Over at sapiens.org, an anthropology magazine, author Elaine Guevara (a lecturer in evolutionary anthropology at Duke) takes modern genetics education to task. Making a number of assertions about what students from high school to college learn in their genetics courses, Guevara claims that this type of education imparts “zombie ideas”: outdated but perpetually revived notions that prop up biological racism. Her main topic is race, and she does offer some insights that modern genetics has given us about differences between geographic populations (I prefer to use “populations” rather than “races”), but these insights have been known for a long time. By failing to tell us that the errors earlier biologists have made about race have been refined and, to a large degree, dispelled, Guevara is herself deficient in describing the state of modern genetics.
Click the screenshot to read:
Guevara makes several accusations that, I think, are misleading. I’ll group her misleading conclusions under bold headings (the wording of those is mine). Quotes from her paper, or my paper with Luana Maroja, are indented and identified
1.) Human populations are not as different as we think, and the concept of “race” is incorrect: classical “races” are not genetically distinguishable. Guevara first cites a famous 1972 paper by my Ph.D. advisor, Richard Lewontin, “The Apportionment of Human Diversity“. The paper looked at genetic variation of 17 proteins detected by gel electrophoresis, apportioning the worldwide variation of proteins among individuals within a population, among populations within a classical “race”, and then between seven “races”. He found that of the total genetic variation seen worldwide, 85% occurred among individuals within one geographic population, 8% among populations within a race, and only 6% was found among races.
Thus races were not as genetically different as some people assumed. Lewontin concluded this (bolding is mine):
It is clear that our perception of relatively large differences between human races and subgroups [JAC: note that Lewontin’s “subgroups” correspond to what I would call “populations’], as compared to the variation within these groups, is indeed a biased perception and that, based on randonly chosen genetic differences, human races and populations are remarkably similar to each other, with the largest part by far of human variation being accounted for by the differences between individuals.
Human racial classification is of no social value and is positively destructive of social and human relations. Since such racial classification is now seen to be of virtually no genetic or taxonomic significance either, no justification can be offered for its continuance.
The first paragraph is correct. Later studies using better methods (DNA) have shown that yes, the apportionment of human diversity shows most of it within populations and only a fraction among populations or among “races”. The classical view that races like “Caucasion”, “Asian” or “Black” showed large and diagnostic genetic differences at single genes was wrong
But the second paragraph is wrong, too, because Lewontin did not raise the possibility (as I’m sure he realized) that small differences among populations (or the groups of populations that constitute classical “races”) can, taken across many, many genes, add up to significant statistical and biological differences. The failure to recognize the power of using genetic data from many genes (we have three billion DNA nucleotides in our genome) is called “Lewontin’s fallacy.” This fallacy was pointed out in 2003 by A.W.F. Edwards and has its own Wikipedia page.
The power of using many genes instead of just an unweighted average of data from individual genes is shown by several things, as Luana Maroja and I pointed out in our paper published in Skeptical Inquirer last year. For one thing, if there were no meaningful genetic differences between populations, you couldn’t use genetic differences to diagnose someone’s ancestry. Yet you can, and with remarkable accuracy, as anyone knows who is aware of their family history and has taken a genetic test like those offered by 23andMe. My test showed that I have complete Eastern European ancestry, with 98% of it from Ashkenazi Jews, which comports with what I know of my family history. (I also have a small percentage of genes from Neanderthals.)
Now this tells you the area of the world—the population—from which your ancestors probably came. It doesn’t deal with “races” as classically defined. Yet a multiple-gene analysis using four races that Americans themselves use in self-identification (African-American, white, east Asian, or Hispanic) can indeed be diagnosed with remarkable accuracy. As Luana and I said in our paper (I’ve bolded the money quote):
Even the old and outmoded view of race is not devoid of biological meaning. A group of researchers compared a broad sample of genes in over 3,600 individuals who self-identified as either African American, white, East Asian, or Hispanic. DNA analysis showed that these groups fell into genetic clusters, and there was a 99.84 percent match between which cluster someone fell into and their self-designated racial classification. This surely shows that even the old concept of race is not “without biological meaning.” But that’s not surprising because, given restricted movement in the past, human populations evolved largely in geographic isolation from one another—apart from “Hispanic,” a recently admixed population never considered a race. As any evolutionary biologist knows, geographically isolated populations become genetically differentiated over time, and this is why we can use genes to make good guesses about where populations come from.
