The genetic differences we obsess over, the basis for nationalism, racism, and tribal thinking, emerged in just the last 70,000 years. David Reich is about to show us that the past is far stranger, more violent, and more interconnected than we could ever have imagined. 90% population replacements happened in just a few centuries, a blink of a cosmic eye. David and his team reconstructed human migration patterns and discovered ghost populations, entire civilizations that vanished without any archaeological trace. These people built Stonehenge. These people crossed into the Americas. These people left no trace of who they are except for the code that lurks within each of us.
What if I told you that everyone alive today, all 8 billion of us, can trace their DNA back to a group so small they could fit in a modern football stadium? And that this near extinction event 70,000 years ago explains everything from why you get certain diseases to why humans are so similar genetically compared to other species. Soon we'll discuss with Harvard geneticist David Reich how he has discovered all these facts and many more from extracting DNA from bones. They shouldn't even have any genetic material left in them, but still do. And he's performed almost miraculous work in the laboratory and in the field. But first, let me tell you about what is called a ghost population. An entire civilization that vanished without leaving a single archaeological trace. Yet David and his team have tracked genetic fingerprints still hiding in your genome right now.
David, I just came back from the UK. I was at many different sites there, including the British Museum, which has an awful lot of objects that we're going to talk about. But one place I went to that struck me very powerfully was Stonehenge, and sort of a clear indication of how civilizations, massive and mysterious as they are, they can collapse and not leave a remnant other than perhaps what they leave behind in their DNA. So let me ask you this question to open up: If ghost populations left no trace in the landscape, how can we be sure that places like Stonehenge weren't built by cultures that disappeared entirely from the genetic record? Or conversely, if you like, how many Stonehenges are we missing because their creators only left DNA, not stones?
Yeah. So, I think that, thank you for having me on. It's a pleasure to be here. I think that one of the things that is a big open question in our understanding of our human populations and how things have changed in the past is how people have moved to get to where they are today. And what's interesting in this context is with ancient DNA, with the ability to get DNA out of ancient human remains and see how people are related to each other and to people living today, you could actually ask the question: are people who lived in a particular place a long time ago the same as the people who live there today? And so what we found again and again in applying this technology to many places in the world is that people who lived in particular places in the past are not directly ancestral or contributed very little to people today, what you might call a ghost population.
So concretely, in Great Britain, which is actually a place with a relatively simple history compared to more continental places which have had many more populations moving around, in Great Britain there have been relatively smaller numbers of exchanges with the continent, but even there there's very dramatic episodes of change. So if you take the case of Stonehenge, Stonehenge is built by the descendants of farmers who came to Britain about 6,000 years ago from the continent. Before that there were hunter-gatherers from about 14,000 to 6,000 years ago. And before that there was a big block of ice on top of Great Britain and there was no one there. Before that great block of ice there were people there, but they were cleared out of people during the last glaciation. So these farmers who came to Britain after 6,000 years ago displaced the hunter-gatherers, so then almost none of the hunter-gatherers contributed to the later population. They're the people who built the stones. 1,500 years later the last big stones of Stonehenge go up, and then if you look at DNA from people who live just 200 years later, it's completely different. It's a completely different group of people all over Great Britain. It's at least a 90% population turnover, possibly as high as 100%. It's a very dramatic change. And what seems to have happened is there was a stream of people coming from the continent that largely displaced people who were there before. So today, Stonehenge is a sort of cultural treasure in Britain. It's sort of seen as part of the heritage, but it's not built by people who are largely the ancestors of the people who live there today.
So let's get into that, because what you've shown with your work and your colleagues in your lab and in the field is something that challenges how most people think about human identity. The fact that was so striking to me, if I remember correctly, is that every person outside of Africa, whether they're in, you know, Tokyo, Stockholm, or São Paulo, shares ancestry from basically the same tiny founding tribe or population 70,000 years ago. And so before we get to the details, the nitty-gritty technical details that my audience loves so much, I want to get to how you did this using ancient DNA, I want to explore the philosophical implication first and foremost. What does it mean for what we think about human diversity, for nationalism, for racial categorization, if we're all basically relatively recent by cosmological standards, descendants of the same tiny little tribe?
