Reich: Bronze Age Selection Boom Rewrites Evolution

Natural Selection's Unexpected Recent Surge

David Reich overturns decades of consensus that natural selection has been dormant in humans since agriculture began 10,000-12,000 years ago. Using massive ancient DNA datasets from Europe and the Middle East—now feasible due to industrialized sequencing—Reich and Ali Akbari developed statistical methods to distinguish directional selection from genetic drift and migration effects. They identified ~7,200 genomic positions (50% confidence) under selection in the last 10,000 years, with the genome 'vibrating' from widespread weaker signals. Only 2% of allele frequency changes are due to selection, but its effects are pervasive, explaining prior observations of low differentiation between Europeans and East Asians: selection has accelerated recently, not steadily over 40-50k years.

Reich attributes this to environmental shocks post-hunter-gatherer life. Populations shifted to farming, dense living near animals, and pastoralism, pressuring adaptation. Host Dwarkesh Patel probes why group replacements (e.g., Yamnaya steppe migrations replacing 40-80% ancestry) aren't selection; Reich clarifies these cause genome-wide shifts, uninformative for pinpointing variants, unlike stable periods where alleles steadily rise/fall across isolated 'archipelagos' of populations.

"The genome is reacting much more strongly to these events that happened 5,000 years ago," Reich notes, countering the cartoon view that Neolithic farming was the biggest transition.

Bronze Age as Peak Evolutionary Inflection

Selection intensified dramatically ~5,000 years ago in the Bronze Age, more than during initial Neolithic farming spread ~8,500 years ago. Reich links this to 'wrenching' lifestyle changes: higher densities, animal herding, and cultural intensification among food-producers. Immune traits show 4-5x enrichment in selection signals, metabolic traits (obesity, fat, Type 2 diabetes) strongly too. This aligns with pathogens from density/domestication and dietary shifts.

Behavioral/cognitive traits show no strong single-site enrichment—Reich explains they're polygenic (many weak-effect genes), beyond current power to detect amid noise. Yet evidence confirms selection on them; polygenic scores for cognitive performance rose ~1 standard deviation over 10k years, mostly 4-2k years ago. Patel asks why no intelligence maximization: Reich cites time constraints (evolution lags environment), pleiotropy (genes affect multiple traits), and no population bottleneck limiting variation.

"Humans... were wrenched into a way of living... so different... that the organism had to adapt very strongly. Maybe the degree... moving into the Bronze Age was qualitatively greater than... the initial transition to growing plants."

Neanderthal Puzzle: A Heretical Modern-Human Origin

Reich sketches a radical model challenging Neanderthals as a separate archaic lineage with minor interbreeding. Instead, ~300,000 years ago, a small Middle Stone Age (MSA) tech-using population near the Caucasus expanded: European branch admixed heavily with local archaics, becoming genetically 'swamped' Neanderthals; African branch mixed with diverged archaic Africans, yielding modern humans. Both share MSA cultural ancestry; differences stem from post-expansion archaic mixtures.

This fits evidence better than the standard story, per Reich: shared advanced behaviors, DNA patterns. Patel captures the whiteboard sketch post-recording. Reich admits it's heretical but evidence-driven.

"Neanderthals are essentially genetically-swamped modern humans."

Methodological Leaps Unlocking Insights

Prior ancient DNA excelled at history (migrations, mixtures) via single genomes representing thousands of ancestors, but biology needed thousands of samples per era for frequency tracking. Reich's lab scaled high-quality, cheap sequencing. New stats partition selection from 98% drift/migration noise, using isolated population intervals (e.g., Britain or Hungary for centuries between waves).

They confirm ~3,600 real strong signals (half of 7,200 at 50% confidence), clustered non-randomly. Lower thresholds reveal tens of thousands more. Workflow: Genome-wide association studies (GWAS) map signals to traits; compare eras (last 5k vs prior 5k years) for acceleration.

