the counts of some alleles. There may be reasons to expect these models to be valid well beyond the observed genotype-phenotype regime, but it could be that locus-locus interactions have effects that pop up when the genome is well outside the normal distribution. Even a GWAS that adds terms for locus-locus interactions could still allow for locus-locus-locus interactions that become relevant when optimizing. (I'd like to get more examples of this and get more clear on the math.)
  • Traits outside the regime of adaptedness.
    • Many biological systems exhibit strong nonlinearities in performance that lie only somewhat outside their usual operating regime. E.g. if you can exert yourself and burn some calories to deadlift an X-pound weight, that doesn't imply that you can burn four times the calories and deadlift a 4X-pound weight; given twenty days worth of food, your stomach can't just go ahead and process it, as usual, over the course of twenty days. Within the regime of adaptedness, everything works, even as things change; outside the regime, nothing works.
    • Robert Wadlow was almost 9 feet tall. He had circulation problems and died at age 22. Since, by these numbers, Wadlow was in the ballpark of $(272-176)/6.35 \approx 15$ standard deviations above the mean male height, his phenotype wasn't the result of normally distributed polygenic variation (which would put his rarity in the ballpark of 1 sample in more than $10^{40}$); rather, he had a hormonal abnormality. Still, assuming that some of his major health issues were caused by his height rather than directly caused by the hormones, we can infer that the human developmental program can't cope with such an extreme phenotype. This could be described as a very-high-order epistasis: any given 50 height alleles have no epistasis with each other, because the circulatory system can perfectly well develop and function in a body that's a few centimeters taller than normal; but a set of 1000 height alleles all together, is in extreme epistasis with each other, in the sense that if they are all present there's a strong negative effect. Selection for a trait causing severe issues has precedent in plant and animal breeding.
    • If GS were used to select for height, there would be issues like with Wadlow (to the extent that his issues were caused by his height, rather than through some other pathway from growth hormone). What if GS were used to select for intelligence? Some possible problems: skull problems (size, closure), blood flow problems to brain tissue, birthing problems, brain cancer (maybe correlated with neurogenesis / plasticity?), metabolic demands (glucose, material for myelination and neurotransmitters, etc.), mechanical overpacking in the brain (e.g. restricting CSF, neurotransmitter flow, etc.), interneuronal "conflict" (if humans are tuned to be near a threshold that allows exploration while avoiding intractable conflict), plaque / other waste, exhausting capacity of some shared structures such as the corpus callosum, exhausting physical room for receptors / pumps / synapse attachment, disrupted balance of ions. How likely are these? What are others?
    • There may be limited capacity to be heterozygous in general, and heterozygosity in general is increased to the extent that one selects in favor of rare variants. E.g., perhaps extreme heterozygosity would interfere with pairing during the prophase of meiosis, rendering these people infertile.
    • Conceivably, even just removing a large number of variants which are purely deleterious in normal humans, could have this same effect. E.g., if height is controlled by many such genes, this change would make a child who's extremely tall, with the accompanying problems. It could also conceivably be the case that this genomic selection would interfere with beneficial hormesis. E.g. if a child is genomically predisposed to have two different layers of zer immune system both be very strong, then maybe one of the layers is so strong that it fends off almost all infections, which makes the second layer never develop properly, and then the second layer fails catastrophically when the person gets really sick.
  • Restricted mating. Depending on the GS method, GS children might be at high risk for having genetically diseased children themselves if they reproduce naturally, forcing them to themselves use some form of GS.
  • Survivorship bias. In the pipeline from primordial germ cells, to gametes, to fetuses, to babies, very many would-be offspring-contributions are lost. Often this is due to genetic mutations. Mutations that would be deadly to, say, an oocyte, would not show up in the adult population. PGSes usually measure people who were already born alive. So those PGSes will not account for variation that would have definitely prevented reaching adulthood. So, in principle, selecting too hard on such a PGS, or generally making the genome too unnatural, could drive the genome into a region that's developmentally harmful, without warning.
  • Selecting for culture viability. Many GS methods involve making a baby from a cell that comes from stem cells cultured in vitro. Culturing cells applies selection pressures which lead to genetic aberrations. E.g., oncogenes are often prevalent in culture lines. With more effort, culture lines can be filtered to discard bad mutations, but we might not know all the relevant mutations. For example, some GS methods ask for culturable haploid cells. That might apply an abnormal selection pressure towards genes that make haploid cells viable in culture, with unknown effects.

