David Reich – Why the Bronze Age was an inflection point in human evolution
In a Nutshell
Ancient DNA analysis of ~16,000 Europeans/Middle Easterners over 18,000 years reveals rampant natural selection, with ~3,600 genomic positions (99% confidence) showing directional shifts, especially intensifying in the Bronze Age (~5,000-2,000 ya) across immune, metabolic, pigmentation, and polygenic traits like predictors of cognitive performance, years of schooling, and reduced obesity—far stronger than the Neolithic farming transition or recent centuries. This "evolutionary mismatch" from hunter-gatherer genetics clashing with high-density pastoral/urban life drove adaptations like lactase persistence, depigmentation, and ~2 standard deviation gains in executive function proxies, countering expectations of quiescent selection post-Out-of-Africa. A new model posits Neanderthals as culturally modern hybrids (~95% archaic + 5% early modern input ~300,000 ya), explaining shared Levallois tools, mtDNA/Y lineages, and wavefront expansions, simplifying convoluted admixture histories.
These notes were generated by AI and may contain inaccuracies.
Humans in this part of the world were wrenched into a way of living so different from their hunter-gatherer ancestors that the organism had to adapt very strongly. The degree of this wrenching process moving into the Bronze Age was qualitatively greater than the initial transition to growing plants, despite the cartoon picture emphasizing farming as the big transition. Genetic data shows the genome reacting much more strongly to events 5,000 years ago.
David Reich is a professor of ancient DNA at Harvard. He is a geneticist working on human history and how ancient people relate to each other and people living today. They did an interview two years ago that was one of the most popular.
People found the previous interview compelling because ancient DNA techniques reveal much unknown human history. New preprint is exciting. The dream 16-17 years ago when the ancient DNA field started was to learn about biology and how people's biology changed over time by tracking changes in DNA from ancient human remains. This dream has not been realized; the field succeeded in human history with surprising findings on migrations, mixture being common, sex-biased processes, and results unexpected from archaeology.
Success in history but not biology due to small sample sizes. One person's DNA provides information about history because it represents two parents, four grandparents, etc., effectively tens of thousands of ancestors, positioning the individual exquisitely relative to others. For genetic variants affecting skin pigmentation, lactase persistence, or behavioral traits, one person gives only one or two samples. High-resolution frequency changes over time require very large sample sizes, only available in the last few years.
Now with large numbers, they can track frequency changes. Frequency changes are interesting to use the experiment of nature in history over tens of thousands of years to understand biologically significant DNA. Environmental changes like shift to agriculture, living close to domesticated animals, moving from cold to warm or low to high altitude create pressure to adapt. Detect by systematic pushes in frequency of variants allowing high-altitude living or advantageous behaviors, extreme enough not due to chance. Hard to detect slight shifts without big sample sizes.
Seven years ago, Ali Akbari (postdoc then permanent staff scientist in Reich's lab) used data to learn biological change. Lab focuses on industrializing ancient human data production: inexpensive, high quality, large numbers. Mainstream view: natural selection quiescent over last several hundred thousand years. Evidence: comparing Europeans and East Asians (descended from common population 40,000-50,000 years ago out of Africa/Middle East), almost no mutations 100% different in frequency, as expected by genetic drift over 1,500-2,000 generations. No rocketing mutations expected under selection (e.g., for alcohol or milk digestion).
Ancestral population reached optimum; small selection to remove bad mutations, but not directional selection on newly arising mutations. They partitioned changes in ~10 million variable positions: 98% due to other factors like genetic drift, migrations, population structure; 2% directional selection/adaptation. Signals hard to detect as tiny fraction, but natural selection rampant.
Population admixture/replacement could be selection at group level, with huge changes in areas over time. But might be cultural (technology), not genetic; whole-genome shifts uninformative for selection. Best detection when populations isolated for few hundred years, like archipelago of little populations in Europe/Middle East (e.g., Britain or Hungary isolated between migrations). If mutation increases consistently across such periods, indicates selection. Example: 4,500 years ago Europe, huge changes from Yamnaya steppe pastoralists (40-80% ancestry north of Black/Caspian Sea), due to prior evolution, not necessarily selection.
