What If Everyone Was A Clone?

A highly unsatisfying area of science fiction is its handling of human clones. With few exceptions, the use of ‘cloning’ draws on effectively no science, and seems to imagine human clones not as they are (identical twins separated in time) but as sort of AIs being copied in a computer. (SF also usually makes bizarrely weak, inept use of the possibilities there too, but that’s a different discussion.) Even the few works which do use cloning, do not show much insight into it (like Brave New World). Given all that we know now about heritability, the personalities & lives of identical twins or ‘multiples’, asexual reproduction in complex organisms (sometimes spontaneously emerging out of sexually-reproducing organisms), and the results of large-scale commercial cloning in animals, we can do better in imagining alternative worlds where humanity reproduces primarily or entirely conally (perhaps asexually).

single mutation to create clonal reproduction in bees: “A Single Gene Causes Thelytokous Parthenogenesis, the Defining Feature of the Cape Honeybee Apis mellifera capensis”, Yagound et al 2020

rapid population explosion: https://en.wikipedia.org/wiki/Marbled_crayfish

example of strange coordination possibilities: the giant pyrosome https://en.wikipedia.org/wiki/Pyrosome https://oceana.org/marine-life/giant-pyrosome/

if unethical to split up triplets, why is it ethical to implant single embryos when it could be twins or triplets instead?

• female clones like crayfish, bees https://www.newscientist.com/article/2280153-a-single-honeybee-has-cloned-itself-hundreds-of-millions-of-times/#Echobox=1623202651 https://royalsocietypublishing.org/doi/pdf/10.1098/rspb.2021.0729

• snakes: eg. the now-global https://en.wikipedia.org/wiki/Indotyphlops_braminus

• “Haplotype divergence supports long-term asexuality in the oribatid mite Oppiella nova”, Brand et al 2021 https://en.wikipedia.org/wiki/Matthew_Meselson

• condors: https://academic.oup.com/jhered/advance-article/doi/10.1093/jhered/esab052/6412509

• reproductive cloning

• https://en.wikipedia.org/wiki/Klepton https://www.popsci.com/female-salamander-kleptogenesis/

• all-female termites https://www.pnas.org/content/118/51/e2009533118 https://gizmodo.com/drywood-termites-clone-all-female-colonies-1848452516

• “Virgin Birth: A genetic basis for facultative parthenogenesis”, Braun et al 2022 (genetically engineering fruit flies partially capable of clonal reproduction)

overall picture: much more predictable yet more diverse

~50% heritability, maybe 20% shared, but clones can arrange standardized environments above average similar psychologically on intelligence/personality, physical appearance/attractiveness, but more conflicting on ideology/politics, various low-h² exceptions like longevity/homosexuality

benefits: faster tailored training (see polo clones—may well be able to develop good flow charts pace SSC), likely failure modes, prediction of health issues

intensified testing, accelerated maturation: apple breeding example

https://www.royalroad.com/fiction/25137/worth-the-candle/chapter/491050/the-aviary failure mode: Doris Finch defecting

inclusive fitness like eusocial insects lineages/clans: deep loyalty, idiosyncratic rituals, media/leisure time preferences (highly heritable see /Amuse) allowing clan recommendations but still markets, division of labor

(“luckily, we are only men—and cowards”)

identical twins: secret languages!

cost of variety: product variety will fall by orders of magnitude, yielding several-fold falls in cost due to economies of scale + experience curve effects

occupational castes on steroids? winner-take-all effects, particular clans will dominate particular niches—over many lifetimes, clans will steadily assort towards occupations/niches they are particularly suited for, because much cheaper to train & recruit reliably see (see /Clone Special forces dogs example) clans may possibly use an extreme level of internal selection based on their needs; a useful tactic might be carefully calibrated emotional neglect/abuse to create a permanent drive for accomplishment… extreme honor codes might develop where failures might see it as honorable to commit suicide to free up resources for the clan—not as extreme as eusocial insects permanent aristocracies/monarchies? factionalism would make democracies much more difficult

highly mobile human capital: clan can just relocate and raise next generation in the local language/culture to carry on

selection against individualism/high discount rates: clans with low discount rates and high solidarity will willingly stick together and invest in growing the next generation, while short-sighted individualists won’t if (expensive) reproductive cloning, then regular attrition of low-earning clans as they hit lower bound of 0

male vs female variance: most male lineages will die out except for a few elites?