And this:
More recent work, taking advantage of our ability to easily sequence whole genomes, confirms a high concordance between self-identified race and genetic groupings. One study of twenty-three ethnic groups found that they fell into seven broad “race/ethnicity” clusters, each associated with a different area of the world. On a finer scale, genetic analysis of Europeans show that, remarkably, a map of their genetic constitutions coincides almost perfectly with the map of Europe itself. In fact, the DNA of most Europeans can narrow down their birthplace to within roughly 500 miles. [See below for the European data.]
You can also identify the “classical” races used in self-identification using some morphological traits. As we wrote:
But you don’t even need DNA sequences to predict ethnicities quite accurately. Physical traits can sometimes do the job: AI programs can, for instance, predict self-reported race quite accurately from just X-ray scans of the chest.
Population differences summed across genes can tell us more, too:
On a broader scale, genetic analysis of worldwide populations has allowed us to not only trace the history of human expansions out of Africa (there were several), but to assign dates to when H. sapiens colonized different areas of the world. This has been made easier with recent techniques for sequencing human “fossil DNA.” On top of that, we have fossil DNA from groups such as Denisovans and Neanderthals, which, in conjunction with modern data, tells us these now-extinct groups bred in the past with the ancestors of “modern” Homo sapiens, producing at least some fertile offspring (most of us have some Neanderthal DNA in our genomes). Although archaeology and carbon dating have helped reconstruct the history of our species, these have largely been supplanted by sequencing the DNA of living and ancient humans.
Finally, there are nearly diagnostic differences between populations in genes that evolved in an adaptive way, like known genes for resistance to low oxygen, short stature or skin pigmentation. Here’s a figure from a 2015 Science paper by Sarah Tishkoff:
None of this would be possible if there were not significant genetic and biological differences between populations. We did not maintain that there are always diagnostic differences between populations at single genes that can group them into races, but that there are statistical differences in frequencies of variable genes among populations that are biologically meaningful. Nor did we claim that the classically-defined races are absolutely geographically distinct with little intermixing, or have nearly fixed differences in frequencies of variable genes. That’s not true, and all geneticists realize this now. (But note that even the classically defined “races” generally differ in gene frequencies and in some biological traits to an extent that they can be diagnosed.)
The reality is that we should be dealing with populations, and populations—roughly defined as geographically different groups of people that largely breed among themselves—show diagnostic genetic and morphological differences.
Yet Guevara misleads the reader by relying solely on Lewontin’s paper and neglecting all the work done since that showing that yes, there is diagnostic geographic variation among populations (note that Lewontin implied that the concept of “population” is about as meaningless as “race”). Here are cxcerpts from Guevara’s paper:
Lewontin published his calculations in a short paper in 1972 that ended with this definitive conclusion: “Since … racial classification is now seen to be of virtually no genetic or taxonomic significance either, no justification can be offered for its continuance.” His results have been replicated time and again over the last 50 years, as datasets have ballooned from a handful of proteins to hundreds of thousands of human genomes.
But despite huge strides in genetics research—leaving no doubt about the validity of Lewontin’s conclusions—genetics curricula taught in U.S. secondary and post-secondary schools still largely reflect a pre-1970s view.
This lag in curricula is more than a worry for those in the ivory tower. Increasingly, genomics plays a leading role in health care, criminal justice, and our sense of identity and connection to others. At the same time, scientific racism is on the rise, reaching more people than ever thanks to social media. Outdated education fails to dispel this disinformation.
Leaving “no doubt about the validity of Lewontin’s conclusions”? Nope. The apportionment of variation is without doubt, but not his conclusion that populations or races are without biological meaning.