§ Social failures (how GS children might be especially harmed by their social environment)

  • Missing local context. GS children might lack the peers, mentors, institutions, and culturally accumulated knowledge that would hypothetically nurture them. Parents might have a lot of trouble helping their GS children.
  • Hostile broader context. GS children might be targeted by non-person anti-abnormality forces such as anti-intellectualism (compare violent anti-Semitism) or envy (compare people vandalizing Teslas).

Strain on parents

Birthing and caring for a GS child would be challenging. E.g. selecting for intelligence might select for cranial size, which would make birthing more difficult and harmful for the mother. The child might have higher metabolic needs, have difficultly sleeping, have more extreme and less regulated emotions, be more cleverly troublesome, be more difficult to communicate with, and so on, taxing the parents. In general the parents would bear the brunt of many of the other potential downsides.

Messing with the gene pool

§ Decreasing genetic variation

GS doesn't necessarily reduce variation across the board, because e.g. it could increase the frequency of a rare variant, but PGS targets are likely to be heavily weighted on decreasing variation. This could weaken the ability of humanity to genetically evolve in response to a changing world. It could also lead to diseases. (Intuitively, some of these risks are only much of a risk in extreme scenarios where most people are using strong forms of GS for multiple generations, but I have not checked the math and this intuition could be incorrect.)

  • Holstein cattle. As an example of the power of selection, take Holstein Friesian cattle, a widespread breed of dairy cattle. There are many millions of Holstein cattle, but their effective population size is less than 100. A (presumably same/similar?) measure of effective population size in wild populations of various species found that usually the ratio of effective (genetic) population size to census (headcount) population size is more like 1 to 1, or 1 to 10, or 1 to 100, but almost never 1 in 10⁵ or 10⁶ as with Holstein cattle. (More detail and thought would be needed to see the implications of this for human GS, but we see that at least one real-world breeding program has produced an extremely abnormally low effective population size.)
  • Selecting against invisibly heterozygous-advantaged variants. E.g., a mutant HBB gene is neutral or slightly harmful when heterozygous in normal environments, is very harmful when homozygous (sickle-cell anemia), but protects against malaria when heterozygous. If that variant shows up as mildly negative in a GWAS on a non-malarial population, and then is selected against by GS, then the population is more vulnerable to malaria.
  • Linkage disequilibrium blowup. Even without being directly selected against, a variant might be eliminated by coincidentally not being present on DNA segments that are strongly selected for. For example, if chromosome selection is widely used, chromosomes that score extremely highly on whatever PGS is used will be propagated to a great extent, and variants not appearing on that chromosome will be driven down in frequency.
  • Increasing homozygosity. Most GS methods will strongly select for certain segments of DNA, making those segments very abnormally widespread. Subsequent generations would then be at much higher risk of large amounts of homozygosity. This might be especially bad if the GS selects for rare variants that are very deleterious if homozygous; even if the GS method can orchestrate that those variants only show up heterozygously, it still could greatly increase their raw frequency and hence greatly increase the chance of very deleterious homozygosity subsequently.

§ Decreasing phenotypic variation

  • Selecting away weirdness. By their nature, PGSes derived from the usual sort of GWAS will entirely or almost entirely ignore weird people. If those weird people use GS, then in general the use of GS shrinks the weirdness. For example, there could be some epistasis that makes people have sort of quirky phenotype that's valuable in some way to that person or to others. That epistasis probably wouldn't show up in GWASes, and then might be selected against very strongly (if each individual component is low-scoring on some PGS).
  • Discarding dimensions of variation. For example, it might be that intelligence has relevantly more than one dimension, and it might be that selecting for measured g-factor selects against variation in some of those dimensions. So the resulting GS children will all have the same kind of intelligence, and problems in the world that are much more profitably addressed with some weird kind of intelligence will go unaddressed.
  • Converging on "optimal" points. Depending on the kind and power of GS, it might produce children with a fixed genome or small set of genomes. For example, chromosome selection, depending how it's used, could use up all the available inter-chromosome variation in one generation, with nowhere further to go.

§ Blocking the flow of information coming from natural selection

Natural selection——the filters on surviving, thriving, and mating that determine which organisms reproduce with which organisms——induces a complicated, changing breeding structure that takes many things into account, and which could be described as pumping information into the gene pool. Artificial genome selection cuts off this flow of information to some extent, replacing it with selection on human-measurable features. So if GS is widespread, then much of the information about fitness is lost, except insofar as the PGSes used also can provide that information.

