Razib Khan's Unsupervised Learning

On this episode of Unsupervised Learning Razib talks to Tade Souaiaia, a statistical geneticist at SUNY Downstate about his new preprint, Striking Departures from Polygenic Architecture in the Tails of Complex Traits. Souaiaia trained as a computational biologist at USC, but also has a background as a division I track and field athlete.

Razib and Souaiaia discuss what "genetic architecture" means, and consider what we're finding when we look at extreme trait values in characteristics along a normal distribution. Though traits like height or risk for type II diabetes can be thought of as represented by an idealized Gaussian distribution, real molecular and cellular processes still underlie their phenotypic expression. Souaiaia talks about how genomics has resulted in an influx of data and allowed statistical geneticists with a theoretical bent to actually test some of the models that underpin our understanding of traits and examine how models like mutation-selection balance might differ from what we've long expected. After wading through the depths of genetic abstraction and how it intersects with the new age of big data, Razib and Souaiaia talk about race and sports, and whether there might be differences between groups in athletic ability. Souaiaia argues that the underlying historical track record is too variable to draw firm conclusions, while Razib argues that there are theoretical reasons that one should expect differences between groups at the tails and even around the memes.

On this episode of Unsupervised Learning Razib talks with Washington Post columnist Shadi Hamid. A native Pennsylvanian of Egyptian ethnic background, and Islamic faith, Hamid completed his Ph.D. in politics at Oxford University. He is an assistant professor at Fuller Seminary, co-host of the Wisdom of Crowds podcast and website, and now the author of his own Substack and a recent book, The Problem of Democracy: America, the Middle East, and the Rise and Fall of an Idea. Hamid is also the author of Temptations of Power: Islamists & Illiberal Democracy in a New Middle East and Rethinking Political Islam.

Hamid and Razib discuss the tail end of the war in Gaza, from the explosion of 10/7 and the wave of atrocities against Israelis surrounding the Palestinian enclave, to the brutal counter-attack that has resulted in tens of thousands of Gazan civilian deaths. While Hamid points to the deep structural issues that divide the two parties, and make final resolution of the conflict difficult, Razib highlights the many pitfalls of third parties becoming involved in such a highly polarized and fraught topic. They also discuss the growing identification of the global Left, including American progressives, with the Palestinian cause, the difficulties of grappling with and containing anti-Semitism within the movement. Though Israel's counter-offensive is finally reaching a denouement, Hamid strikes a fundamentally pessimistic note about long-term possibilities.

Then they pivot to domestic politics, and recent cultural trends that culminated in a Trump victory in the 2024 USelection, and the alienation of many nonwhites in the Democratic coalition from the hegemony of woke cultural elites. Hamid reiterates his long-standing critiques of racial identitarianism on the Left, and the irony that the progressive awareness of racial minorities only tends to extend to them when these minorities cosign woke nostrums. In contrast to the seemingly interminable nature of the conflict in the Middle East, Razib and Hamid both see hope for a path forward with reduced racial polarization and a reorientation of politics around substantive material interests rather than symbolic racial or ethnic categories.

On this episode of Unsupervised Learning Razib talks to Conn Carroll, the author of Sex and the Citizen: How the Assault on Marriage Is Destroying Democracy. Caroll is currently an editor for the Washington Examiner, but previously he was the communications director for Senator Mike Lee of Utah, an assistant director at the Heritage Foundation, White House correspondent for Townhall.com and a reporter at National Journal. Carroll wrote Sex and the Citizen in response to what he felt was misleading and biased reporting in the mainstream media on the origins and implications of marriage and monogamy.

Razib asks Carroll how he refutes the ideas presented in Christopher Ryan and Cacilda Jethá's Sex at Dawn, which argues that prehistoric humans were non-monogamous. Carroll outlines the current mainstream thinking in evolutionary anthropology and primatology, and all the biological reasons that indicate that Homo sapiens is far more monogamous than our common chimpanzee and gorilla cousins, most clearly in our reduced sexual dimorphism.

But while our hunter-gatherer past was defined by monogamy, Sex and the Citizen argues that the rise of agriculture resulted in the explosion of polygamy, as high status males in societies defined by incredible inequality began to monopolize access to women, culminating in the explosion of Y chromosomal "star phylogenies," where supermale lineages exploded all over Eurasian 4,000 years ago. Carroll then explains that the Romans and Greeks took steps toward enforcing monogamy as a legal institution, and Christianity introduced the idea of sexual fidelity upon men. After Christianity popularized egalitarian monogamous marriage, Sex and the Citizen follows in Joe Henrich's wake in The WEIRDest People in the World: How the West Became Psychologically Peculiar and Particularly Prosperous. Carroll discusses the Catholic Church's strict policies on incest and adoption, which destroyed the power of elite related clans in the West, and hastened the emergence of the Western European marriage norm of independent and separate nuclear families, rather than extended families as the primary unit of kinship in society.