Found 7,200 positions with 50% confidence under selection last 10,000 years (half real: ~3,600). At 25% cutoff, tens of thousands, many real; genome vibrating with selection. Almost every position correlated to another under selection. Analyzed ~100 traits from GWAS (immunity, autoimmunity, behavior, metabolism): 4-5 fold enrichment for immune traits, strong for metabolic traits (obesity, fat, Type 2 diabetes), almost none for behavioral/psychiatric.
Not that behavioral/cognitive traits not under selection; individual sites not especially among strong signals. Behavioral traits polygenic (many genes of weak effect), immune fewer strong-effect genes; lack power for weak signals. Immune most selected category, but behavioral also selected.
Previous: no 100% frequency differences Europeans/East Asians. Now: hundreds positions rocketing (1%+ selection rates, doubling dozens generations). Explained by: (1) accelerated selection in Europe/Middle East; immune/metabolic enrichments intensified last 5,000 years (Bronze Age onward) vs prior 5,000. Whole period increased vs before. Plausible due to huge shock: farming invented Middle East 11,000-12,000 years ago, exploded Europe after 8,500 ya; Bronze Age higher densities, living next to animals, disease exchange. Evolutionary mismatch: hunter-gatherer genetics in agricultural/pastoral/urban context.
AGES browser by Ali and co-author queries 10M positions/trajectories/selection evidence. Most signals constant selection, but handful show reversal/radical change 5,000-2,000 ya (Bronze Age/Iron Age): rapid growth, intensive technologies. TYK2 variant: major risk for severe tuberculosis (top infectious killer); rose to 9-10% 8,000-6,000 ya, then fell last 3,000 ya (both directional selection). Possibly protected before, then TB endemic 2,000-3,000 ya pushed opposite.
Selection intensifies Bronze Age across traits. Lactase persistence intensified: time of cattle for milk/wool secondary products. FADS1/2 (converts plant fatty acids to long-chain needed; vegetarian/meat adaptation, ancient even in archaics): strong selection 5,000-3,000 ya, identified 2015 by Ian Mathieson. ABO blood system: B to 10% at expense of A; A/B in human/gibbon ancestor, fluctuating. TYK2/MS risk variant increased pre-Bronze Age, reversed 2,000-3,000 ya (stronger North vs South Europe). Hemochromatosis (pathogenic iron buildup) reversed around period. Complex traits also intensified.
In some complex traits, there are periods of intensification of natural selection. For example, depigmentation: Europeans have gotten lighter skin over the last 10,000 years, visible in the data. The period of strongest depigmentation is between about 4,000 to 2,000 years ago, after which it much lessens. This period seems very impactful, where many processes became very powerful. It's surprising on first principles: one might expect the big change from starting to grow plants and farm animals in the Neolithic, beginning 10,000-12,000 years ago and spreading into Europe after 8,500 years ago. But intensification happens 5,000 years ago, 4,000 years ago. This observation marks it as an inflection point, indicating when humans in this part of the world were wrenched into a way of living so different from hunter-gatherer ancestors that the organism had to adapt very strongly. The wrenching into the Bronze Age may have been qualitatively greater than the initial transition to growing plants. The biological readout in the genome reacts much more strongly to events 5,000 years ago than to farming.
In 2014 work with Bhatia and colleagues on 20,000-30,000 African American genomes today (80% West African DNA, 20% European DNA), allele frequencies were not much different than expected from admixture. Over 200-300 years of intense environmental change—from chattel slavery to a new environment—there's no effect of natural selection. Yet the Bronze Age had a stronger effect, stronger than Africans migrated to the New World under slavery. This may be because 200-300 years is too short: in the Bhatia et al. paper, no places in DNA had significantly more or less than 80% West African ancestry, as expected if selection favored variants from Europeans or Africans. With only about five generations, even 2% selection per generation yields only 10% compounded effect, not detectable. The Bronze Age spans 3,000 years—like compound interest—enough time for strong effects. This period is very impactful in human history, visible in complex traits.