None of the critiques of Lewontin’s paper, including Edwards’s famous clarification, are even mentioned by Guevara. And, in fact, I don’t know of any biologists in post-secondary genetics education who still teach the view that “Race and ethnicity are social constructs, without scientific or biological meaning.” (This is a quote from JAMA reproduced in the Coyne and Maroja paper. And perhaps some people teach this erroneous view, but no biologist that I know of.) That JAMA statement is completely misleading, as I hope I’ve shown above. The delineation and definition of classical races was itself misleading and often tied to racism in the past, but, as we see, even self-identified classical races can be diagnosed through genes or morphology, and generally do fall into clusters using analysis of multiple genes.
The last paragraph of Guevara’s quote above shows the ideological motivation behind her paper: we must dismiss the existence of biological races and genetic differences between populations because it emphasizes differences between humans, and thus could lead to ranking of human populations, and thence to racism. But, as Ernst Mayr recognized, accepting differences does not mean you have to view groups as being morally or legally unequal. We give a quote by evolutionist Ernst May quote in our Skeptical Inquirer paper:
Equality in spite of evident non-identity is a somewhat sophisticated concept and requires a moral stature of which many individuals seem to be incapable. They rather deny human variability and equate equality with identity. Or they claim that the human species is exceptional in the organic world in that only morphological characters are controlled by genes and all other traits of the mind or character are due to “conditioning” or other non-genetic factors. … An ideology based on such obviously wrong premises can only lead to disaster. Its championship of human equality is based on a claim of identity. As soon as it is proved that the latter does not exist, the support of equality is likewise lost. (Mayr 1963)
Thus, the second conclusion of Guevara is wrong:
2.) “High genetic variation exists within geographic regions, and little variation distinguishes geographic regions.”
Well, that’s sort-of true, but, as we said, that “little variation among geographic regions” can, when added up, diagnose populations sufficiently to not only tell you your geographic ancestry, but also to reconstruct the evolutionary and migratory history of human populations. Guevara dismisses these ancestry tests, though she doesn’t tell us why they are wrong:
Helping the zombie persist, direct-to-consumer genetic tests, like those offered by 23andMe and AncestryDNA, can reinforce misconceptions about human variation. These services have become many people’s primary reference point for human genetics information. To be marketable, the companies must communicate their results in simple, familiar ways that also appear meaningful and reliable. This usually entails simplifying genetic ancestry to bright, high-contrast colors, pinned definitively to geographic regions.
And yet, at the same time, Guevara admires the same kind of data—genetic differences between living populations (as well as “ancient fossil DNA”)—as being of value:
In addition to genomes from living humans, DNA extracted from ancient humans over the past two decades has revealed incredible insights. Across time, past humans frequently migrated, mated with, or displaced people they encountered in other regions—resulting in a tangled tree of human ancestry. The ancient DNA results refute any notion of deep, separate roots for humans in different geographic regions.
Well, there are deep roots for some groups (the Neanderthal lineage, for example, separated form the lineage leading to modern humans about 400,000 years ago), and this comes from both fossil and DNA evidence. The “tangled tree” may be correct in some ways (we did hybridize with Neanderthals, and other populations exchanged genes to different degrees), but it’s not tangled enough to completely efface the evolutionary history of human populations.
All this leads to a third misleading conclusion:
3) Races are social constructs. Any differences between races are largely caused by racism rather than genes. As Guevara says:
As laid out by a major professional association for biological anthropologists, race is a social reality that affects our biology. For the last several hundred years in the U.S. and other colonized lands, racism has influenced people’s access to nutritious food, education, economic opportunities, health care, safety, and more. As a consequence, and precisely because of the environmental influence on most traits, the social construction of race is a risk factor for many health conditions and outcomes, including maternal and infant mortality, asthma, and COVID-19 severity.
This again shows both an ideological motivation and a misleading conclusion. Even the classical biological races (and even more so worldwide populations) are NOT social constructs, but are associated with genetic, morphological, and adaptive differences. If races are purely socially constructed, how could you tell them apart in the first place? You need some kind of genetic marker. In the case of racism in America, the differences between African-Americans and whites were “constructed” based on skin pigmentation, hair texture, and other traits—traits based on genetic differences. Those differences served to mark out which people were considered different, and then “inferior”, though, as I said, genetic differences among people say nothing about moral or legal equality. THAT is the lesson that needs to be imparted, not the falsity that there are no genetic differences among groups.