Humans are pretty non-diverse compared to our relatives like chimpanzees. So the population structuring that exists amongst chimpanzees into different groups goes back hundreds of thousands of years. Whereas humans become relatively homogeneous amongst almost all living people today by maybe about 70,000 years ago. There's a few exceptions that go back a little bit more, but basically by that time, and this is true comparing West Africans to most East Africans to other Africans, going back 50 to 70,000 years ago. So what seems to have happened is that there was a group that went through some kind of cultural development, I don't know if it's biological in particular, but that then led to expand in dramatic ways, probably out of Africa and the Middle East, both within Africa and outside of Africa, displacing the humans that were there before, maybe mixing with them a little bit. And that's where we all come from, or where we nearly all come from living today. So the great majority of the changes that have accumulated across human populations owe their origin to events that occurred during this.
So I want to walk through the technological developments and how you fascinated me when I saw you for the first time at the Simons Foundation in June at a tribute to Jim Simons, my late great mentor and I'm sure influence on you and many many of our audience members as well. But I want you to walk us through, because as this guest who I never had on, Carl Sagan, this is him with a huge neck and his trademark professorial patches on his elbows, but I had on his wife, his widow Andrea, and his daughter Sasha Sagan. So I feel kind of close to him, but he said famously, 'Extraordinary claims require extraordinary evidence.' So here's what I find most intriguing, and I'm just a simple, humble experimental cosmologist. How do we get actual DNA samples out of things that are essentially bones and they almost look like they could be in a museum? And how do you use algorithms to turn these fragments that have been degraded, perhaps mishandled or whatever, into detailed human migration maps? So walk us through the technical details of how do you actually get DNA, which I think about as coming from, you know, foolishly or not, vials like this that I scoop up down at the beach, but how do you get DNA to begin with from an ancient almost mineral substance?
Sure. So this was something that 35, 40 years ago a few people were working on to try to understand whether you could get DNA out of remains. And what they were looking at initially were Egyptian mummies or museum specimens 100 years old, especially Spitzmuller, but also others, and it was very challenging to get convincing DNA. There were a lot of false starts. There were a number of cases where it didn't yield anything even though people thought that there were successes. Some people thought they got dinosaur DNA and so on. There was a breakthrough at some point where people started getting convincing DNA. One of these was a paper in 1997 by Svante Pääbo and his team which got DNA out of Neanderthals. So these are archaic humans who lived in Europe 50, 60, 70,000 years ago and more. And they got a sequence from the mitochondrial DNA, the part of the DNA that you get from your mother and she gets from her mother, and occurs in thousands of copies per cell.
And the way this works is you take a little bit of sample, so some milligrams of bone powder, maybe a 40th of the weight of a paperclip, and in a clean room where the goal is to get beneath the surface of this old remain, the part that may be not contaminated by archaeologists or by animals that might have touched it, but might be more representative of the individual whose bone or tooth it was, you then take a little bit of this powder and then you remove the protein and the minerals and you extract DNA. So why is there DNA left there at all? When you extract from an Egyptian mummy, there's usually almost no DNA in it. And the big lucky break seems to be that the biochemistry or the structure of bone contains mineral content that serves as an excellent preserver of DNA. So the hydroxyapatite mineral content of bone seems to catch DNA and adhere it to its surface, to absorb it when cells shatter upon an organism's death and you have these fragmented bits of DNA that are stuck to the hydroxyapatite, and it's a much better preservation even than natural soft tissue. And so the things that we have left from ancient individuals, the bones, the teeth, actually often have DNA intact, so much so that for example for some cases you can get DNA out when you can't even do a radiocarbon date, there's not enough carbon preserved. And so this discovery that DNA was preserved meant that it was possible to go back in time to people who lived thousands, tens of thousands, or in some cases even hundreds of thousands of years ago and extract DNA.
And then the other big development was the technological revolution in DNA sequencing that occurred in the late 2000s when it became possible to sequence absolutely vast numbers of little sequence fragments from DNA samples for about 100,000 or even a million times less than that cost to sequence DNA when the Human Genome Project was moving forward. So this was a transformative change and it meant it was possible to take these millions, billions of highly fragmented bits of DNA that come out of this bone and sequence it. What you get are little fragmented damaged sequences. Most of them are less than 30 DNA letters, bases long, but a handful of them are longer than 30 or 40 or 50 or 60 DNA letters long. And that's long enough to reliably identify where in the human genome or other organisms' genomes they come from. And that's what we do.