Pigmentation shows the strongest signal of selection for a complex trait: dozens or hundreds of genetic mutations affecting it, with strongest selection 2,000-4,000 years ago. For other traits too, strongest selection is 2,000-4,000 years ago. For genetic variants affecting measures of cognitive performance (e.g., performance on intelligence tests in white British people today), there's very strong natural selection for this combination, highly correlated with years of schooling or household wealth. These are strange traits for the past—no intelligence tests, no school, no wealth—but predictors today show strong systematic movement, about a standard deviation on modern variation scale. Using a 2,000-year sliding window on the data, selection strength maxes out in the Bronze Age, 5,000-2,000 years ago. Impact in last 2,000 years is almost nothing—no evidence of selection. Bias might expect strong selection in last 2,000 years (industrialization, greater need), but evidence shows very strong selection 2,000-4,000 years ago: two standard deviation strength vs. one.
The standard deviation measures how much the polygenic score for the trait moves over 10,000 years within a population held constant for ancestry. The dataset is heterogeneous (Southern/Northern Europeans, hunter-gatherers, farmers) with changing ancestry over time, but Ali Akbari's methodology corrects for that. It divides the dataset into an archipelago of little populations in space and time (e.g., Britain 4,000-3,500 years ago, Hungary, Italy 2,000-1,500 years ago), where ancestry is relatively stable short-term. It checks if genetic changes blow in the same direction across these pockets after correcting for population changes. The effect is huge: one standard deviation above median is 85th percentile; over 10,000 years, median shifts to 85th percentile. This is 2% of allele frequency change from selection, 98% from migration—stupendous impact, especially with migration's huge role.
Migration impact is huge: for cognitive performance predictors (intelligence test scores in white British today), European hunter-gatherers estimate three standard deviations below modern mean; farmers at mean (zero); steppe pastoralists lower. Huge fluctuations over time are migration—different set points—not proving selection. The test detects consistent natural selection direction beyond migration fluctuations across places/times.
The collective intelligence hypothesis posits selection against intelligence as society develops (more specialization, less need for broad understanding), ancients smarter, we've evolved down. Results point opposite: no selection last 2,000 years as society complicated, but early civilization increased selection on what predicts intelligence today. Surprising vs. hunter-gatherers needing vast info (food processing, shelters, fire—per Joseph Henrich's book) vs. modern narrow skills (e.g., setting up mics). Data power shows opposite of expectation: hunter-gatherers wouldn't predict high on this trait; complex societies selected for it despite collective brain.
Genetic predictor of intelligence or years of schooling (highly correlated, measured better) confuses: 2017 Iceland study shows 0.1 standard deviation decrease in genetic predictor of years of schooling over last 100 years (huge short-term effect), selection against it. Not intelligence, but correlated traits: strongly to women's age at first child (control for it, schooling signal vanishes); also body mass index, obesity, walking pace. Measures delaying gratification/planning? Years of schooling meaningless in past, but underlying predictors under strong selection.
Years of schooling predictor (from millions) correlates to age at women's first kid, obesity, walking pace, household wealth—surprising. Not pure intelligence/studiousness; some general trait like executive function or propensity to defer gratification, pushing disparate traits, advantageous/disadvantageous at different times.
Incredulous at signal for genetic propensity predicting more schooling in white British, tests done: checked variants' effect on years of schooling in Chinese people in China—huge statistical correlation (five-six standard deviations) to their trajectory in Europeans over 10,000 years, despite disconnected populations. As strong as European variants to European trajectory—not chance. Confirms real natural selection increasing changes manifesting as more years of schooling today. Same variants robustly predict more schooling in modern Europe and China—not GWAS artifact.
What changed in environments over last 18,000 years? Surprising intelligence or lack of schizophrenia (seemingly robustly good) not maxed before Bronze Age; population diversity huge (European hunter-gatherers three SD less predicted intelligence). Yet they existed where intelligence matters—why not strong prior selection like on body traits? Selection effective within hundreds/thousands years if adaptive.