Harm to animals

Some GS technologies would benefit from animal experimentation, including observing adult GS animals. This could cause animal suffering by creating malformed animals. Also, GS technologies might be used to create more efficient farm animals, which could in theory cause a Jevons effect, e.g. increasing the total number of individual cattle farmed, even though each individual cow produces more.

Destabilizing institutions

Current social institutions are adapted to the current distribution of genomes, whether by conscious design or just by people coping with their local reality in a way that co-adapts their behavior with the behavior of other people. Changing the distribution of genomes might at least temporarily disrupt these adaptations, causing institutions to decrease in capacity or disintegrate.

4. Objectification

An attempt to speak from the center of this intuition:

Genomic selection takes humans and humanity as objects to be manipulated. This cuts off the relating-to that's necessary to deal rightly with other beings that are the same sort of being as you; the empathy, the openness to co-creating language and ideals, the willingness to be moved by negotiation, the care for them for their own sake, the enjoyment for its own sake of interacting with another. Making humans the target of strong optimization pressure both incentivizes and socially evokes optimization pressure that treats humans as instrumental, which unjustly suppresses respect that should be shown towards the self-sovereignty of the internal locus of agency of each human. That harms humans who are treated that way, creates conflict with them. Parents don't want to think of their kids as things to optimize.

Children can't give prior consent to being created. GS is in some ways more risky than natural childbearing, and GS is a deviation from the natural state of affairs. So consent is more of an issue with GS. (More so, to the extent that the necessity of consent depends on deviation from a prior default or natural state of affairs.)

Harm involved in conscription

GS children might have high capability in some domains, and so might be especially targeted for conscription in general. E.g., conscription by governments into literal wars or other conflicts, conscription by ideologies into missions, conscription by social class into class conflict, conscription by non-person emergent social forces into self-reinforcing non-person behavior, or conscription by parents or local social context into some specific life path. Conscription in general involves harmful behavior such as deception, punishing agency, withholding help, threats, isolation, sacrificing epistemics in favor of superficial coordination, "pointless" harm to reinforce surrender mindsets and cripple direct capacity to resist, applying these tactics to people related to the target, and controlling in general. So GS children might be especially targeted for harm.

Less intensely, but more likely, GS children may be socially pigeonholed, and may be socially pressured to take on tasks. This restricts their freedom.

Deficient social context of intrinsic care

Parents who use GS for their children might be selected for treating children as means to an end rather than ends in themselves, or for being in a local social context that does that. Also, the act of using GS might cause parents or their social context to change the value it places on children's lives for their own sake in favor of the instrumental value of children: by providing social proof that communicates "our values are whatever would have recommended to use GS", using GS says that children are worthwhile and worthy of care because of the traits selected for, and perhaps implicitly not because of their own experiences and values. A social context like that could leave the child starved of what is called love, acceptance, and security.

Moving the Overton window towards coercive eugenics

Taking "eugenics" very broadly to mean any effort to affect the genes of future humans, we can distinguish coercive from non-coercive eugenics. GS is non-coercive, while for example forcibly sterilizing or killing people to keep them from reproducing is coercive. (This is different from the distinction between positive and negative eugenics (encouraging vs. discouraging certain genes from being passed on): banning abortion for some subpopulation is positive eugenics but is coercive and undesirable, while elective mild embryo selection to screen for severe disease is negative eugenics but is non-coercive and desirable.)

Non-coercive but extreme eugenics, such as strong forms of elective GS, might break the wall that society puts up as a mind-control field to prevent people from gathering political will around coercive eugenics. Public non-coercive eugenics could put into reflective common knowledge the belief that manipulating the gene pool, in general, is acceptable, which implies (within the blunt, slippery, dream-like logic of political will and large-scale common knowledge) that coercive eugenics is acceptable.

Further, eugenics, more narrowly construed to mean influencing a population's gene pool and reproductive patterns, even if it's non-coercive or not transparently coercive, is tied up with the idea of controlling groups of people. Though GS is about creating individuals rather than controlling groups, if GS is practiced widely, the conceptual distinction between creating individuals vs. controlling groups could, like the distinction between coercive and non-coercive efforts to affect genes of future humans, be blurred in the public sphere. That could add to political will around controlling groups of people in general.

Sacrificing well-being for competitive advantage

To make genes controllable is also to make GS an available move in competitions. GS to win in a competition may select against well-being by selecting in favor of traits like single-track thinking and self-sacrifice. Competitive pressure to increase selection power could drive people to cut corners on safety, resulting in suffering GS children.