Reich reconciles with quiescence views: recent speedup hides long-term subtlety; no 100% frequency fixes across groups because selection episodic, recent.

Why Farming Delayed Until Post-Ice Age

Patel asks why no pre-Ice Age farming. Reich notes hunter-gatherers thrived in Ice Age scarcity; post-glacial plenty reduced incentive. Farming inventors in Middle East ~12k years ago exploded into Europe.

Notable Quotes

• "I am back with David Reich... there's so much about human history we don't know and are just learning about now." — Dwarkesh Patel, framing the field's revelations.
• "Natural selection has actually sped up... went especially bonkers during the Bronze Age." — David Reich (from description, echoed in talk).
• "The dream was... going to learn a lot about biology... that dream has really not been realized... until the last few years." — David Reich on ancient DNA's biology shortfall.
• "Instead of being quiescent, natural selection is everywhere." — David Reich on genome-wide effects.

Actionable Insights

• Scale ancient DNA to thousands per era for trait frequency tracking; industrialize sequencing for cost/quality.
• Use stats isolating stable population intervals to filter migration noise: test allele trajectories across 'archipelagos'.
• For polygenic traits, aggregate weak signals via predictors (e.g., cognitive polygenic scores) rather than single sites.
• Cross-reference selection hits with GWAS for trait enrichments; compare eras for acceleration patterns.
• Test archaic admixture models with expanded global ancient genomes, focusing shared tech timelines.

Relevance Assessment

For dev/AI/tooling pros tracking cutting-edge science: Reich's stats mirror ML signal detection amid noise (drift as confounders); polygenic selection insights inform AI-genetics models (e.g., predicting cog performance shifts). Bronze Age 'wrenching' parallels rapid tech shifts demanding adaptation; Neanderthal rethink challenges deep learning priors on lineages. Helps skip hype, grasp evolution's pace for human-AI augmentation debates.

People & Entities

• David Reich: Harvard professor of ancient DNA; leads lab generating massive datasets; co-author of selection preprint.
• Ali Akbari: Postdoc turned staff scientist in Reich's lab; co-lead on selection study.
• Dwarkesh Patel: Podcast host/interviewer; probes naive questions, captures Reich's Neanderthal sketch.

Tools, Products & Workflow

1. Ancient DNA sequencing: Reich's lab workflow—industrial scale, low-cost, high-coverage from bones/teeth.
2. Selection detection stats: Partition frequency changes; flag >1% rates (doubling in dozens generations); enrich vs GWAS traits.
3. Polygenic scoring: Aggregate variants for traits like cognition; track ~1 SD rise. Sponsors mentioned: Cursor (parallel LLM research), fastokens (tokenizer speedup), Jane Street (compute auction).

Tensions & Open Questions

• Bronze Age shock > Neolithic? Genome reacts stronger to pastoral density than initial farming—surprising culturally.
• Cognitive selection real but undetectable at sites; polygenic proof via scores, but causation unclear.
• Neanderthal model: Fits DNA/culture but awaits Caucasus 300kya MSA confirmation; vs standard interbreeding.
• Pop size not limiting, but why no super-intelligence? Pleiotropy/ tradeoffs unresolved.
• Global generalizability? Study Europe/Mideast; Africa/Asia patterns?

Key Takeaways

• Natural selection ramped up post-agriculture, exploding in Bronze Age due to density/pathogens/pastoralism.
• Immune/metabolic traits most hit; behavioral/cognitive polygenic, but scores rose 1 SD recently.
• Distinguish selection via stable intervals amid migrations—98% noise from history.
• Neanderthals: Admixed MSA moderns swamped by European archaics; we by African ones.
• Large samples + new stats unlock biology from ancient DNA; field shifting from history to adaptation.
• Evolution time-limited; no optima reached—ongoing amid shocks.
• Farming post-Ice Age: Plenty dulled HG efficiency incentive.
• Method: GWAS enrichment, era comparisons reveal acceleration.
• Genome everywhere under tug-of-war selection, not quiescent.