From modern society's view intensely valuing IQ/schooling—unprecedented. Hebrew/Christian Bible (esp. Old Testament, when intelligence selection peaked) values strength, courage, religiosity—not intelligence. Homer/other religions: beauty, etc. Hyper-focus on smarts not common past value. IQ performance not directly what mattered (e.g., new-world food/shelter processing—Henrich notes modern underestimate difficulty). Intuitively valuable regardless.
Traits (obesity, years schooling, walking pace, IQ, household wealth) share genetic variants. Iceland 20th century: selection for many kids low investment vs. few kids high investment (defer kids, more wealth/resources per kid = lower fertility but better outcomes; or max kids less investment—fits plenty times like Iceland). Toggle like ecology: mammals (few, invested) vs. fish (many, most die). Moves back/forth by conditions.
For schizophrenia/bipolar, maybe spectrum advantageous contextually: subclinical anxious/imaginative/neurotic helpful in shamanistic/religious traditions valuing visions/creativity. Selection for different creativity/thinking types valuable differently.
Julian Jaynes' The Origin of Consciousness in the Breakdown of the Bicameral Mind: up until Homer, people schizophrenic—thought gods real people communicating. Ancient texts show this; even today, some religious communities value visions/supernatural communions.
Most mutations for schizophrenia and bipolar disease are disadvantageous, shown by low-frequency variants with small effects.
Another trait under selection is the trend away from body fat since the agricultural revolution. There is a reduction in genetic mutations increasing risk for obesity, body mass index (BMI), fat mass, waist-to-hip ratio, and type 2 diabetes risk. Clear selection of about one standard deviation (on modern variation scale) reducing these traits over the last 10,000 years in Europe and the Middle East.
Thrifty gene hypothesis: Hunter-gatherers moving to farming environments with plentiful food no longer need to build body fat to survive stress periods, as food stores are constant. Natural selection acts against body fat in agricultural environments with food plenty.
Europeans are genetically better protected against type 2 diabetes than populations like African Americans and Native Americans, possibly due to longer exposure to agriculture and stable food.
This contradicts the idea that hunter-gatherers had more stable diets due to variety, mobility, and not relying on single crops. Selection against fat storage suggests agricultural famines were more stable than hunter-gatherer conditions.
In traditional hunting societies, people gorge after hunts, build temporary fat, then go days without meat—a boom-and-bust cycle. Farming has famines (e.g., every 3-5 years), shown in farmer bones with stress markers, but timescales differ: fat from a hunt won't carry through a 3-year famine. Survival of periodic famines differs from short-term fat storage needs.
Traits like immune system (post-Bronze Age) mattered more for fitness than intelligence, under stronger selective pressure. This suggests humans could have been much smarter if selected for intelligence.
Relevant to AI: Training focuses solely on intelligence, without needing immune systems or avoiding schizophrenia. If intelligence wasn't dominant under selection over last 10,000-100,000 years, there's more room at the top for this trait.
There's more room for many traits: height could shift extremely, but with trade-offs sacrificing other traits. Natural selection could push means further.
Evolution since Out of Africa acts on alleles from initial group of ~10,000 people. Surprising that this pool had latent variation for cognitive profiles, disease resistance, height, etc.
Human populations contain tremendous variation for complex traits:
- Huge variation affects height, BMI.
- Setting all mutations to high-height makes someone extremely tall.
- For schizophrenia, pointing variants one way yields extreme risk or protection.
For polygenic complex traits (many mutations), all variation exists to shift to new adaptive set points in hundreds/thousands of years in new environments.
Unusual traits like lactase persistence (digest cow's milk) or sickle cell protection require rare single mutations, waiting dozens/hundreds of generations in small populations (~10,000).
In large populations (8 billion people, ~30 new mutations per generation/person = 240 billion new point mutations/generation), every possible mutation (across 3 billion genome bases) occurs ~100 times per generation—no mutation limit.
