David Reich and Dwarkesh Patel discuss ancient-DNA evidence for selection, the limits of modern trait predictors and possible models of Neanderthal origins. Eight research notes distinguish measured genetic patterns from statistical interpretation and speculative history. The closing discussion examines methods, confounding and the expanding scale of ancient-DNA datasets. They also consider climate, agriculture and the limits of using genetic evidence to explain cultural change.
HostDwarkesh Patel
GuestDavid Reich
About these timings. Publisher-supplied VTT cue times for the official audio. Automated caption recognition contains errors; Discussion summaries use clear passages, checked against the publisher’s editorial transcript. No independent audio audit or cross-platform offset is claimed. Duration is rounded up from publisher metadata.
Reich explains why a large ancient-DNA sample is needed to distinguish selection from drift and population movement.
Background and context
Akbari and colleagues’ April 2026 paper analysed 15,836 ancient West Eurasians, including 10,016 with newly generated data. Their method tests whether variant frequencies change consistently through time after accounting for genetic relatedness. Simulations and comparisons with independent trait associations help assess whether the signal could instead reflect population structure or other processes.
The geographic and statistical scope matters. These are sampled West Eurasian histories, not a uniform census of humanity. A model that distinguishes sustained change from migration does not automatically identify the historical cause of every selected variant. Nor do present-day associations reveal precisely what a trait meant in an ancient environment. The paper expressly notes that effects measured in industrialised societies may not describe the phenotypes favoured in the past. Reich’s opening is strongest when understood as a methodological result with those limits attached.
The inference is regionally sampled and model-dependent; historical causes require additional evidence.
Reich proposes tuberculosis as a possible explanation for the later decline of a TYK2 variant, while calling the interpretation speculative.
Background and context
Kerner and colleagues’ March 2021 study examined the TYK2 P1104A variant in 1,013 ancient human genomes. Using demographic models and approximate Bayesian computation, they inferred a marked frequency decline after the Bronze Age and strong negative selection beginning roughly two thousand years ago. The interpretation drew on separate clinical evidence that people carrying two copies have increased susceptibility to tuberculosis.
This combines a measured genetic trajectory with a model of selection and an identified biological vulnerability. It is not a direct count of ancient tuberculosis cases, and the proposed disease pressure does not by itself explain every earlier rise in the variant. The distinction also matters for present-day readers: the paper’s claim concerns a particular genotype and population history, not a diagnosis from a podcast example. Reich’s proposed reversal of advantage remains a historical explanation to evaluate against independent evidence.
A variant-frequency model is not direct surveillance of historical infections or individual medical advice.
Reich corrects the discussion from IQ to educational attainment when describing a recent Icelandic selection study.
Background and context
Kong and colleagues’ January 2017 paper related an education-associated polygenic score to reproductive histories of 109,120 Icelanders. Higher scores were associated with later reproduction and fewer children, including after adjustment for participants’ education. In a larger sample born from 1910 to 1990, the average score declined by about 0.010 of its standard deviation per decade.
That quantity is a change in a constructed genetic score. It is not a measured decline in schooling, a direct estimate of intelligence, or a scale on which ancient populations can simply be ranked. The authors discuss environmental contributions and the possibility that associations change with social conditions. Their observational evidence does not identify one universal psychological mechanism connecting every variant to fertility. Keeping the score, population, period and outcome explicit preserves the study’s finding without turning it into a claim about human worth or an inevitable future.
Modern Icelandic associations cannot be interpreted as ancient IQ measurements or an innate hierarchy of populations.
Reich distinguishes a proposed universal genetic change underlying recent behaviour from gradual changes involving many existing variants.
Background and context
The 2016 Simons Genome Diversity Project reported high-quality genomes from 300 people representing 142 populations. Its analyses asked, among other questions, whether a recent selective sweep shared by all living humans could explain the archaeological emergence of recognisably modern behaviour. The study found no strong signal fitting that history, while acknowledging imperfect detection power and excluded genome regions.
The scope of the test is narrower than saying humans stopped evolving. A strong new mutation spreading to fixation across all populations is one possible process; changes in the frequencies of many older variants can leave different patterns. Genetic and archaeological dates also concern different observations, so their relationship requires an argument rather than an automatic match. Reich’s distinction between those mechanisms helps explain why evidence for recent regional selection can coexist with this earlier result. Neither result assigns an immutable cognitive capacity to any present-day population.
