Razib Khan's Unsupervised Learning

How is it that babies across entirely different cultures seemingly elicit one single sort of “baby talk” from adults? To answer this question, Razib talks to Cody Moser, coauthor of a recent paper on the topic, and an evolutionary psychologist and cultural evolutionist at UC Merced. Moser first discusses what cultural evolution today means in the context of American anthropology, and how it relates to the new field of evolutionary psychology. He observes that some of the conceptual ideas that underpin modern cultural evolution actually have roots in naturalistic frameworks dating back decades, though out of fashion in American cultural anthropology since just after World War II. Razib and Moser compare and contrast the descriptive and interpretive methods of most cultural anthropology, and the formalistic evolutionary paradigm of cultural evolution. Then Moser gets to the meat of the paper on which he was the second author. He points out that humans have noted the similarities across cultures for decades, with evolutionary psychologists concluding this is a human universal. But it takes evolutionary and cognitive frameworks to understand how this phenomenon emerges naturally out of the common biological heritage of our species. Moser outlines the structural conditions that result in the universality of baby talk across cultures, and what benefits this universality might confer upon us as a species.

When Jack Dorsey stepped down as Twitter CEO last year, I wondered what we could expect from the new leader, Parag Agrawal. Luckily, I knew Nikolai Yakovenko, who worked at Twitter on deep neural networks in the mid-teens. Yakovenko told me Agrawal was not a rock-the-boat kind of guy, and perhaps that’s why Dorsey tapped him to head Twitter after some tumultuous years. Now that Twitter and its leadership is in the news again, due to Elon Musk’s status as “chief twit,” I wanted to talk to Yakovenko about his time at Twitter, discuss the application's upsides and downsides, and get his take on what we can expect going forward. Yakovenko is now the CEO of Deep NFT Value, and has extensive experience in crypto, machine learning and deep learning. We discussed the nitty-gritty of how Twitter’s algorithm works to prioritize and deprioritize certain types of content, what precisely deep neural networks are, and how they are relevant to what Twitter does. More generally, we discussed why machine learning has become so important in the technology space over the last decade, and why a company like Twitter has become heavily invested in the field. Finally, Yakovenko talks about the general prospects of Twitter going forward under the new Musk regime. Note: the archive of podcasts (2 weeks delayed) now also resides on YouTube as well as Apple, Stitcher and Spotify (though my “monologues” will not be posted in full ungated).

On this episode of the Unsupervised Learning podcast, Razib discusses the history and genetics of Anatolia, from the first farmers to the Ottoman conquest of the peninsula. He focuses on the underappreciated reality that prehistoric Anatolia was the font of the first wave of farmers that built the majestic Neolithic societies of Europe, from arid Iberia north to the shores of the Baltic. These people left the vast stoneworks that dot Europe’s Atlantic coasts to this day, beginning with the megaliths of Brittany and culminating in the enigmatic site of Stonehenge. Razib also points out the role of Anatolia in the emergence of historical states, like the nearly forgotten ancient empire of the Hittites, plus the storied Byzantines, who held the armies of Islam at bay for nearly 1,000 years. Finally, he addresses the ethnogenesis of the Turkish-speaking population in Anatolia and its transformation from the eastern frontier of Greek speech to the western edge of the Turkic world.

Today, on the Unsupervised Learning podcast Razib talks to Erik Hoel, author of the novel The Revelations, and host of  The Intrinsic Perspective Substack. Hoel is a neuroscientist at Tufts who is interested in the problem of consciousness. Hoel admits right off that the questions and answers around consciousness motivate neuroscience in the first place, but throughout the conversation, he also points out that the discipline has a long way to go before it uncovers deep and insightful counterintuitive findings. In the early years of the 21st century, neuroscience was driven forward by amazing new technologies like functional magnetic resonance imagining (fMRI) that seemed to offer a window onto the brain’s activity, but over the last few years, most researchers agree that many of these papers did not live up to the hype (getting caught up in the replication crisis and underpowered studies).  Razib also talks to Hoel about his recent paper, The overfitted brain: dreams evolve to assist generalization, which argues that by “hallucinating out-of-distribution sensory stimulation every night, the brain is able to rescue the generalizability of its perceptual and cognitive abilities and increase task performance.” In plainer English, dreams allow the brain to experiment with novel possibilities outside of the range of experience and let it be more flexible and well-prepared in the face of surprising stimuli. Razib and Hoel also discuss his unique perspective as a humanist and a scientist. Hoel’s mother owned an independent bookstore, and he spent most of his childhood exploring its shelves. He reflects on how his Substack has grown (his piece The gossip trap won Scott Alexander’s book review contest), to the point where he wonders if perhaps in the next decade he will be more a writer who does some neuroscience than a neuroscientist who does some writing. Share