In small populations (10,000), wait dozens/hundreds of generations for key mutations.
By 3000 BC, population ~50 million; high gene flow makes even weak selection visible.
Unlikely Bronze Age change due to population size alone—1 million already suffices (every mutation occurs in few generations). Even pre-Bronze Age Europe/other regions hit this.
In small populations (1,000), mutations drift randomly (±1/1000 frequency/generation); selection < that is drowned by genetic drift.
0.1% selection coefficient weak; 1% strong, works in 1,000-10,000 populations.
Weak effects rising only in large populations (1/10,000-1/100,000) take 10,000-100,000 generations (hundreds of thousands-millions years)—not relevant timescale.
Strong coefficients (0.5%+) work regardless of population size. Dominant factor post-threshold: time span.
Mallick 2016 paper: No fixed differences between modern and archaic humans ~50,000 years ago. This period saw cognitive revolution, modernity, art.
Suggests no biological genetic switch; perhaps cultural change.
100,000-50,000 years ago: Quickening cultural pace—first extensive representational art, bead necklaces, wall drawings, rapid tool innovation.
No genome-wide evidence of key selective sweeps ancestral to all today (common ancestor <400,000-500,000 years ago). No sweeps shared by all humans in this period, unlike none between Europeans/East Asians.
Possible polygenic adaptation: many mutations shift direction without single high-frequency change.
This population ancestral to West Africans, most East Africans, all non-Africans. Some Africans (Khoisan, Central African rainforest hunter-gatherers) have ancestry from groups diverged ~200,000 years ago.
All groups capable of college, etc.—no key mutations lacking. Variance between modern humans latent in small group (~10,000, 50,000-100,000 years ago).
Human genetics: Population has clay for almost any trait; means shift with environment/selection.
New work: Last 18,000 years in this region, significant movement in >500 traits (~100 complex traits systematically).
Genetically equipped ~50,000 years ago (common ancestors had ingredients). Humans reached Americas (15,000 years ago), New Guinea (40,000), East Asia, Europe, West Africa—no farming until 11,000-12,000 years ago (Holocene, end of Ice Age).
Climate experts/archaeologists: Holocene is unusual on 2 million-year scale—not just warmer, but stable (less fluctuation year-to-year, decade-to-decade, century-to-century, per pond isotope data).
Multiple independent agriculture origins follow this stability, despite same genetics from 50,000-300,000 years ago.
300,000-400,000 years ago: Levallois technology (stone tools from cores); Middle Stone Age/Middle Paleolithic Revolution in Africa/Eurasia (shared by Neanderthals, modern humans; absent East/South Asia)—cognitive change.
100,000-50,000 years ago: Upper Paleolithic/Later Stone Age—second transition, less revolutionary.
Diversification of lineages (Khoisan, rainforest hunter-gatherers) ~200,000-300,000 years; all capable today.
Cognitive/behavioral/genetic toolkit possibly in place 200,000-300,000 years ago, even in Neanderthals.
Post-diversification, agriculture ignites independently ~12,000 years ago.
Surprising across environments: maize (New World), cereals (Old World).
Common view: Ancestral population ~70,000 years ago had cognitive ingredients for farming/state societies—descendants independently develop in West Africa, East Africa, Americas, Europe, South Asia, East Asia, New Guinea. Long 40,000-60,000 year fuse; possibly 300,000 years from Neanderthal split.
Population stable 10,000-5000 BC (early Neolithic)—no explosion until Bronze Age.
Unlikely hidden agriculture without metallurgy (copper/tin from Middle East); archaeology would show it.
Americas had extraordinary Stone Age developments (e.g., Teotihuacán in Mexico—massive, without metal, animals, wheels; marble hauled by hand). As impressive as ancient Egypt.
These from groups split 20,000 years from East Asians, 40,000 from West Eurasians—same toolkit, long delay.
Genome sequences from archaics in Europe, West Eurasia, Central Eurasia (Neanderthals); Denisovans (now with skeleton/skull).