In the second half of Sex and the Citizen Carroll addresses the social history and policy changes in relation to marriage in the US. While Western European societies took a significant step away from familialism, Carroll explains that American marriage was even more individualistic and radical, as nuclear families spread out to the frontier, away from their extended kin networks. He also contextualizes the rise of the 1950's nuclear family, which some scholars have argued was an aberration in American history. Carroll argues that actually it was an extension of earlier American norms, but the rise of the wage-based capitalist economy allowed for couples to set up separate households earlier in their lives. Carroll concludes the discussion outlining the 1970's policy changes in welfare provision that discouraged marriage, noting the decline of the institution across American society over the last 60 years, and how government policy might reverse it.

On this episode of Unsupervised Learning Razib has a wide-ranging conversation with Dan Hess, the man behind the More Births account on social media. An engineer with a large family in the DC area, Hess' essays on topics like Israelis' high birth rate have gained the attention of X, with an account that has come from a few hundred followers to more than 30,000 in 2 years.

Razib and Hess first review the birth-rate collapse seen worldwide in the past two decades. They discuss the relatively abrupt cultural pivot that has occurred since the turn of the century, with the end of the "overpopulation" narrative typified by Paul Erhlich's Population Bomb, the rise of the "birth dearth" and the natalist movement. They talk about the most extreme cases of low total fertility rates (TFR) in Europe and East Asia, but also the decline in societies like the US, Latin America and the Middle East. Hess addresses both possible causes and possible solutions. They also discuss historical and demographic factors that impact fertility; for example, which religions have been the most pro-natalist? Hess also puts a particular focus on South Korea, the world's most extreme case of a sharp decrease, with a TFR of about 0.70 children per woman (vs. 2.1 replacement), as well as exceptions to the rule like Haredi Jews and the Amish. Finally, Razib and Hess tackle why we should care about slower population growth in this century, from dependency ratios to the impact on cultural vitality.

On this episode of Unsupervised Learning Razib talks to Brian Chau, who writes at the From the New World Substack. A graduate of the University of Waterloo and former software engineer with a background in pure mathematics, today Chau is executive director of the Alliance for the Future, a think tank that believes artificial intelligence will transform our world for the better.

Chau addresses the great "doomer vs. anti-doomer" debate, and argues for an anti-catastrosophist position. He also makes the case that increasing scaling has started to hit diminishing returns, and the expectation that artificial intelligence will continue to gain power purely through throwing more resources at the same problems. Then, they discuss the revolutionary impact of the recent advances DeepSeek has made in China (an issue he addresses on his Substack). Chau breaks down the technological nuts and bolts, as well as geopolitical and economic consequences.

On this episode of Unsupervised Learning, third-time guest John Hawks returns after two years to discuss what we've learned in paleoanthropology since he and Razib last talked. Hawks obtained his PhD under Milford H. Wolpoff, and is currently a professor in anthropology at the University of Wisconsin. Hawks has also co-authored Almost Human: The Astonishing Tale of Homo naledi and the Discovery That Changed Our Human Story and Cave of Bones: A True Story of Discovery, Adventure, and Human Origins with Lee Berger.

Razib first presses Hawks on what we know about archaic human admixture into modern populations, and particularly what we've learned about Denisovans. They discuss how many Denisovan populations there were, how many Denisovan fossil remains we have, and why it has taken so long for researchers to assign a species name to this lineage of humans. Hawks also address the puzzle of the phenomenon of why there are at least two pygmy hominin populations in Southeast Asia. Perhaps humans too are subject to island dwarfism like many other mammals? Also, Razib wonders why Southeast Asia was home to such a startling variety of humans at once prior to the arrival of modern populations. They discuss all of this in light of the framework of Out-of-Africa, the recent spread of anatomically modern humans outside of Africa. Razib questions how robust this model is today given our understanding of modern humans' extensive and repeated interactions with both Neanderthals and Denisovans. Finally, Hawks covers some controversies over fossils being sent into space that roiled the archaeological world last year.

Three years ago David Mittelman came on Unsupervised Learning to talk about emerging possibilities on the frontiers of genomics, and his new startup at the time, Othram. Since then, Othram's work has been featured widely in the media, including in a Law & Order episode, and the firm has solved thousands of unsolved cases, with nearly 500 public. For over a decade, Mittelman has been at the forefront of private-sector genomics research. He trained at Baylor College of Medicine and was previously faculty at Virginia Tech.