Now Guevara may be correct that the “social construct” view is the one taught, erroneously, in high school and college. But she’s wrong in thinking that Lewontin’s paper supports that “social construct” view. In fact, the social construct view is largely wrong, with some exceptions centered on the outmoded view of “classical races”, but it appears to dominate anthropology and the social sciences. Anybody holding that view for either populations or groups of geographically contiguous populations needs to read the Coyne and Maroja paper.
4). Humans aren’t peas. According to Guevara, Mendel’s work on peas, as taught in school, buttresses scientific racism, too:
I, along with others, am concerned that this focus instills and reinforces a false pre-Lewontin view that humans, like Mendel’s peas, come in discrete types. In reality, early studies of peas and other inbred, domesticated species have little relevance for human genetics.
Indeed, it is of little relevance to human genetics, but I’m not aware of any teacher who describes Mendel’s work—which served to show how genes sort themselves out during reproduction—and uses it to conclude, “See, human races are as distinct as round and wrinkled peas.”
In the end, both races and populations of humans show genetic and evolved morphological differences—less than we thought, say, a hundred years ago—but differences that are still significant in useful ways. To say that races or populations are purely social constructs is simply wrong, and to use Lewontin’s paper to reinforce that conclusion is doubly wrong.
Now reader Lou Jost has argued that Lewontin couldn’t really mathematically partition genetic variation the way he did because Lewontin used the wrong method. Regardless, it’s clear that there is more genetic variation at a given locus within a population than between populations or the groups of populations once deemed “races”. But in the end there is a tremendous amount of information of biological and evolutionary significance to be gained by adding up the small genetic differences we see between human populations.
To end, here’s a map of genetic variation among populations in Europe, showing how the genetic variation (grouped by principal components analysis) lines up nicely with the geographic variation in populations. That’s because genetic differences evolved between semi-isolated groups of people, and that is why we can tell with considerable accuracy where our ancestors came from
Paper: Gilbert et al. 2022
Geography (populations sampled are in black)
Genetics (grouping of individuals using two axes of a principal components analysis. Look how well the geography (identified by color above) matches the genetics!
Mmm hmm – this is excellent, I wish YouTube channels would write like this. Great to review all this.
There wasn’t much sign of deviation from scientific epistemology in Guevara’s piece until “… the social construction of race …”
social construction is, in brief, a gnostic temptation. It asserts there is no objective reality, because reality itself is a product of thought. Critical constructivism, in modern critical thought (Joe Kincheloe).
This statement – this spell – immediately shows the subjective forest of Hegel’s Leftism for the objective trees of modern scientific epistemology (gel electrophoresis, genome wide arrays, etc.). No scientific degree program teaches this, and rightly so – but it leaves a missing plate of armor to an inverted epistemology.
I guess I’ll conclude with this quote which I’ve put up a lot (emphasis added) :
“[..] and so the dialectic continues.”
-Delgado and Stefancic
Critical Race Theory – An Introduction, p.66, 3rd Ed., 2017
Also of course :
Science, Politics, and Gnosticism
Eric Voegelin
1968, 1997
Regenery Press, Chicago;
Washington D.C.
“social construction is, in brief, a gnostic temptation. It asserts there is no objective reality, because reality itself is a product of thought.”
I believe the point has been made before in these forums that this is not how social scientists use the concept of “social construction.” No social scientist that I am aware of asserts that there is no “objective reality.” The point of social (or “cultural”) construction is that statements/beliefs/claims about objective reality – not objective reality itself – have to be discovered (a social process), tested (a social process), accepted by the [scientific] community, disseminated, taught, and learned. None of the knowledge expressed in those statements/beliefs is given to us at birth, fully formed and valid. Indeed, in this season Nobel Prizes are given out for scientific discoveries. If those discoveries were not the product of social processes and social construction, they would not be discoveries. My body automatically recoils from burning heat. That is hard-wired. But what “heat” is was a problem for science for centuries; only a social process of discovery, testing, etc. offered an account of heat that met scientific standards.