Big mysteries in relationships. Denisovans and Neanderthals are sisters from common ancestor 500,000-600,000 years ago; that group from modern human ancestors 700,000-800,000 years ago.
Whole-genome data: Neanderthals and Denisovans from common archaic population.
But many things shared between
Neanderthals and modern humans share many traits not found in East Asians, including Middle Stone Age stone tools, Levallois technology—a cognitively unique method of making stone tools—and the same mitochondrial DNA and Y chromosome sequences. The Y chromosome and mitochondrial DNA of Neanderthals match modern humans, introduced through interbreeding 200,000–300,000 years ago that reached 100% frequency.
Neanderthals and modern humans both result from mixture events between archaic and modern humans 200,000–300,000 years ago, shown by patterns in ancient and modern DNA. Despite genome-wide data clustering Denisovans closer to Neanderthals, Neanderthals and modern humans share something distinct from Denisovans. Possible connections include interbreeding or cohabitation events we have missed.
A known interbreeding event contributed ~5% modern human DNA to Neanderthals' ancestors. This may signal a deeper modern impact, with Neanderthals retaining modern traits despite archaic genome dominance. The shared Middle Stone Age and Middle Paleolithic Revolution suggest this culture is fundamental to their identity.
The interbreeding occurred 300,000–200,000 years ago, making the common ancestor of Neanderthals and most modern humans more recent than that of all modern humans today. Divergence from archaic humans, including Denisovans, falls within human variation. The average time to the common ancestor of any two human genes is 1–2 million years ago; for copies of chromosome 3 from mother and father, it's also 1–2 million years ago—predating the Neanderthal-Denisovan split. In many DNA regions, individuals are more closely related to Neanderthals (e.g., on mother's side) than to their other parent.
This occurs because ancestral DNA 500,000–1 million years ago was highly variable; Neanderthals split recently on human timescales, so in some genomic regions, modern humans are closer to Neanderthals than to each other.
Standard model: Neanderthals and Denisovans descend from a common ancestor ~500,000–600,000 years ago, splitting from modern human ancestors ~700,000–800,000 years ago. Evidence shows modern humans contributed ~5% DNA to Neanderthals' ancestors 200,000–300,000 years ago.
Archaeologically, Neanderthals resemble modern humans more than Denisovans. Pre-2010, Neanderthals were seen as our closest sisters; Denisovan genome sequencing showed Denisovans closer to Neanderthals than to modern humans.
Proposed model: Neanderthals as culturally modern despite mostly archaic (Denisovan-like) genetics, motivated by the Middle Stone Age Revolution/Middle Paleolithic/Levallois technology. This began 300,000–400,000 years ago in the Caucasus (e.g., Georgia) or East Africa, involving high-quality stone (e.g., flint) sourced far away, widespread fire use, and long-distance stone transport.
A population invents Levallois technology and expands into Europe, mixing with local archaics (~5% modern input 200,000–300,000 years ago). Genetically, they become ~95% archaic but retain modern culture.
Genome-wide, Neanderthals cluster with Denisovans, but mitochondrial DNA (maternal) clusters Neanderthals with modern humans (~300,000–450,000 years to common ancestor). Neanderthal Y chromosome (paternal) also ~300,000–450,000 years, while Denisovan mtDNA/Y with modern humans >700,000–800,000 years.
Current view: 5% mixture introduced mtDNA/Y to 100% frequency (improbable by chance, ~0.25% probability). Sima de los Huesos site (Spain, 300,000–400,000 years old) has Neanderthal-like nuclear genome but Denisovan-like mtDNA/Y, suggesting population replacement of those lineages.
Modern humans enter (e.g., Middle East), expand into Europe as a wavefront. Simulations (mammals, birds) show small interbreeding at frontiers leads to massive local gene introgression; pioneers' DNA gets swamped, resulting in ~95% local ancestry by expansion's end.
