Razib Khan is a geneticist, science blogger, and podcaster who writes about genetics, history, evolution, and culture at Unsupervised Learnings (razib.substack.com). In this wide-ranging conversation, he discusses the genetics of intelligence, endogamy in India and among Ashkenazi Jews, the social structure of science, the trajectory of human history, and the future of gene editing and brain-computer interfaces.
The Dysgenics of Intelligence
In the developed world, there is currently negative selection on genes for educational attainment: people who carry those genes tend to delay childbearing to invest in education, and many end up having fewer children or none at all.
This is not a long-term equilibrium—reproductive differentials have not stabilized—but it could persist for centuries.
The math is straightforward: longer generation times and lower fertility among the highly educated mean their genetic representation declines each generation.
Razib is skeptical that natural selection will correct this, because there is little reproductive advantage to very high intelligence in modern society, while lower cognitive ability correlates with higher fertility.
Endogamy in India
Genetic studies (including work by David Reich in Andhra Pradesh) show that Indian jāti (sub-caste) groups have maintained extraordinarily high endogamy rates—around 99.5% per generation—for thousands of years.
This is far more extreme than any other known population; for comparison, Black American endogamy is ~85%, and it was ~95% in earlier decades.
No one fully explains how this is maintained. The caste system is an obvious candidate, but sexual exploitation of lower-caste women by upper-caste men has also occurred, so the mechanism is not purely normative.
Some exceptions exist: the Nair group in Kerala historically had consensual relationships with Nambudiri Brahmin men, producing offspring fathered by Brahmins but raised in Nair households.
India is unusual in that an entire society is stratified this way. The only rough parallels are Ashkenazi Jews and the Roma (who are themselves partly of Indian origin).
Why So Many Indian (Brahmin) Tech CEOs
Many prominent Indian-American tech CEOs (e.g., Satya Nadella at Microsoft, Sundar Pichai at Google) are South Indian Brahmins, specifically Telugu Brahmins.
This reflects a long-standing pattern: South Indian Brahmins migrated to North Indian cities during the colonial period to work in the Indian civil service, while North Indian Brahmins tended to stay local.
South Indian Brahmins have been disproportionately represented in engineering and software for generations—a pipeline that carried over into the Indian-American professional class.
Razib contrasts this with North Indian elites (e.g., UP Brahmins, Rajputs), who tend to prefer being “big fish in small ponds”—dominating local power structures rather than taking risks in unfamiliar environments.
He cites Subrahmanyan Chandrasekhar (the astrophysicist) as an example: a Tamil Brahmin born in Lahore who faced discrimination as a South Indian in the North, then later faced prejudice from figures like Arthur Eddington in Britain.
Male Variance in Intelligence
Men show greater variance in cognitive ability: more geniuses and more individuals at the low end.
One mechanism: males have only one X chromosome, so deleterious X-linked mutations are not masked by a second copy. Women randomly inactivate one X chromosome per cell, providing a buffer.
Males are the heterogametic sex (XY), which introduces extra developmental steps (e.g., the testosterone burst in the first trimester that masculinizes the fetus). More developmental steps mean more opportunities for things to go wrong.
Testosterone also suppresses immune function, contributing to higher male vulnerability.
More males are conceived (~104–105 males per 100 females born), but male fetuses are overrepresented in miscarriages, suggesting higher baseline developmental instability.
At the high end, there is no clear reproductive advantage to extreme intelligence; it may simply be a byproduct of greater male developmental instability—a “freak” outcome of a system that is less tightly canalized.
Brain Size Decline Since the Pleistocene
Human brains have shrunk by roughly 10% since the Pleistocene, alongside a general reduction in body size.
Agriculture likely contributed: reduced nutritional quality (even if more reliable) and smaller body size meant smaller brains were more metabolically optimal.
There is evidence of negative selection for small body size in Southern Europe and Asia.
In South Asia, populations like Bengalis are genetically shorter, partly due to East Asian ancestry and partly due to local selection pressures (e.g., cholera resistance in flood-prone Bengal).