As for “race,” there are two semantic realms that we access by the term “race”, one I’ll call the popular discourse, and the other a scientific discourse. Considering that the parsing of “race” in the popular discourse has changed over the past 150 years, and differs in other countries, it seems reasonable to consider that, in the social construction of “race” over that period, a lot of cultural stuff gets packed into popular beliefs about “race.” Prof. Coyne mentions that biologists prefer to refer to populations rather than races, a nod to the fact that the scientific concept is quite different from the popular concept.
What’s the point of all of this? It is that history has shown that the processes of social construction leave even science vulnerable to bias and error, and good scientists are alert to those risks as they conduct science. I’ll leave lecture mode, but not without pointing out again that “social construction” is not about objective reality, but rather about our beliefs, assumptions, and claims about objective reality.
I thought I noted before that anthropological literature uses this term. Science is not anthropology.
Social construction of science would mean that a different set of “social” inputs would produce different, alternate discoveries. Nullius in verba means “on no one’s word”. That does not mean on everyone’s word, as gnostic thought sees things.
Social construct is a Deepity a mile wide and an inch deep to elicit a defense of dialectical gnostic thought, which is how gnostic cults work – by dialectical manipulation.
Sorry but I fail to see the relevance of your comment to the issue.
“Science is not anthropology.”
But science is a social and cultural system that anthropologists presumably can study (I am a cardiologist, and in addition to being a physician I have a PhD in physiology, so I know little about anthropology)
“Social construction of science would mean that a different set of “social” inputs would produce different, alternate discoveries”
And over the course of history that is exactly what has happened.
The German expresses it best :
Wissenshaftlicher sozializmus.
Capitalization of the “H” on “History” is a nice touch as well.
It seems to me that the above conclusion could easily be adopted by people who want to minimize or eliminate DEI initiatives. No more statements about race, no more data about race. From now on we’re a colorblind society paying attention only to merit or specific situations of adversity on a case by case basis.
If that were to happen I wonder if Guevera would change her tune.
I think the question all these targeted journals need to ask is something like :
Where is the citation of social construct?
Or
What theories produced the social construct argued in this piece?
I think all roads will lead to Hegel or Rousseau, and a certain author under “M”, with some distraction along the way like in anthropological literature.
It’s funny how social construct gets a pass every time.
Would she agree that Rachel Dolezal is African-American because she says she is?
What about the Pretendian phenomenon? People who claim native ancestry when they actually have none?
Thanks for mentioning my objection to Lewontin’s method of partitioning what he calls “diversity”. Any reader who would like more info can find it here:
https://www.researchgate.net/publication/24036324_GST_and_its_relatives_do_not_measure_differentiation#fullTextFileContent
and my other articles cited in that article.
Thank you. And Dr PCC(e) for the refs and critique.
Comment by Greg Mayer
Lou– You may know of the theme issue in the Phil Trans Roy Soc (vol. 377, 2022) entitled “Celebrating 50 years since Lewontin’s apportionment of human diversity”. I have a few of the papers in it, but none that I have mention your 2008 paper. Do you know if your criticism has been substantially addressed?
GCM
I didn’t know about that special issue. Thanks for pointing it out. No, in my opinion there has been no serious criticism of my partitioning theorem, and there really can’t be, since it is just math. In ecology it is widely accepted as definitive now.
There is however one valid response: it may be true that the actual diversity within groups is so low that the bad method used by Lewontin is not too far wrong. The right and wrong methods differ most when diversity is high.
I’ve now read the background article by John Novembre. It shows no signs of mathematical critical thinking. How could they not notice that the method would give dramatically wrong answers when “diversity” is high? It is as if they never bothered to “test-drive” the theory. The theory gave them the answer that they wanted, and I suspect this made them less skeptical than they should have been.
Thanks for returning to this topic. The Gilbert paper is amazing.
I think Guevara misunderstands how geneticists (and everyone else!) relate geography, ethnicity, and genetic variation. She imagines that a researcher looks at a person, classifies him to a racial group from a particular place, then looks to see how many genetic markers (out of millions of SNPs) distinguish that racial group. She thinks researchers (and teachers) in genetics are *trying* to make people assigned to different races (whatever that means) seem different because a person’s assigned race predicts genetics, and this is racist.