Matrilineal expansion retains mtDNA; patrilineal retains Y. Modern humans form ~80% of ancestors + ~20% archaic African ancestry (diverged ~1.5 million years ago, mixed 200,000–300,000 years ago). This group expands into Europe (5% modern + 95% local archaic) and Africa (80% modern + 20% local).
Key population drives Levallois/Middle Stone Age expansion ~200,000–300,000 years ago, contemporaneous with African population breakup (Khoisan, Central African rainforest hunter-gatherers). Neanderthals as close cousins via shared mtDNA/Y, mixture timing, toolkit.
Modern Africans/non-Africans show deep substructure: split >1 million years ago (at least two groups, likely more), remixed ~few hundred thousand years ago into anatomically modern humans (including Khoisan). 80% from Caucasus/Northeast Africa (Middle Stone Age origin); 20% from diverse archaic groups across Africa (Southern, Western, Central, Eastern).
Africa: 7 million years of human occupation, landscape of archaic groups. First interaction ~300,000 years ago shares culture/technology without full replacement.
Analogy: Yamnaya expansion traces Indo-European languages/culture despite dilution (e.g., 25% in Corded Ware, <10–5% in India). 5% ancestry traces major cultural signals.
mtDNA/Y clustering paradox unique; explanations: matrilineal (retains mtDNA, incorporates local males, culture via mothers) or patrilineal (retains Y, incorporates local females, culture via fathers). Patrilineality/matrilineality common in humans.
Social selection: Male reproductive variance high; archaic fathers disadvantage offspring in mate competition. Female mate choice favors non-archaic males. Seen in Central African hunter-gatherers (differential treatment by parental origin).
mtDNA/Y survival at wavefront: Modern mtDNA persists in offspring; archaic mtDNA lines may fail due to fitness/social discrimination.
~300,000 years ago: Diffusion, archaic genome survives. ~70,000 years ago: Modern humans replace Neanderthals (genetic ancestry gone). Contingent on divergence time.
400,000 years separation (early interaction): Low incompatibilities. Later (~50,000 years post-divergence): Incompatibilities evident. African archaics (~1.2 million years diverged): Greater barriers (infertility, unfit offspring). Eurasians (~400,000 years diverged): More permeable.
Khoisan-Bantu (e.g., Xhosa) mix despite ~200,000 years separation. Explosion from one origin impacts Eurasia/Africa simultaneously, sharing toolkit (Levallois not independently invented, like farming). Shared mtDNA/Y, timing, culture.
The current model of relationships between archaic and modern humans has accreted over time, starting with modern humans distinct and Neanderthals and Denisovans as sisters. Additional mixture events detected: modern human into Neanderthal, super-divergent lineage into Denisovans.
Despite whole genome indicating Neanderthals and Denisovans as sisters, mitochondrial DNA and Y chromosome show odd patterns explained by invoking natural selection.
This patching resembles Ptolemy's geocentric model where sun revolves around Earth but planetary movements required epicycles—extra rotations to fit observations. Model became convoluted.
Copernicus' heliocentric model (everything revolving around sun) simplified everything. As astronomical data accumulated, it contradicted standard model but was patched with complications.
New alternative model is simpler, explains more, no more complicated than current one.
It's probably wrong, but it's much more plausible. It explains many more things, and it's no more complicated.
Nobody thinking about this model now; seems obviously natural.
Aristarchus proposed heliocentric theory but dismissed because relative movement of stars required them to be incomprehensibly far away (which is true).
Equivalent here: linkage between cultural transformations in Africa and Eurasia, connecting substructure in Africans with ancient DNA of archaic relationships to modern humans. They line up in substructuring time.
Parsimonious: different human groups capable of adopting Stone Age technology; genetic differences not too big to teach stone tools. Possibly genetically driven.
Time to common ancestor of human genes: nothing at 100,000-150,000 years, much at 400,000-500,000 years. Mutation in Caucasus, Middle East, or Northeast Africa could enable this; expands, swamped by local genes in Europe but retained via selection.
Archaeologists see 400,000-500,000-year event as equally or more profound than 50,000-100,000-year anatomically modern human behavior event.