Making society overly prioritize industrially optimizable traits

When a desire is easier to fulfill, it unfolds into the structure to fulfill it, e.g. the technical knowledge, the legal environment, the social norms, and the social institutions for coordinating to fulfill it. This reciprocally causes and is caused by the desire, and the common knowledge among people of the desire, and the resulting political/coordinative willpower. Making it feasible and inexpensive to strongly genomically select for some trait might make society desire and/or pursue that trait more than would be good. E.g., being able to select for intelligence could make society value intelligence over morality; as in, "we don't need to worry about raising people to be good people because we can just make them intelligent".

Making society deprioritize non-GS children

By a similar token, if people think that GS-children have qualities that make them worth investing in, they might uninvest in non-GS children.

Fetishizing traits or genetics

Some people incorrectly attribute desirable concomitant qualities to traits such as intelligence in a rigid way that is not suitably responsive to evidence. Those people have some reason other than truth-tracking for their opinions. That reason may cause them to be reckless with GS technology, and to have incorrect assumptions about how to interact with and what to expect from GS children. Likewise, some people rigidly attribute desirable qualities to genetics, with the same potential problems. For example, such a person might rigidly assume that a child genetically predisposed to be intelligent will be creative, curious, clever, and compassionate, without guidance; and they might fail to notice that the lack of guidance is the issue, instead blaming the child for not realizing the potential that is surely there.

5. Transgression

An attempt to speak from the center of this intuition:

Manipulating the genes of future children is not something that society accepts, and doing things that society doesn't accept creates conflict. Conflict causes harm.

Punishment

  • Parents, scientists, technologists, funders. People involved in doing GS might be punished by reputational damage, loss of jobs, loss of influence, harassment, violence, stigma, sanctions, fines, or imprisonment. For example, He Jiankui and his collaborators were fined and imprisoned after "widespread criticism" for using CRISPR/Cas9 to edit out a gene from two twin girls to protect them from HIV. (His experiments were indeed dangerous and irresponsible.) Steve Hsu was made to resign from an administrative position (though retained his tenured faculty position) due to protests about his discussion of the genetics of intelligence. This might spill over to other projects; e.g. if a funder funds both GS and AI alignment research, then AI alignment research might suffer reputational damage "by association".
  • Children. GS children might be stigmatized and treated poorly by institutions.
  • Future GS, other biotechnology. A GS project that goes poorly might cause backlash and legal restrictions, making it more difficult for future GS projects to succeed. This might spill over, so that other related biotechnologies such as IVF, IVG, adult gene therapy, GS in animals, and so on, are also restricted or banned more than they'd have been otherwise.

Criminality

  • Keeping secrets. Because of the threat of punishment, people who work on GS are pressured to keep secrets. Keeping secrets pressures people to lie, and creates a social context that makes it easier to behave harmfully because there's less oversight. Keeping secrets forces people to lie and expect to be lied to and expect to not be able to resolve contradictions, so it makes it harder to have a shared social reality that's coherent and grounded in physical reality and goodness. That context reinforces anti-patterns that rely on social reality being incoherent or false, such as fractal deference to incoherent power-seeking behavior.

Fanaticism

In the same way that people who work on GS are selected for transgression, they're also selected for having very strong motivation, strong enough to overcome the avoidance of transgression and to spur difficult feats of bioengineering. Ideology (group-reinforced stories about overarching meaning) and other strong motivators might spur people to be reckless with GS, e.g. by not being cautious to avoid creating suffering children or by trying to use GS to gain power rather than to achieve life-increasing ends.

State conflict

States might go to war over the use of genomic selection.

6. Misalignment

The human capacity created by GS might be directed by values that we don't like. And the creation of GS people might negatively affect how all human capacity is directed.

Intentional misuse

§ ...of component technologies

In developing GS technology, other technologies might be developed that could be dangerous by themselves. I'm not aware of many of these for GS, and would like to know if there are more.

  • Viral vectors. If viral vectors are used to deliver DNA into cells, those viral vectors might threaten accidental or intentional viral outbreaks.
  • Genome synthesis. This could be misused to create pathogens.