So we apply this process that's evolved over time in a clean room where the goal is to protect the sample from the people analyzing it and who might contaminate it, with specialized chemicals that maximize the amount of DNA that comes out but not protein and minerals, with sequencing technologies that maximize the amount of these very rare fragments that get into the sequencing reaction, and then deep sequencing and sometimes enrichment for parts of the DNA that we're interested in. So this combination of things allows us to get sometimes millions or tens of millions or hundreds of millions of DNA fragments from a person. So much so that if you're looking at people who lived at the time of Stonehenge and you're looking at bones, for more than half those people we're able to get genome-scale data of a quality similar to what you get when you send your results to a direct-to-consumer ancestry tester.
You analyze the data. I could tell you a little bit about how you analyze the data, but what you're left with is tens or hundreds of millions of DNA fragments that are human in origin and you can tell are from the individual whose bone it was, and then you try to make sense of that in terms of how they relate to other people you have data from.
Today's episode is sponsored by Shortform, the smartest way to keep up with the world's smartest ideas. Look, I'm a professor, podcaster, and parent, and I don't have hours to spare. When I needed to prepare for a guest like David Reich, the geneticist redefining what it means to be human, I started with their Chrome extension to instantly summarize Reich's papers on ancient DNA and human migration. Then I dove deep into Shortform's full guide on his groundbreaking book, Who We Are and How We Got Here. For today's interview, I leaned on Shortform's summaries of works connected to Reich's research like Sapiens, The Selfish Gene, and Guns, Germs, and Steel to see how genetics, archaeology, and evolution all intertwine. I was on a deadline to release this episode the same week this book got published, so Shortform was indispensable. I needed to make the connections fast, and they didn't disappoint. Reich's ideas connect with thinkers like Richard Dawkins, Jared Diamond, and Yuval Noah Harari, helping me frame how ancient DNA rewrote human history. But it's not just books. Shortform has ads, articles, podcasts, and even AI-generated summaries. I'll use it for research on AI hype cycles, for catching up on the latest astronomy papers, or yes, even sometimes for parenting questions, diving into their podcasts and articles. It's become my daily driver, my curiosity companion, helping me save hours and hours a week. And it will for you, too. You can try it for free. And if you go premium, you'll get three extra months on me. Whether your curiosity has been sparked to explore Reich's discoveries about our shared ancestry, or you just want to dig into the new insights into genetics, evolution, and civilization, Shortform's got you covered. Start your free trial plus three bonus months at shortform.com/impossible or click the notes below. Now, back to my conversation with David Reich.
So they essentially were looking for echoes in the genome and then maybe supplementing that with artifacts of culture. You know, soon we'll get to discuss creatures and cultures that are listed in the Torah and Tanakh and the ancient Hebrew Bible and what they might have been referring to, if that's some kind of echo in our cultural or memome. I don't know. I know I've talked to Richard Dawkins many times, but I don't think he talks about the memotype or the memome. He is of course known for the meme, but before we get there, I want to ask you about the algorithms themselves and how you do... You've described the laboratory kind of processing, but what's actually happening? I mean, I talked to Craig Venter here, you know, last year and he went into great detail about the Human Genome Project, but he basically said it would have been impossible without large-scale computing advances and kind of Moore's Law. How much are you relying on algorithms? Do you rely on LLMs or any other kind of techniques, diffusion models? What kinds of advanced algorithmic technologies are you using? You mentioned before we started with discussion about Nick Patterson and all. Can you talk about the algorithmic, mathematical, code-breaking analyses that you and your team members use?