Razib and Mittelman discuss the changes that the rapid pace of genomic technology has driven in the field of genetics, from the days a $3 billion dollar draft human genome in the year 2000 to readily available $200 consumer genomes in 2024. One consequence of this change has been genetics' transformation into information science, and the dual necessities of increased data storage and more powerful, incisive data analysis. Genomics made information acquisition and analysis so easy across the research community that it allowed for the pooling of results and discoveries in big databases. This has pulled genetics out of the basic science lab and allowed it to expand into an enterprise with a consumer dimension.

Mittelman also discusses the improvements and advances in DNA extraction and analysis techniques that allow companies like his to now glean insights from decades-old samples, with bench sciences operating synergistically with computational biology. Razib and Mittelman talk about how he has helped solve hundreds of cold cases with new technology, in particular, at the intersection between new forensic techniques and both whole-genome sequencing and public genetic databases. They also discuss the future of genetics, and how it might touch our lives through healthcare and other domains, passing from inference to fields like genetic engineering

On this episode of Unsuperivsed Learning reviews what we know about Indo-Europeans as 2024 comes to a close. This is prompted by a new preprint Ancient genomics support deep divergence between Eastern and Western Mediterranean Indo-European languages, which finally establishes that populations in Northern and Southwestern Europe derived from a different steppe-origin population than the Greeks and Ilyrians of the Balkans, as well as Armenians. Razib talks about how ancient DNA is resolving long-standing disputes in historical linguistics, and coming down on the side of very particular sets of hypotheses. He discusses Peter Bellwood's First Farmers: The Origins of Agricultural Societies, and its models about the origins of Indo-European languages, and how they have been falsified by paleogenomics. Razib also steps through the relationship of particular Indo-European groups to ancestral archaeological cultures like the Corded Ware, Bell Beaker and Catacomb Cultures. He also talks about the connections between charioteers and the early Mycenaeans, and looks at Robert Drews' ideas in Coming of the Greeks. Finally, he addresses outliers in the ancient DNA data that indicate connections between locales as disparate as Scandinavia and Cyprus 4,000 years ago.

This week on Unsupervised Learning Razib talks to Megan McArdle, author of The Up Side of Down: Why Failing Well Is the Key to Success and Washington Post columnist and op-ed board member. McArdle was raised in New York City and attended Riverdale Country School. She obtained an undergraduate degree in English from University of Pennsylvania and an MBA from the University of Chicago. A pioneering blogger based out of New York City and covering the site of the WTC in the wake of 9/11, McArdle went on to work at The Economist, The Atlantic and Newsweek.

In this episode, the discussion largely focuses on McArdle's research about the cultural history of food and cooking in the US. But first they discuss the economic implications of Donald Trump's election, and the domestic consequences shifting toward a tariff-heavy trade regime. McArdle lays out the case that a massive tariff would have the same impact as a tax, not to mention the broad disruptive economic effects on large companies' supply chains.

Then they move on to the changes in American cuisine over the last few centuries, and the shifts driven by technology and innovation. McArdle points out that in the 19th century, gelatin dessert was a luxury and an exotic treat because it was labor intensive to prepare. But by the middle of the 20th century industrial-scale food processing made gelatin, in particular Jell-o, a cheap commodity, and it became associated with the lower classes. Similarly, before factory farming, chicken and eggs were more expensive than red meat, and thus viewed as high-end ingredient (whereas today, chicken is far cheaper than beef). Finally Razib and McArdle talk about how the plentitude of food available in the 21st century contributes to the obesity epidemic that has only ceased its relentless expansion with the advent of Ozempic.

On this episode of Unsupervised Learning, Razib catches up with Nikolai Yakovenko about the state of AI at the end of 2024. Yakovenko is a former professional poker player,and research scientist at Google, Twitter and Nvidia. With a decade in computer science, Yakovenko has been at the forefront of the large-language-model revolution that has given rise to multi-billion dollar companies like OpenAI, Anthropic and Perplexity and hundreds of smaller startups. Currently, he is the CEO of DeepNewz, an AI-driven news startup that leverages OpenAI's latest model. Full disclosure: Razib actively uses and recommends the service and is an advisor to the company.

Razib and Yakovenko first review what makes the last few years special, the rise of large-language-models on top of neural network architecture of transformers. Yakovenkoi discusses how far they've come since OpenAI released ChatGPT to the public in the fall of 2022, and how people have been using the underlying technology to develop applications atop it. Despite predictions of mass unemployment, Razib points out that two years later America is at full employment, and only niche fields like translation have been impacted. In contrast, Yakovenko points out that most software developers use artificial intelligence in some form to aid in their daily engineering work, noting the possibility that the AI revolution is integrating itself seamlessly as a utility for preexistent jobs. They also discuss the fact that though AI is a booming field, only one brand-name company has so far emerged in the industry, OpenAI. Though they agree that the current hype cycle is now abating, it is clear that the major investments in the field like data centers will continue from major players as AI-driven applications like self-driving cars become more and more mainstream.