§ ...of GS technologies

  • Fraud by GS providers. Apparently multiple doctors have used their own sperm to sire hundreds or thousands of children without parents's consent.
  • Selection for usability. Conceivably, a government, or other large organization such as a social movement, might create GS children selected for traits that make them useful as tools for the organization, such as obedience or willingness and ability to do violence. This would make the GS children potentially harmful and/or take away some of their agency.
  • Selection against strength. A deaf couple intentionally selected a deaf sperm donor so that their child would be deaf. Other parents might want to use GS to have a child who's blind, a dwarf, or has some other non-normal trait that prima facie makes them more likely to have a more difficult live.
  • Selection for values. Apparently, a wide variety of central political attitudes have substantial genetic components, and likewise for personality traits ("The genetics of human personality", Sanchez-Roige et al., Sci-hub). Parents or governments might select children to hold certain attitudes or have certain personality traits. In many cases this would influence the center of mass of the values of humanity in a seemingly non-epistemic way, and would take away agency from the children.
  • Competitive selection and inequality. (Cf. the Objectification section.) To make genes controllable is also to make GS an available move in competitions. A competition over genes could, Moloch-style, emergently lead to outcomes no one wants. E.g., selecting for tendency to gain power or money, might select in favor of sociopathy, single-track thinking, non-epistemic charisma / persuasiveness, and so on, which could lead to a world that's not desirable to live in for any of its inhabitants, as in The Rapacious Hardscrapple Frontier. Alongside, such a competition could make the world even worse for people who don't want to or don't have the resources to select for competitiveness. Even if the world has some people who are living worthwhile lives, many or most people could end up left behind and excluded from gains.

§ ...of technologies developed by GS children

Free GS children might use their capabilities to develop technologies that are dangerous. Even if the GS children don't have any intention of doing harm, those technologies might be misused by others. In other words, GS children might increase the speed and volatility of technological development without much increasing the overall wisdom of humanity, and since technology tends to spread, that might greatly increase the total amount of unwise use of technology.

Conscription

The film Gattaca (1997) depicts a world with widespread genomic selection, where people with high-scoring genomes are conscripted into a social class that excludes and mistreats the non-GS people. Huxley's Brave New World depicts a similar class dynamic enforced with coercive eugenics, poisoning babies, drugs, sexual abuse, and strong norms against thinking. Whatever intelligence can survive that state of affairs, is put toward maintaining it.

More generally, governments, ideologies, classes, and other coalitions (including whoever develops GS technology) might deceive and/or coerce GS children into fighting on some side in a conflict. This takes away agency from the children, reinforces conflict, and helps coercive forces thrive.

Even more generally, one might think that humanity as a whole has net-bad motivations by default, i.e. wants to do things that are bad on net. (I disagree.) If so, then empowering humans is bad on average because it enables them to do more of what they want to do, which would be bad.

Internal misalignment

  • Antagonistic pleiotropy with unmeasured traits. Some crucial traits, such as what is called Wisdom and what is called Kindness, might not be feasibly measurable with a PGS and therefore can't be used as a component in a weighted mixture of PGSes used for genomic selection. If there is antagonistic pleiotropy between those traits and traits selected for by GS, they'll be decreased.
  • Traumatizedness. Due to being potentially abnormal, invidious, conscripted, or around harmfully transgressive people, GS children might be traumatized and then harm others out of fear.
  • Antagonistic coalition of enhanced humans. Humans selected to have very high physical or mental capacity might, by their nature, be misaligned with the values of the rest of humanity. In some cases that would be good, since humans selected to be intelligent will predictably make moral progress. But, for example, it could be the case that humans are kind when they're within some regime of interdependence, and outside that regime they are cruel. In that case, highly enhanced humans would be a danger to what we care about. Intrinsic misalignment is also related to intuitions about being "left behind" by GS. Also, enhanced humans might engage in "runaway assortative mating" strategies, meaning that they progressively decrease mating with non-enhanced humans; reproductively isolated populations might (be feared to) have their values less coupled with the values of the rest of humanity.

Destiny change

  • Generally, by changing what sort of humans exist, GS would change what humanity is, where it is going, and what forces determine where it is going. That change could be a change for the worse.
  • For example, one might think that there's a sort of god of humanity, constituted by default-human-evolution as it has been happening up until now. In this view, the default process of natural selection will select for fitness, which is goodness; humanity is thus improved over time. (I strongly disagree with this view; there's no good reason to think that survival of the fittest is remotely compatible with what humans or humanity like, care about, or would prefer after more thought.) On such a view, GS would be bad because it would disrupt natural selection.
  • Similarly, one might think that it's good for some people to have children and others not, because that reflects the values of humanity as expressed in mate choice. (While most people would profess to be against eugenics, many of those same people would also happily joke, in private, that an ugly person "doesn't need to contribute to the gene pool" or similar. Such people should be thrown in moral debtor's jail.) On such a view, GS would be bad because it would disrupt the "soft eugenics" of mate selection.