potential increased total variance & range: selection processes will select out and preserve clonal extremes (eg extreme body types in world-class sports). could total variance increase anyway? I think this might be true even without any selection, because of regression to the mean. imagine you draw a multivariate normal and then you fix each point and draw normals centered around each; that’s clones. but with regular reproduction, you pick a random other point to mate with, and that will likely be not just smaller on any given variable, but potentially the opposite sign, even. the panmixia reproduces the original distribution, but the clonal clumpy distribution expands out at the tails. if you only start with a few clones so the tails aren’t represented, then yeah, it does have smaller variance, but if you have a few million or billion like I’m imagining…

advantage: very cheap easy science—see the milk power analysis for example. primary problem: external validity! “In our world, scientists can cure cancer in mice; in Clone World, scientists can cure cancer in Mikes.” clan secrets

Clans would be pushed into institutions by other pressures, like the accumulation of trade secrets (clan secrets), which would be guarded jealously, and for the elite clans in any field, may be far more valuable than any mere genes would be. Clan secrets are qualitatively and quantitatively different from any mere family wisdom.

A member of a clan will benefit from enormous implicit experimentation and knowledge. All clansmen will constitute an ongoing longitudinal study of preferences, options, tactics, health problems, weaknesses & strengths across the whole range of human existence. If there is a propensity for, say, celiac disease, it will not come as news but will be watched for from birth; with such knowledge, the clan will reshape its environment to best suit its genes—the chefs of a clan prone to celiac will not even use gluten-containing recipes in the first place, to avoid triggering it through autoimmune reactions (and because it is hard to miss tasty gluten-containing treats if you never ate them). Bored and need a movie or book recommendation? There will be a vast library of tailored recommendations, far more accurate than guessing from some movie critic’s reviews or trusting in some algorithm. Senior members will have built up stores of tricks to cope with their psychological quirks and weaknesses. The clan might run its own schools, with equally tailored curricula, whose teachers, having taught thousands of students alike as peas in a pod, can iterate with each class towards better explanations, better quizzes, better classroom management skills, and head off their likely misunderstandings from the start. Given cryptophasia in identical twins, how families tend to have their own peculiar catchphrases & in-jokes, and the rapid drift of small insular groups into jargon then accent then argot then dialect then language, and the high internal uniformity, clans over a certain size will doubtless develop their own private language, speaking a more global tongue with outsiders. (Examples here would be minorities which speak one language at home, and another at work, for example the Amish originally spoke German but now speak a Pennsylvania German language while reluctantly learning English for speaking with “the English”—which makes it that much harder for any Amish to leave the community…) Even ‘deviant’ members will likely deviate in ways that are, in aggregate, fairly predictable, and the knowledge of how to help rather than harm can be built up like any other. All of this knowledge will be implicit, encoded into the environment; a member who leaves can’t take it with them, and may not realize that the clan cuisine was gluten-free until they develop celiac years later. This will represent an enormous deterrent to defection.

This holds true for explicit experimentation & knowledge. From a statistical perspective, small increases in variance require painfully large increases in n to overcome them; the use of identical twins or clones can easily reduce an experiment’s required n by an order of magnitude (or two) for highly heritable traits (as Student pointed out, for the Lanarkshire milk experiment, use of twin pairs would have reduced the sample size by TODO). Or to put it another way: a clan of n members could potentially run 10–100× more experiments of equivalent quality than an organization using random population samples totaling n people. And this would hold for every time period. Over centuries, this could result in a huge knowledge gap. This explicit knowledge is also resistant to defection or imitation. The catch is that within-clan implicit & explicit learning will have extremely high internal validity, but the external validity will be lower, for precisely the reasons that the internal validity is high—much of the efficacy will be clan-specific. In our world, scientists can cure cancer in mice; in Clone World, scientists can cure cancer in Mikes. This reduces the value of intellectual piracy: even if clan secrets are stolen, the tricks may not necessarily apply to anyone else, and no one, including the original clan, necessarily knows which is which. Clans could invest in trials of hypotheses verified within-clan which are designed to keep the statistical power of twins while testing external validity (eg trials using pairs, with each pair drawn from a different clan), but this would require a lot of coordination between clans, would tend to harm the clan that verified the hypothesis (they already know it’s true & are exploiting it, and establishing its external validity helps their competitors), so “trade secrets” would be a far larger fraction of knowledge, and this would be a serious impediment to progress and science.¹

Example of novel research: directed mutation experiments, a bit similar to the mutation lagoons. Clones would have an extremely strong sense of identity with their clan and be willing to undergo painful experiments to ‘pay it forward’ (for all the benefits they reap from centuries of self-optimization) and so it might be a standard thing for each generation of clones to receive a handful of unique mutations each which have been selected for their high prior Value of Information. with a shared generation, highly optimized environment, extraordinarily detailed historical statistics on every trait, genetic uniformity, and extremely low discount rates, continual experimentation may be quite profitable.