So the genome is basically a big code and the activity we're engaged in as geneticists, especially historical geneticists but also molecular geneticists, is to read the message. And so it's a cryptographic problem in some way and there are all sorts of levels to this. So once you have the DNA sequences, there is a bioinformatic problem which is just processing the sequences and that involves things like, you might have a data set of 100 million sequences each 40 bases long on average and you want to know what species they come from and where in the DNA they come from. And so that involves performing a computation 100 million times where you compare to a database that is absolutely huge because it consists of all the sequenced genomes of all the organisms that you know, and you want to place this sequence uniquely in this universe of possible sequence spaces that it could be aligned to. So that's an alignment problem and there's very fast mathematical methodologies that will rapidly take a sequence and place it, even though the target is huge and even though the operation has to be repeated tens or hundreds of millions of times.
So technologies have been developed in the 2000s to do this incredibly rapidly and honestly at a scale that you can deal with billions, even trillions of sequences. And so we do this in our laboratory. We use compute clusters that have thousands of nodes that do these operations in parallel, align the sequences. Once we have sequences aligned, typically our question is human-focused in our laboratory. So we'll only analyze the human sequences. We might sometimes analyze even pathogen sequences if there are some, and we will then try to process the data to try to remove the errors or identify the errors or identify the content. So we'll do various operations on the data to try to establish the authenticity of the data. That's a whole other series of bioinformatics analyses and then we'll create a representation of the sequences that we have that is in some sense clean for analysis.
So one version of this might be, let's take a random sequence from our sequencing at every position in the genome we want to analyze and use that random sequence to represent what that person has. That's one possible representation. There's other possible representations that perhaps extract more information but have strengths and weaknesses relative to what it is, and so we might do that. So that's the bioinformatic processing and create a data set where the positions in the genome are analyzed, you know, who we want to compare to other humans we've sequenced from the past and present, we can put them all in the same framework and compare.
Excellent. So keep in mind I'm a simple experimental cosmologist and I know nothing about biology or genetics except for when I did dissections in my high school biology class, the frog came back to life and then suffered immensely. But why are mitochondria, and not, if I remember, Golgi bodies or some other substance in a cell, why is that so much more powerful and effective at doing the reconstructive work that you and your colleagues do?
So most of my work doesn't work on mitochondria at all. We sequence it every time and we collect good data from the mitochondrial sequence, but most of my work does not focus on the mitochondrial DNA. DNA started with mitochondrial DNA work because it occurs in so much higher concentration than anything else in the cell. So at the beginning of the ancient DNA revolution it was magical to even be able to get anything at all and the techniques were so bad that people almost couldn't get anything at all. And in every cell there's maybe about 2,000 mitochondrial genomes for every one nuclear genome and so the copy number is about a thousand times higher.
So if you're not getting almost anything at all, you have a much better shot of covering any part of the mitochondrial genome. And so that was how it started. And the other thing is that mitochondrial DNA is often quite diagnostic. So for example, we now know Neanderthals were units that lived in Western Eurasia up till about 40,000 years ago. Their mitochondrial genomes are completely disjoint and different from mitochondrial genomes of modern humans living today. So it's quite diagnostic of whether you're studying a Neanderthal or modern human whether they have this unique Neanderthal sequence or not. And so that was useful as well. Whereas for the rest of the genome it's a more muddy picture. Typical places in the genome I might be more closely related to a Neanderthal, in many of the genome I have to be, because we're so close.
So, soon we'll get into some of the difficulties you face in terms of biases, calibration, systematic errors. But before we do that, I want to do what you're never supposed to do, which is to judge a book by its cover. I even invented a jingle that will be playing right now. We're going to judge a book by its cover. We're judging books by its cover. Judge books.
David, do you have a copy of How We Got Here with you or should I insert it in post-production? Okay. All right. Great. So, take us through the title, the subtitle, and the artwork, which I have the audiobook. The audiobook has different artwork which I actually, I'm sorry to say, I prefer to that cover that you're showing now. Have you seen the audio cover that I'm talking about here? I'll call it up on screen. I really love this artwork because it... here we go. This is the artwork that I read. I like it a little bit better, but it's got a... it's kind of cool. So, tell us, David, what's the origin of the title, subtitle, and cover art, please?