The important error is the one you emphasized: self-identified ethnicity or race. In studies like that Tang (2005) paper based on older methods (microsatellites), the researcher doesn’t know the ethnicity of the person, and has only the DNA variation. But the genetic variants in each person can almost perfectly predict how that person will self-identify with an ethnicity or race. The causation works the other way: genetics predicts self-identified race, and doesn’t have to involve geography (but the geography is important for explaining how the genetic pattern could evolve in the first place).
This is harmful to the world view in which race or ethnicity is purely socially constructed: if so then how could genetics predict self-ID? But it doesn’t make the genetic data or its presentation in a genetics course racist. And of course the ability to predict a person’s self-identified ethnicity from her genetic variants says nothing about the worth or moral qualities of that individual.
On what data does she make her claim that decades old out of date genetics is being taught in most schools? She obviously doesn’t know what many/most of us teach. My university human genetics course for non majors addresses the previous concepts of race, and follows the current understanding of population differences. Thanks to our host, I have great resources and summaries that guide our discussions. And the students understand.
My impression is that she is projecting. After all, it is she who is uncritically accepting “decades old out of date genetics”.
Yes, indeed.
A nit picky question:
When I read about differences in genes, Are writers referring to those parts of the genome that are protein encoding?
Or are they referring to nucleotides in general?
I don’t have a background in biology or genetics, so learning bit by bit.
Nucleotides in general, including third positions, introns, and dead genes (as far as I know). Some of this is based on the complete genome of people (I think the Europe data above is).
thank you….
A fantastic, almost samurai own of yet another wokester.
Kudos Samurai Coyne.
Very well reasoned and backs up the way I see genetics. I’m no expert but very much an armchair tourist for the past decade.
Their “anti-racism” logic, which is denialist, as David Reich said a few years ago, and buttressed by his latest study out a month or so ago, is going to crash on the rocks soon.
But it won’t stop ’em. Because their motivated less by truth than other things….
Theirs is a modern jihad of the soul rather than a search for truth.
(Reminds me of that Augusto Fuentez? (I forget his exact name right now) fool at Princeton).
D.A.
NYC
Agustín Fuentes: Race, Monogamy, and Other Lies They Told You. 2nd edit., University of California Press, 2022
Her argument seems to be something akin to this:
Homophobia exists. Therefore, homosexuality doesn’t.
Am I misunderstanding her?
As usual, I have nothing to add to the sum of human knowledge, but I do want to thank our host for yet another magisterial science post (and also to thank those making the erudite comments above). I learn something every time I read one of PCC(E)’s articles, but especially the science ones.
It’s kinda amazing that her argument turns almost entirely on Lewontin’s (classic) 1972 paper. A lot has happened since then, and it’s not hard to find research articles that report in detail on human migration patterns based on genetics. I can’t help but wonder if the author is really up to date in her reading.
“Human migration patterns based on genetics.”
Yes! Or otoh how *little* migration has happened relative to the size of the genetic differences that evolved among for example European populations. That last figure in Jerry’s post (the principal components phase space, from Gilbert et al. 2022) is just extraordinary. The correspondence between the spatial distribution of the samples on the map and the clustering of individual genomes in the phase space is surprisingly good. I’m trying to digest the meaning of that ~empty space in the center of the principal components graphic.
Me too.
It seems to suggest a direct gene-flow connection–by sea– between Italy and Spain, whereas gene flow between greater Greece and say France had to move overland through eastern Europe, mixing aling the way. Maybe.
Kudos for another great science post.
“…genetic differences among people say nothing about moral or legal equality. THAT is the lesson that needs to be imparted, not the falsity that there are no genetic differences among groups.”
Precisely.
Guevara doesn’t even understand Mendel’s peas. Like all living organisms, most pea plant characteristics are continuous, not discrete. Mendel was a smart fellow and understood that science is the art of the soluble (as Peter Medawar was to say a century later). He realised that he was more likely to understand inheritance by looking at simple discrete traits, rather than the complex traits that other people were investing (like speed of racing horses). So, he spent years doing preliminary work looking for suitable discrete traits in plants that he and his monk assistants could easily recognise in his breeding experiments. He didn’t find many, which is why some are famously a bit odd, like wrinkled seeds. And his strategy worked – he found universal laws.