The Self-Domestication Hypothesis
The hypothesis proposes that humans underwent a domestication process similar to that seen in dogs and foxes—selecting for traits like reduced aggression, smaller jaws, and neoteny.
Razib finds it plausible but unproven in humans. The genomic evidence has not clearly materialized after a generation of research.
Classic domestication markers in animals (floppy ears, piebald coat patterns) are not obviously present in humans, so the hypothesis remains speculative.
The “Church of Science”
Razib drew a metaphor: if parishioners attend church to bask in the community rather than worship God, the church won’t last. Similarly, if scientists are in the profession for collegiality, comfort, or salary rather than truth, science loses its purpose.
He argues science is genuinely hyper-Pareto distributed: a tiny fraction of researchers produce the vast majority of important work, unlike fields like pediatrics where most practitioners contribute meaningfully.
Much of the talk about “community,” “support,” “inclusion,” and “equity” in science is, in his view, performative—it doesn’t change the fundamental winnowing nature of the profession.
Science is also heavily class-biased: it skews toward children of the professional managerial class, and especially children of academics, who inherit tacit knowledge about how to succeed.
He gives examples: a professor’s child publishing a peer-reviewed paper in high school under parental guidance; a progressive academic on Twitter whose uncle was a Nobel laureur and who worked in that uncle’s lab as a teenager—then signaling virtue online while being demanding and difficult in the lab.
Publication inflation is massive: postdocs today often have multiple papers before finishing graduate school, whereas a decade ago, many had none until their third or fourth year. Much of this reflects metric distortion rather than genuine productivity gains.
The Long View of History
Razib sees history as mostly gradual and exponential, though we retrospectively rarefy it into “step functions” (agriculture, industrialization, etc.).
The Industrial Revolution and the Axial Age were both more gradual than commonly portrayed.
However, he thinks we may currently be at a genuine step: information technology is advancing at an exponential rate (à la Kurzweil), and the changes are radical enough to constitute a discontinuity.
His own children are already confused by older phone form factors (flip phones, rotary phones), illustrating how quickly the technological environment shifts.
He references the idea that this could be “the last century of humans as we recognize them”—or a century of regression.
Civilizations have collapsed before (Chinese dynastic cycles show progressively shorter collapse periods, suggesting increasing institutional resilience), but now we are global, so a collapse would be far more consequential.
Legacy systems are already a problem: much military software runs on COBOL, and there are almost no COBOL programmers left—analogous to Babylonians using Sumerian liturgy centuries after the language died.
The Future: Biology vs. Virtuality
Razib thinks gene editing of humans is more feasible in the near term than full brain-uploading or virtual existence.
In the next 10 years, gene editing will focus on Mendelian diseases (cystic fibrosis, sickle cell, ALS)—conditions with large-effect loci where the risk-benefit calculation favors intervention.
In ~20 years, parents will likely begin editing polygenic traits in offspring, including intelligence, though this is harder because intelligence involves thousands of SNPs.
One strategy: rather than trying to add “gain of function” intelligence mutations, focus on fixing deleterious de novo mutations in the child’s genome by comparing both parents’ genomes to a reference and correcting copy errors—a finite, tractable problem.
Brain-computer interfaces represent a potentially larger leap—more like a qualitative transformation than incremental Smithian growth—but they are untested and will likely have high failure rates initially.
He speculates that the first successful person to fully interface with a computer could have an enormous advantage—like Columbus arriving in the New World—potentially becoming a “god” of that new domain.
Cultural Evolution as a Field
When asked where historical expertise could create a new niche, Razib points to cultural evolution—applying evolutionary principles to historical and cultural change, as done by Peter Turchin and Joe Henrich.
This field is nascent and growing, though it faces resistance from traditional historians who see it as turf infringement.
Writing About Technical Topics for a Broad Audience
The key is to anchor technical methods in domains people already care about.
Example: instead of writing abstractly about machine learning, write about using machine learning to classify personality—personality is interesting, machine learning is the method.
People are interested in topics like Ashkenazi Jewish genetics, the genetic architecture of skin color, astronaut genetics—find the compelling domain, then apply the technical toolkit.