Well, the origin of this... I forgot what the original proposed title was when I originally set out to write this book. The subtitle, 'Ancient DNA and the New Science of the Human Past,' was I think part of the original proposal. But 'Who We Are and How We Got Here' was not. And a reader of the book, actually Skip Gates, when he read the book he said, 'Why don't we call this book Who We Are and How We Got Here? You use that language many times in the book and you know it's a pretty good description of what the book is about. The book is about how we got to where we are today as a varied group of people around the world and it's explaining this from the point of view of migrations and movements of people through the lens of ancient DNA. So why don't you consider calling it Who We Are and How We Got Here?' So I thought about it for a while and then I eventually thought that's actually really a great formulation. So that's how the main title came about.
Great. And then the artwork that I have has a DNA or a double helix with the kind of digital staircase or something like that. But that one's just a green and white cover. So it's...
This... the artwork is done by Oliver Udy. And there's a few different ones. So yeah, show...
Here's one. So this one has a kind of very subtle picture of the Dali skull.
Which is a possible Denisovan, one of these archaic groups. And then people did lots of different artwork in different languages it was translated into, with various variations or different ideas altogether about this idea. The original artwork was based on these lines which are supposed to be suggestive of not some of the bonds in DNA, but... and is done by this amazing artist I got to work with, Oliver Udy, who worked with me, collaborated with me and drew the images that are in the book, the black and white images.
Excellent. What's the difference between what a Neanderthal is and a Denisovan? Are they just one is from Europe and one is from the steppes of Asia or what is the difference? How do we think about them?
So Neanderthals were known from the archaeological record and the skeletal record since the 1850s. And in Europe they were discovered in the Neander Valley in Germany for the first time and in many other places as well eventually, and they were recognized as close relatives of humans, big brain, as tall as us, and so on. There was no other known close relatives of modern humans until 2010 when the Denisovans were discovered. So what happened when Svante Pääbo set out to sequence the genome of Neanderthals was everybody knew that it would be phenomenally interesting to get a sequence from our closest relatives and that if we could do this by sequencing DNA we would be able to learn what has been specific to our lineage of hundreds of thousands of years since the split from Neanderthals. Maybe we could learn what makes humans distinct from Neanderthals and, you know, of course also chimpanzees. We learned that from sequencing the chimpanzee.
And so that was the goal and it was incredibly exciting to begin to get this data and to try to analyze it and to learn about how the Neanderthals related to modern humans and whether they had interbred with people living outside of Africa as modern humans spread out of Africa at least. That was all very exciting. But there were surprises in that work very strongly. But one of the things that was most surprising was that Svante got a bone shared from a cave in Siberia by some Russian archaeologists and anthropologists, Anatoly Derevyanko and colleagues from the Altai Mountains in Siberia. It was a finger bone from what was probably a little girl that was not classifiable as either a modern human or a Neanderthal, and they thought it was probably one of these two groups. But when Svante sequenced it, it turned out to be something completely different. So it was not a Neanderthal, it was not a modern human. In fact, modern humans and Neanderthals on the mitochondrial DNA are related within about three or four hundred thousand years, but this thing was a million years separated from either. So it was an incredible stroke, incredibly interesting to be able to get the DNA sequence from just a bone without a whole skeleton attached to it, without artifacts and so on, and yet it was something else entirely.
So the history with the Neanderthals, we knew they were going to be interesting and so these were fossils and we were trying to get a genome out of them. But with the Denisovans, we got a genome first, it was a high quality genome, and then we were looking for a skeletal record, cultural record, remains to connect to them, which actually just in the last year it's beginning to accelerate to find those connections.
So the next topic I want to get into is a subject that we usually gloss over, which is the presence of biases. Not just talking about cultural or racial or any kind of biases like that, which are things we talk about in academia ad nauseam unfortunately, but I do want to talk about how you confront the biggest sort of systematic errors. I always tell my students the most important thing is not the value you quote for the Hubble constant or something like that, but how you characterize your ignorance of that. And that ignorance component, the error bar so to speak, has at least two, maybe more different components. One is called systematic and that has something to do with the system, the environment, in our case the galaxy. If we're studying the Big Bang we have to live in the fact that we live on a planet in a galaxy and there's all sorts of contaminants out there as I know all too well. But what sorts of contaminants or biases, systematic, do you have to deal with and how do they compare to the statistical uncertainties that you have from just the number of experimental samples and evidentiary instantiations of these creatures, individuals, just by virtue of the fact that there are at least 8 billion, more of us now than there were back then, these smaller populations. So what are the biggest challenges and how do you mitigate against them?
