Dwarkesh PodcastRazib Khan - Genomics, Intelligence, and The Church of Science
CHAPTERS
- 0:00 – 0:47
Cold open: “The Church of Science” and why institutions dissolve without truth-seeking
Razib frames science as a quasi-religious institution where “God is the truth,” arguing that labs and research communities only survive if truth-seeking is the core motive. He also drops a quick travel aside about England feeling like “the United States, but whiter,” setting the conversational tone.
- •Science as an institution held together by commitment to truth
- •Skepticism of participation motivated by status or community vibes rather than inquiry
- •Why passion for research is a necessary stabilizer
- •Quick travel/cultural observation about England
- 0:47 – 2:15
Razib’s background: genetics as a tool for history, demographics, and self-directed analysis
Razib explains how his long-running interests in history and demographics converged with modern genomics. He describes his hands-on style—replicating analyses himself and using population genetics methods to answer historical questions.
- •Scientific training plus long-standing interest in history/demography
- •Genomics’ rise over the last two decades as a general-purpose inference tool
- •Doing independent replications (e.g., pairwise genetic distances) for blogging
- •How his work straddles biomedical and historical inference
- 2:15 – 5:04
Dysgenics and the long-run equilibrium of intelligence without gene editing
Dwarkesh asks what happens to intelligence over centuries if current fertility patterns persist. Razib argues we’re not at equilibrium and that present-day selection in developed countries appears negative on educational attainment due to delayed and reduced reproduction among the highly educated.
- •Why “equilibrium” may take centuries (generation time + fertility differentials)
- •Evidence for negative selection on educational attainment polygenic scores
- •Modern incentives: thriving vs reproducing are decoupled
- •Fertility constraints and social dynamics (partner search, delay, childlessness)
- •Idiocracy as a pop-culture shorthand for selection differences
- 5:04 – 9:14
India’s extreme endogamy: how jati boundaries persist for millennia
They dig into genetic evidence suggesting extraordinarily high endogamy rates among Indian jatis across thousands of years, even within the same villages. Razib notes the phenomenon is unusually strong compared to other societies and admits there’s no fully satisfying explanation.
- •Reich’s findings: endogamy rates ~99%+ per generation in some regions
- •Comparison to U.S. racial endogamy and resulting admixture levels
- •Why such strong barriers are surprising given typical human mating behavior
- •Speculative mechanisms (taboos, punishment, infanticide) but no consensus
- •The scale of stratification as the truly unusual feature
- 9:14 – 9:37
Endogamy elsewhere: Ashkenazi Jews, Roma, and why whole-society stratification is rare
Razib identifies a few global analogs to Indian-style endogamy (notably Ashkenazi Jews and Roma), while emphasizing how unusual it is for an entire large society to be organized that way. The discussion sets up how endogamy can amplify genetic distinctiveness and group-level traits over time.
- •Ashkenazi Jews and Roma as notable endogamous comparisons
- •Why “a whole society” built on endogamy is historically odd
- •How endogamy can preserve distinctiveness across centuries
- •Transition to thinking about other potential high-achievement minority dynamics
- 9:37 – 15:08
Could there be a ‘new Ashkenazi’ group? China, middleman minorities, and assimilation dynamics
Dwarkesh asks whether other endogamous or bottlenecked groups might surge in achievement once prosperous and “liberated.” Razib discusses the historical niche of Ashkenazi Jews and explores Chinese regional patterns (Fujianese success, Hakka distinctiveness) alongside the Han tradition of assimilation.
- •Ashkenazi Jews as a middleman minority in Central/Eastern Europe
- •Fujianese historical examination success and state responses (quotas)
- •Hakka as a partial analogue; limits of ideological endogamy comparisons
- •Han identity as highly assimilative (paternal lineage/clan logic)
- •North–south Chinese genetic patterns (e.g., minor West Eurasian signals in the north)
- 15:08 – 20:37
Why so many Indian (often Brahmin) tech CEOs: literacy elites, migration pipelines, and regional dynamism
They explore why Indian Americans are overrepresented among tech leaders, and why South Indian Brahmins in particular show up frequently in elite educational and professional pipelines. Razib contrasts South Indian Brahmin mobility with North Indian elite preferences for local dominance and lower migration dynamism.
- •Brahmins as historical “symbolic manipulators” and literate specialists
- •South Indian Brahmins’ colonial-era migration into civil service roles
- •Engineering/software as the modern pipeline for Indian American leadership
- •North Indian elites (e.g., UP/Bihar) as ‘big fish in small ponds’ dynamics
- •Historical example: Chandrasekhar’s trajectory and experiences of prejudice
- 20:37 – 25:31
How Razib learns the world: minimal travel, high context, and the geography ‘party trick’ (plus jati guess)
Razib describes doing most of his work from the U.S. and how deep background knowledge can make present-day behaviors feel legible rather than random. He shares anecdotes about impressing scientists by guessing their origins via geography cues, culminating in accurately guessing Dwarkesh’s jati mix.