A test for universal selective sweeps does not exclude polygenic adaptation, cultural change or every form of selection.
Reich explains why mitochondrial DNA, Y chromosomes and the rest of the genome can imply different relationships among archaic humans.
Background and context
Meyer and colleagues’ March 2016 study recovered nuclear DNA from two Sima de los Huesos specimens in Spain, approximately 430,000 years old. Those sequences connected the specimens more closely with Neanderthals than Denisovans. Yet a recovered mitochondrial sequence retained the earlier reported affinity to Denisovan mitochondrial DNA. The authors discussed later mitochondrial replacement among possible explanations.
The disagreement is useful evidence, not a reason to treat one small inherited component as the whole population’s ancestry. It also does not uniquely establish the expansive scenario Reich develops later in the interview. A successful historical model must account for the nuclear and uniparental results together, with the samples and dates available. The paper supplies a concrete example of the puzzle while leaving open how much migration, drift, selection or other population processes contributed to the eventual pattern.
Two nuclear specimens and a mitochondrial sequence constrain models; they do not uniquely determine an entire population history.
Reich says modern African genomes contain evidence of deep population structure but acknowledges that its geography remains uncertain.
Background and context
Ragsdale and colleagues’ May 2023 study compared demographic models using genomes from contemporary African populations. Models with weakly differentiated ancestral populations connected by gene flow fitted the data better than the particular long-isolated archaic-admixture models tested. The researchers stressed that model assumptions can change inferred parameters and that more complex alternatives may still fit better.
This matters for the whiteboard discussion: an inferred ancient branch is not itself a recovered fossil population, a known location, or a uniquely established encounter. The paper also cautions against mapping its ancestral stems directly onto present-day populations. It offers an alternative to treating every deep genetic signal as evidence of a separate archaic group that later mixed with modern humans. Reich’s broader range-expansion scenario therefore needs to distinguish what these modern genomes constrain from the additional historical narrative attached to them.
The analysis uses contemporary genomes and model comparisons, not comparably ancient African DNA identifying the proposed groups.
Reich presents a possible shared expansion behind Neanderthal and modern-human histories while repeatedly acknowledging that it could be wrong.
Background and context
Reich’s March 2026 preprint explicitly presents a hypothesis. In its first version, he proposes that an expanding population mixed extensively with resident archaic people in Europe and more modestly in Africa. He argues that this could connect otherwise puzzling similarities in inherited lineages, technology and inferred dates. The version matters: this note refers to the March 13 text, rather than silently substituting a later revision.
The paper’s proposed explanatory fit is not independent confirmation that the expansion occurred. Shared tool forms do not by themselves identify the makers’ ancestry, and the suggested sex bias cannot simply be assumed to explain both maternal and paternal lineages. These are parts of the mechanism to test. Considering the preprint alongside the interview’s repeated qualifications preserves its value as a research proposal while keeping its numerical mixture scenario and archaeological connections separate from established observations.
This cited version is a preprint hypothesis, not peer-reviewed confirmation of the proposed migration or ancestry percentages.
Reich returns to the selection study’s statistical checks and explains why larger datasets reveal signals earlier work could miss.
Background and context
Mathieson and colleagues’ November 2015 analysis used 230 ancient West Eurasian genomes, including 163 newly reported individuals, to search for selection. It detected signals associated with diet, pigmentation and immunity, while also documenting substantial ancestry changes. That historical study illustrates how ancient DNA already combined biological and demographic questions with a much smaller sample than the later project.
Comparing the number of discoveries across papers is therefore not a direct measurement of a change in the past rate of evolution. Samples, covered periods, statistical models and detection thresholds also change. The 2026 analysis asks additional questions within its own time series; those comparisons must be evaluated through its model and robustness checks. Reich’s methodological conclusion is stronger when improved detection is kept distinct from the separate claim that the intensity of selection varied between ancient periods.
Different discovery counts across studies cannot by themselves measure differences in historical selection intensity.
Remember a word from the conversation? Find an approximate listening time without reading through a transcript.
Times follow the Publisher audio edition. All times are approximate. Use up to six words; matching uses 30-second windows, not exact phrases. Transcription can be wrong, so listen to check. Searches stay in your browser.
PodCompanions