crime: surveillance and eyewitness even more useless; fingerprints still work, but not cheap DNA testing (distinguishing identical twins requiring extremely deep WGS); likely requiring collective punishment (gimmicks like Siamese twins going free on technicalities won’t wash in an entire world)

having a natural child: highly risky—unaligned competitor

militaries? I’m sure there would be specialized clans for that. like ex-special forces operators. although that leads to some pretty serious political problems, like ‘what stops them from launching a coup?’ in general, it’s really hard to see how politics would work in clone world. political/religious values are not that heritable so even with shared-environment there will be considerable within-clan disagreement, but the clans themselves would be enormously powerful political actors and it would be difficult for them to reach any kind of agreement or engage in regular politics. plus the question of clones creating aristocracies which actually work. even if you have wild-types regularly being born and starting new clans, creating to some degree a ‘circulation of the elites’, that simply creates a ratchet effect.

romance is not heritable, high twin discordance in mates: https://slatestarcodex.com/2016/10/21/the-heart-has-its-reasons-that-reason-knows-not-of/ http://cogprints.org/773/3/155.pdf

against ems:

• machine teaching

• brain imitation learning

• style transfer

• gradient ascent

• model distillation/sparsification

• meta-learning

library(skimr)
originals <- rnorm(1000000)

h2 <- 0.5
clonesAll <- as.vector(sapply(originals, function(o_pheno) {
    o_gene <- o_pheno * sqrt(h2) # regress back to mean
    rnorm(100, mean=o_gene, sd=1 - sqrt(h2)) }
    ))

threshold <- 0.35
clonesSelected <- as.vector(sapply(Filter(function(o) { o > threshold || o < -threshold; }, originals),
    function(o_pheno) {
        o_gene <- o_pheno * sqrt(h2) # regress back to mean
        rnorm(137, mean=o_gene, sd=1 - sqrt(h2)) }
    ))
skim(originals); kurtosis(originals)
#   variable missing complete       n     mean sd    p0   p25    p50  p75 p100     hist
#  originals       0  1000000 1000000 -0.00022  1 -4.55 -0.67 0.0014 0.67  4.7 ▁▁▂▇▇▂▁▁
# [1] -0.00488809495
skim(clonesAll); kurtosis(clonesAll)
#   variable missing  complete         n     mean   sd    p0   p25     p50  p75 p100     hist
#  clonesAll       0 100000000 100000000 -0.00016 0.77 -4.04 -0.52 0.00042 0.52 4.13 ▁▁▂▇▇▂▁▁
# [1] -0.00301335287
skim(clonesSelected); kurtosis(clonesSelected)
#        variable missing complete        n     mean   sd    p0   p25     p50  p75 p100     hist
#  clonesSelected       0 99393226 99393226 -0.00026 0.88 -4.07 -0.68 0.00086 0.68 4.13 ▁▁▃▇▇▂▁▁
# [1] -0.614314287

so looks like in clone world, overall ‘diversity’ does go down as measured by standard deviations, but with even slight selection processes, you can get higher kurtosis. so clone world resembles more of a square or a V than a bell curve ∩.

library(skimr)

originals <- 200
cloneMultiplier <- 30
h2 <- 0.8
df <- data.frame(Clan=numeric(), Phenotype=numeric(), Clan.mean=numeric(), Clan.sort=numeric())
for (i in 1:originals) {
    o_pheno <- rnorm(1)
    o_gene <- o_pheno * sqrt(h2) # regress back to mean
    c_pheno <- rnorm(cloneMultiplier, mean=o_gene, sd=1 - sqrt(h2))
    df <- rbind(df, data.frame(Clan=as.ordered(i), Phenotype=c_pheno, Clan.mean=o_gene, Clan.sort=-abs(o_gene)))
    }
plot(density(df$Phenotype))
qplot(Phenotype, Clan.sort,                             data=df) + theme_bw() + theme(legend.position = "none")
qplot(Phenotype, Clan.sort, color=as.ordered(Clan), data=df) + theme_bw() + theme(legend.position = "none")

qplot(jitter(Phenotype), Clan.sort, data=df) + theme_bw() + geom_point(size=5)

library(mvtnorm)
library(MBESS)
## simulate:
mean(replicate(100000, max(rowSums(rmvnorm(10, sigma=cor2cov(independent, sd=rep(sqrt(0.33*0.5),35)), method="svd")))))

library(mvtnorm)
library(MBESS)

rowSums(rmvnorm(10, sigma=cor2cov(independent, sd=rep(sqrt(0.33*0.5),35)), method="svd"))