- •Travel is optional; analysis can be done remotely with strong priors and data
- •How historical/geographic antecedents explain modern cultural patterns
- •Conference anecdotes: guessing Wuhan/Shandong from minimal hints
- •Americans’ low geography baseline makes knowledge seem like magic
- •Jati guessing as pattern recognition (Patel + Bania)
- 25:31 – 30:57
Greater male variance: X chromosome exposure, developmental instability, and skewed tails
Dwarkesh asks why men appear overrepresented among both geniuses and low performers. Razib outlines biological mechanisms that can increase variance in males—especially X-linked vulnerability, additional developmental steps in masculinization, and higher fragility during gestation.
- •Single X chromosome in males exposes deleterious alleles (less masking)
- •X-inactivation mosaicism in females as a buffer
- •Extra developmental steps (testosterone burst, masculinization) add failure points
- •Male overrepresentation in miscarriages and early-life culling
- •High-end outliers as ‘going off target’—variance without direct adaptive need
- 30:57 – 35:21
Why our brains shrank after the Ice Age: body size, agriculture, nutrition, and (maybe) self-domestication
They discuss evidence that human brain size decreased alongside overall body gracilization, especially with agriculture and changing diets. Razib considers the self-domestication hypothesis as plausible in spirit but argues genomic evidence hasn’t strongly validated it in humans.
- •Post-Ice Age warming and smaller bodies correlating with smaller brains
- •Agriculture’s role in gracilization and potentially lower nutrition quality
- •Selection for smaller body size in parts of Europe and Asia
- •Bengali height as an example of strong genetic component vs environment
- •Self-domestication: compelling cross-mammal patterns, but weak/unclear human genomic confirmation
- 35:21 – 45:07
“The Church of Science” unpacked: collegiality, incentives, class inheritance, and metric distortion
Razib explains his tweet: science fails when community comfort replaces truth-seeking. He argues academia is ‘up or out,’ deeply Pareto-distributed, class-biased, and shaped by tacit knowledge passed through academic families—while modern incentive metrics increasingly get gamed.
- •Collegiality vs truth: why prioritizing comfort can hollow out science
- •Extreme winner-take-most dynamics in scientific credit and careers
- •High class bias and tacit know-how advantages for academic families
- •Virtue signaling vs real lab behavior; incentives on social media
- •Goodhart’s law in publishing: metrics distort behavior and output
- 45:07 – 53:18
Long history lens: gradualism vs step changes, fragile complexity, and civilizational regression risks
Dwarkesh asks whether history moves via step functions or gradual exponentials. Razib argues we often reify gradual changes into “revolutions,” but that today might truly be a step due to exponential information tech—paired with fragility from accumulated complexity (software, supply chains, expertise loss).
- •Industrial Revolution/Axial Age as more gradual than popular narratives suggest
- •Living through ‘miracles’ we normalize (phones, video calls, rapid iteration)
- •Meta-stability: primate instincts operating atop civilization-scale power
- •Chinese dynastic-cycle example: shrinking interregnums as improved institutions
- •Fragile complexity examples: COBOL legacy, expertise loss, fertilizer/nitrogen dependency
- 53:18 – 57:12
Future humans: gene editing vs brain-computer interfaces, high variance outcomes, and ‘new worlds’
They compare biological enhancement to brain-computer interfacing, with Razib expecting both but seeing interfaces as a bigger discontinuity because it’s unprecedented. He predicts experimentation will be risky and uneven, potentially creating dramatic first-mover advantages and new existential risks.
- •Gene editing as ‘efficiency gains’ using existing variation; interfaces as baseline-shifting innovation
- •Expectation of self-improvement through editing over the next century
- •High-risk early adoption of brain-computer interfaces (failures, mortality, unknowns)
- •Analogy to early Atlantic voyages: many losses before breakthroughs
- •Speculation on emergent threats (e.g., ‘Skynet’ as an uploaded human)
- 57:12 – 1:04:40
Near-term gene editing reality: Mendelian disease first, intelligence later; plus writing/creator meta and closing
Razib forecasts near-term breakthroughs in treating single-gene diseases in adults, with embryo editing arriving later and intelligence being harder due to polygenicity. They close with meta discussion about building a technical-writing audience (find what people care about, then apply the method), Substack economics, and final goodbyes.
- •Next ~10 years: adult therapies for large-effect disorders (sickle cell, cystic fibrosis, ALS)
- •Off-target risk tradeoffs become acceptable when baseline prognosis is dire
- •~20-year horizon for embryo editing; intelligence is hard due to thousands of loci
- •Alternative path: fix deleterious ‘copy errors’/mutational load before chasing gain-of-function
- •Advice for technical writers: hook interest via relevant domains (e.g., personality + ML); Substack reflections and wrap-up