No fixed differences between modern humans and 50,000-year-ago humans, but fixed differences begin at 300,000-400,000 years: all modern humans share common ancestry then.
Anatomically modern, cognitively modern humans by Middle Stone Age start, before breeding with ancient Africans or Neanderthals. Anatomically modern humans and recognizable Neanderthals appear ~400,000-500,000 years ago.
Disconnect: anatomically modern humans in skeletal record vs. behaviorally modern at 50,000-100,000 years. For 200,000 years, made Levallois stone tools like Neanderthals, not more impressive. Then quickening of behavioral traits—could be genetic or not.
Work consistently showed initial biases wrong. Example: Neanderthal genome project—expected no Neanderthal interbreeding into modern human ancestors; non-Africans simple subset of African variation.
Found strong evidence Neanderthals closer to non-Africans than Africans; incredulous, spent years trying to refute, but evidence strengthened.
Natural selection expected quiescent over last several hundred thousand years; early ancient DNA (2015 paper with Ian Mathieson, ~200 Europeans/Middle Easterners) found 12 positions with extreme frequency changes vs. chance.
Hoped more data would detect far more, but 2024 Copenhagen study (better data) found only 21—disappointing given sample size increase, suggesting asymptote.
Study led by Ali Akbari: 14-fold data increase to ~10,000 new individuals; total ~16,000 ancient over 18,000 years (plus 22,000 with modern) from Europe/Middle East (70-80% of ancient DNA literature).
New methodology: predicts genetic type at each position based on relatedness to others in dataset (like disease risk factors in medical studies).
Tests if assuming constant selection direction over time/geography predicts data better than relatedness alone (capturing drift, bottlenecks, admixture).
Analyzed 10 million positions across 22,000 people. Found hundreds changing too consistently over time vs. chance.
At least 479 independent positions (99% confident real); ~3,800 at >50% confident—all pushing same direction. Crazy number vs. prior couple dozen.
Surprised, spent years trying to refute; strengthened. Independent validation: genome-wide association studies (GWAS), UK Biobank (~500,000 people, hundreds of traits).
~15% of 10 million positions predictive of ≥1 trait.
Selection statistic: measures change over time (Gaussian-like, std devs from zero=no selection).
As selection statistic threshold rises (1,2,3,4,5), trait enrichment rises; at >5, ~5-fold (60-70% vs. 15%). Plateau above 5: all signals real (validated by simulations).
Calibrated probability: e.g., 99% to plateau = 99% real.
Concern: signals due to background selection (against bad mutations in genes, which also associate with traits) not adaptive selection.
Repeated analysis in DNA slices with same background selection: same pattern. Same for same-frequency mutations.
Sequencing cost drop: millionfold since late 2000s, 1-2 orders 2010-today.
In-solution enrichment: for low human DNA (<10%, often <1%), use synthesized short DNA fragments targeting >1 million variable/biologically interesting positions (e.g., GWAS hits).
Binds targets; sequencing enriched for informative parts—economically viable.
Other 99%: mostly microbial (bacteria/fungi post-death); sometimes pathogens (plagues, malaria, Black Death, hepatitis B).
Sequences of these pathogens in people's teeth and other parts of their body when they died. But we're focusing here on the human DNA.
This changed the amount of data that was possible to produce from tens per year to hundreds per year, and then we further roboticized and industrialized the process so that there were many hundreds or even thousands per year. Just in our laboratory, we've been generating genome-scale data from more than 5,000 individuals per year.
I know this is true also of several other laboratories in the world now.
This huge jump in data, this semi-exponential or even super-exponential jump in some cases, has made it possible to ask and answer questions. While we were only on the order of 10 genome sequences from humans in 2010, this year it's passed more than 20,000 reported sequences.
There are several orders of magnitude increase, and the questions we were able to ask in 2014 are just not the same as the ones we can ask today.
Awesome. Excellent. David, thanks for your time.
Thank you, Dwarkesh.
Keep Dwarkesh Patel in your library
Save the videos and channels worth coming back to, and find them again in one place.