The point is that humans ARE like peas. In both cases, the vast majority of their characteristics are continuous, controlled in a complex way by many interacting genes and the environment. But in both cases you can also find simple, discrete characteristics if you look hard enough.
This is a particularly penetrating issue (and comments) of the invaluable WEIT. And Coyne and Maroja were particularly astute in quoting Ernst Mayr on the fallacy (and the danger) of conflating equality with uniformity, as so commonly heard from the pop-Left. Impressive how Mayr’s observation—in 1963, long before today’s wokery, and before the postmodernism from which it sprang—captures precisely what is wrong with the various incarnations of the pop-Left. One may wonder whether Mayr’s youth in Weimar Germany, rather than his ornithology, helped equip him to summarize this insight so well.
Hugely interesting as a non-biologist.
I read the paper about AI recognition of race using chest X-rays. It appears that the traits used by the deep learning model are not yet known. The authors of that paper also state “there is more genetic variation within races than between races, and that race is more a social construct than a biological construct”. They state that “demographic factors that are strongly associated with disease outcomes (eg, age, sex, and racial identity), are also strongly associated with features of medical images and might induce bias in model results”. I may be reading it wrong (I’m just an engineer), but it seems that on one hand they’re saying that there are differences in disease outcomes based on race, but on the other hand they are concerned that because AI can accurately detect race that there will be negative outcomes from bias. If the doctor can see from a chest x-ray that the patient is of a race that experiences more negative outcomes from a certain disease, would that not allow the doctor to engage with a more aggressive treatment plan to accommodate for that?
I get the feeling that the writers see the problem as racial bias – White doctors don’t provide the same level of care to Black patients as they do White patience, so they think that by hiding the race it will produce race-neutral outcomes. My thought is, the more information the better; let’s tailor the treatment to the best possible benefit of each individual.
Maybe someone can explain why there is risk in an AI discovering racial identity when humans cannot?
Fascinating. How can they not know which criteria are being applied by the bot? How was the AI-ling trained?
What could it be? Some system of, I don’t know, ratios among anatomical measurements?
Engineers are uniquely trained to avoid b*******. Kudos.
Excellent. Thank you for revisiting this topic.
Another attempt to export the widespread moral commitment from the social sciences into biology. Think this kind of articles/papers may undermine the trust in science because the claims made are not very intuitive.
I really appreciate this summary. I’m teaching an introductory biology course for non-science majors with evolution as the organizing theme. I use WEIT as the text for the first half of the course, and then other readings for the second half of the course, which deals with applications of evolution to everyday life. I deal with the questions this post brings up, and it will help a lot (possibly as a supplementary reading for the students).
Here is another graph, conservatively based on genetic distance calculated from a huge number of SNPS between representative individuals of human populations sampled in the 1000 genome project, which gives imperfect geographic clusters that one must not call races (I agree one shouldn’t, as one should not suggests these groups are different in some cladistically relevant sense), but that have some correspondence to the classic “races”.
https://commons.wikimedia.org/wiki/File:Genetic_similarities_between_51_worldwide_human_populations_%28Euclidean_genetic_distance_using_289,160_SNPs%29.png
Regarding the variance question, Lewontin’s correct but misleading claim always gets misinterpreted as a non-significant ANOVA (statistic test of differences of a measurement between groups) by people who think that ANOVA tests whether the variance between groups is higher than variance within groups. It doesn’t, it tests whether the variance between the groups is above chance from what would be predicted solely from the interindividual variance (if there were in fact no systematic differences between the groups). The ANOVA can have a significant positive result even if the part of the variance that varies systematically between the groups is much smaller than the interindividual variation within the groups.
Lou, I will read yours, but fear it may be above my pay grade.
I’m happy to help you through it if you’d like. I can also invent an example that gives the same 85% figure as Lewontin’s, even if my populations share no alleles at all (they could even be different species). A reductio ad absurdum of Lewontin’s reasoning.