Yeah, I've never thought about it in terms of the kind of physics framework of systematic and statistical error. We don't quote systematic errors in our fields, but we should actually, particle physicists do. So, I think we should.
So, one thing to realize first of all is that the statistics are not so bad for a lot of genetic measurements. Maybe they're often excellent. So, the genome is a very big thing. So, it's three billion DNA letters long. And within those three billion DNA letters, comparing any two people, there's about a difference every thousand DNA letters, bases. That's about three million differences between any pair of people. And they're living on tens of thousands of essentially statistically independent segments that have moved through different genealogical paths connecting people around the world. So you basically have tens of thousands of independent replicates of this genealogical process across the genome. And so if you can make an inference on, for example, a reconstructed genealogy connecting a whole bunch of samples you've sequenced and say, well, what kinds of population histories could this reconstructed genealogy evolve under? You can then repeat that experiment tens of thousands of times and the data exists to do that. I'm not sure I explained that. Let me try another way.
So, another way to think about this is if you ask, 'How do I relate to you?' You might think, well, maybe we're not going to be able to say very much because we're just one sample, one person each. But that's not true because you are not a single person. You're a multitude of people. And the reason you're a multitude of people is you have two direct ancestors and four second-degree ancestors, grandparents, and eight third-degree ancestors, my great-grandparents. So, it doubles going back every generation. So if you go back ten generations, you have on the order of a thousand ancestors. If you go back a hundred generations, you have on the order of a million ancestors, many of them will be actually the same because they'll be in pedigreed ancestral trees and so on. And your DNA also goes back in time and fragments into segments, not multiplicatively but additively. So every generation your DNA fragments maybe forty times on the female side, twenty-five times on the male side, into segments. And so you have about seventy additional fragmentations going back in time. So if you go back ten generations, your genome will have fragmented into something like seventies...
...ten or about seven hundred ancestral chunks just spread over about a thousand people. And so not so long ago in time, you're querying thousands, tens of thousands of independent ancestors and you could independently study how each person is related to each other. And so that's why when you send your DNA to a direct-to-consumer ancestry testing company like 23andMe and test yourself and see how you're related to other people, you can position your DNA with exquisite accuracy with respect to other people because you just have so much data to work with, so many independent replicates of the evolutionary process. So that's the miracle we have. And when you sequence DNA from ancient human remains from Neanderthals or from Stonehenge, what you can do is similarly position their DNA, which we read out across much of the genome, with respect to all the other data that we have. And what we do is we can learn with high accuracy how they're related to each other even with a single sample. So that's what we do. And that's the statistical error.
Then there's techniques that try to extract even more information or less information and there's a trade-off between bias and amount of information you get. One thing that our research group has done a lot of has been to focus on methods that are robust to some of the biases in the types of methodologies we have.
Yeah. Can you say more about that? Does that have to do with the pseudo-haploid or what you call the downsampling?
Exactly. So those are some of the issues. So you know, I think a lot of these techniques we developed to analyze the Neanderthal genome and the Denisovan genome, which we were analyzing as part of our collaborative work with Svante Pääbo's group, and what we wanted to do was build techniques that could deal with really fragmented data that had various sorts of biases. And so we emphasized these methodologies that would not be so affected by these biases. And so for example, one issue that happens is that we have uneven coverage across the genome. So some places we have data from, some places we don't. How do we deal with that? Some places we have lots of data from. So how do you represent every position in the genome? A tempting thing to do might be to do majority rule, like take if you have fifteen sequences covering a particular position, let's take the majority type. But there's subtle biases related to that that actually are really terrible when you actually are carrying out these sensitive statistics.
Like for example, if there's a bias toward when you align the DNA to the genome, it's going to... if you're looking at a position that's variable, some people have one type and some people have another type, that's where your information comes from, the variability. The type that aligns to the human genome reference sequence, the common one that's been sequenced and that everybody uses for DNA alignments, it will align more easily. So if you're a variable person who has one copy of the variant from their mom and the other copy from their dad, the version from say the mom that matches the reference sequence will align better and so will always win in majority rule. So you'll have a reference bias that if you're not careful, toward the majority variant. So this is...