modest <- matrix(ncol=5, nrow=5, 0.3)
diag(modest) <- 1
modest
rowSums(rmvnorm(10, sigma=cor2cov(modest, sd=rep(sqrt(1-0.5),5^2)), method="svd"))

simCorrelatedIndices <- function(n, corMatrix, mean=0, h2=0.5, ntraits=5) {
    rowSums(rmvnorm(n, mean=rep(mean, ntraits),
        sigma=cor2cov(corMatrix, sd=rep(sqrt(1-h2), ntraits)), method="svd"))
        }
simCorrelatedIndices(1, modest)

library(mvtnorm)
library(MBESS)
library(ggplot2)

originals <- 200
cloneMultiplier <- 30
h2 <- 0.8
modest <- matrix(ncol=5, nrow=5, 0.3)
diag(modest) <- 1
simCorrelatedIndices <- function(n, corMatrix, mean=0, h2=0.5, ntraits=5) {
    rowSums(rmvnorm(n, mean=rep(mean, ntraits),
        sigma=cor2cov(corMatrix, sd=rep(sqrt(1-h2), ntraits)), method="svd"))
        }
df <- data.frame(Clan=numeric(), Phenotype=numeric(), Clan.mean=numeric(), Clan.sort=numeric())
for (i in 1:originals) {
    o_pheno <- rnorm(1)
    o_gene <- o_pheno * sqrt(h2) # regress back to mean
    c_pheno <- simCorrelatedIndices(cloneMultiplier, modest, mean=o_gene, h2=h2)
    df <- rbind(df, data.frame(Clan=as.ordered(i), Phenotype=c_pheno, Clan.mean=o_gene, Clan.sort=-abs(o_gene)))
    }

qplot(Phenotype, Clan.sort,                             data=df) + theme_bw() + theme(legend.position = "none")
qplot(Phenotype, Clan.sort, color=as.factor(Clan), data=df) + theme_bw() + theme(legend.position = "none")

qplot(jitter(Phenotype), Clan.sort, data=df) + theme_bw() + geom_point(size=5)

emergenesis: clans of geniuses/conquerors/other extreme outliers?

A caveat to discussions of somewhat reduced diversity is the question of emergenesis: an obscure behavioral genetics idea which is suddenly relevant here. Emergenesis is the observation that, while for the most part genetic & environmental variables can be modeled as independent variables which just add or multiply out, like for height or IQ, there are occasional traits where it’s clear that’s false and an extreme level of interaction/epistasis rules, like facial appearance or voice sound: identical twins are so identical on those because they have every gene in common, and if even a few were different, they would be different, perhaps as different as regular siblings. (This may be connected with the ‘coincidences’ strewn so liberally throughout the lives of identical twins reared apart—I recall one twin researcher telling me that they had reported only a few coincidences in their papers because they worried if they mentioned more, or the more extreme ones, they’d be accused of making it all up!) A trait which is highly heritable, but so non-additive that any genetic distance destroys phenotypic similarity, we might call an example of ‘emergenesis’. The interesting suggestion is that emergenesis may be why some traits appear to run in families but happen rarely, with no particular relationship to (additive, narrow-sense) genetic distance: traits like personality, leadership, motivation, or genius. If extreme outliers like a Napoleon were outliers due to additive traits, one would expect their children etc to at least occasionally rise to an equal level, but, while they are frequently highly competent, they are hardly ever outliers themselves, and this is hard to reconcile with the usual quantitative models like liability-threshold models or with the additive heritability of what seem like relevant traits. But if it was a conjunction of many sub-traits, such that a single missing sub-trait eliminated the possibility of being an outlier, representing a high level of epistasis, that would be consistent with the broad patterns.

For the most part, this is hard to test by looking at whether there are pairs of outlier twins: identical twins represent a very small percentage of the modern population, given child mortality & historical malnutrition/poverty I would guess that surviving healthy well-nourished twins were historically much rarer, and the existence of a twin unavoidably modifies careers (in examining outliers, would one wind up classifying just one twin as the ‘outlier’ regardless of how much the other twin was ‘assisting’, while if they had been raised separately, they would both qualify?). In Clone World, however, any emergenesis would be easier to see: whatever emergenic trait values are in the initial population will be replicated thousands-fold—if there is any clan of budding Napoleons, the potential will quickly manifest. Similarities of face or voice are not important, but if such vital but (hitherto) unmeasurable traits like “leadership” are emergenic, the relevant clans may rapidly proliferate, leading to a world with a level of diversity unimaginable to us—instead of a single gifted individual popping out of a distinguished but obscure lineage, it might be dozens every generation.

net gain: https://www.biorxiv.org/content/10.1101/2023.05.11.540325.full

https://blog.nickwinter.net/posts/what-if-you-could-copy-yourself