I shared your article with Elliott Sober and he sent me the following comment:
“Coyne is wrong about Edwards’s critique of Lewontin. I discuss this briefly
in ch 7 of my new book: https://www.cambridge.org/core/books/philosophy-of-evolutionary-theory/8D3BB3BDD978A94079F41F245D51D262. L was talking about diversity among and within *populations*. Edwards talked about a different question: can you make strong inferences about which of several (pre-defined) groups an *individual person* belongs to.”
Here’s the quotation from Sober’s book:
“Edwards (2003) accuses Lewontin (1972) of reasoning fallaciously – of taking his finding about within- and between-race diversity to entail that it is close to impossible to infer which population or populations an individual’s ancestors came from, given enough data about the individual and the populations. Lewontin did not reason in this way; he was not trying to solve the second inference problem. For further discussion, see Feldman and Lewontin (2008).”
The Feldman and Lewontin paper cited can be accessed here: https://www.researchgate.net/publication/288260752_Race_Ancestry_and_Medicine.
Neil Risch (“Genetic Structure, Self-Identified Race/Ethnicity, and Confounding in Case-Control Association Studies (NCBI/PMC)”) found that race and genetic ancestry are ‘only’ 99.86% the same thing. Tony Frudakis (“The Inconvenient Science of Racial DNA Profiling” in Wired) later found the linkage to be 100.0% (with a smaller sample).
The phrase ‘race is a social construct’ really means ‘race is only a social construct’. This has long (200+ years) been known not to be true. In our era, geneticists and other scientists have shown that race has a biological basis. This list of geneticists and other scientists is long and includes Collins, Frudakis, Hsu, Kahn, Reich, Risch, etc. Race is like sex and age, both a biological and social construct. Of course, this was known 200+ years ago (see the work of Blumenbach). For many years now, anthropologists have been able (with reasonable accuracy) to determine the race, sex, and age (at death). It sad but true, that our understanding of our own species has declined (in some respects), since then. Why?
There is actually a funny version of this. Quote
“Forensic anthropology and the concept of race: if races don’t exist, why are forensic anthropologists so good at identifying them?”
It turns out that we now have AI on this topic. AI has been found to determine race with considerable accuracy from X-rays. If race doesn’t have a biological basis, how exactly does this work? See “Reading Race: AI Recognizes Patient’s Racial Identity In Medical Images” in arXiv.
There is a pretty basic approach to races using just Darwin’s arguments and the subtitle of The Origin of Species:
“The Preservation of Favoured Races in the Struggle for Life”
Given enough time any sub-population of a species will adapt to local conditions (or diverge by genetic drift or founder effect) as long as the sub-populations remain repoductivly isolated.
This idea lies at the heart of evolution, it’s also one of the reasons why new species emerge in the first place. You can’t get race out of biology without denying evolution. The claim that there are no discernable sub-populations/variants/races in humans is like saying evolution has stopped for humans.
You should probably state, in conclusion, that race may not be just a social construct, but racism certainly is.
Fascinating and definitive rebuttal of an article that is either incompetent or of bad faith. I very much appreciate the clarity and succinctness of your writings.
Regarding the question of races =/= populations, it occurs to me that the antiquated sense of “race” hinges not just on observable physical or physiological characteristics. Rather, it was established over time by the combination of culture and these characteristics and reinforced by the relatively low geographical mobility of peoples a few hundred or a thousand years ago.
That is, people who looked alike due to their adaptation to a region would be expected to behave similarly not necessarily as a result of their characteristics, per se, but due to the shared cultural environment.
Or, as is likely, am I totally off here?
Dick Lewontin was my oost-doctoral advisor. Letting your politics interfere with your science isn’t only something the right does. Lewontin was very articulate and a brilliant teacher, but he didn’t hesitate to put down people he disagreed with. I once screwed up the courage to call him out for saying that in some cases the political issue was so important that it was ok to bend the science. He asked where did I write that? I replied that he’d said it in a seminar. He said, well, it’s ok if I didn’t write it. But he thought it and it influenced his science.
The well-established variation in average IQ between populations due at least in part to genetic factors is arguably the topic that upsets the most people. Russell T. Warne’s 2020 book “In the Know: Debunking 35 Myths about Human Intelligence” is an excellent summary of the current state of knowledge.