Jonathan Haidt is the author of The Coddling of the American Mind: How Good Intentions and Bad Ideas Are Setting Up a Generation for Failure, The Righteous Mind: Why Good People Are Divided by Politics and Religion and The Happiness Hypothesis: Finding Modern Truth in Ancient Wisdom. One of the pioneers of Moral Foundations Theory and a founder of Heterodox Academy, over the last few years Haidt has been focused on the impact of social media on our politics and culture (he is writing two books on the topic). Razib and Haidt begin their discussion with the blockbuster piece in The Atlantic, Why the Past 10 Years of American Life Have Been Uniquely Stupid. They both agree that in many ways the 1990’s and 2000’s were an information utopia, where the mind-opening possibilities of the internet were being realized. But Haidt lays out the case for social media, and more precisely functionalities like Twitter’s “quote-tweet” feature, having degraded online discourse, and driven offline polarization. He also argues that government and tech have to protect children from social media, making a case for enforced age restrictions on access. Razib presses Haidt on his theory about the “moral foundations” that differentiate liberals from conservatives. They discuss the possibility that ideological orientations may have been scrambled by the same processes that drove polarization in society more broadly with social media. Haidt also discusses his resignation from an academic society and the climate of intellectual conformity that is now seeping into every corner of the scholarly world.

On this monologue episode of Unsupervised Learning, Razib considers the different roles religion plays in various world civilizations. To explore this topic, he contrasts religion in the West (which includes Christendom and the Dar-al-Islam), on the Indian subcontinent and in China. Depending on which characteristics you focus on, these societies deploy and understand religion quite differently, even though religion as a cultural phenomenon is easily recognizable to all humans. Razib argues that in India religious identity is important both to the individual and essential to civilizational identity. In China, usually, a specific religious identity is neither essential to individual self-conception nor is a specific religion critical to civilizational identity. Finally, in European Christendom and Middle Eastern Dar-al-Islam religious identity holds importance for both the individual and society as in India. Still, that religious identity has a proselytizing aspect that imposes uniformity (unlike in India).

On this episode of Unsupervised Learning Razib discusses approaching politics through philosophy, political philosophy, and what it’s like being an excessively online academic in 2022 with Oliver Traldi. Currently working on a book on understanding politics through a philosophical lens, Traldi explains the relevance of epistemology to the project, while Razib queries the role that deductive, abductive and inductive reasoning might play in political views. Both also consider that political orientation is just a form of tribalism, as made clear when individuals chart a wholesale shift in a cluster of “beliefs” on topics as diverse as abortion and trade within just a few years. Traldi and Razib also discuss ancient political philosophy and its relevance to the modern era, as well as John Rawls and Robert Nozick, the two political philosophers most prominent in late 20th-century America. Traldi also mentions that American academia has developed a recent interest in Chinese and Indian philosophy, both of which have extensive areas of focus on politics. They cap their discussion of politics by discussing the role of intellectual movements like libertarianism on mainstream political parties like the American Republicans. Finally, Razib discusses Traldi’s experience of being a “very online” academic philosopher early in his career, and his contributions to various online publications, and how they relate to his scholarship.

Evolutionary psychology is a field that has made headlines ever since its inception as a distinct discipline in the 1980’s. In this episode of Unsupervised Learning, Razib talks to Dr. Tania Reynolds of the University of New Mexico, who researches intrasexual competition and cooperation, as well as sexual and social selection. Reynolds outlines what evolutionary psychology means for her and explains why she thinks it is helpful in our quest to understand human behavior. In particular, her field of research aims to understand how human females compete and cooperate, the psychological mechanisms driving their behavior, and how our overall evolutionary history informs this behavior (why is this behavior adaptive?). Razib and Reynolds then discuss how men and women have quite different psychologies on average and how that plays out in things as universal and important as friendships. Razib also asks whether it is essential to contextualize evolutionary tendencies within their broader social background. Suppose human males tend to compete through physical aggression. How does it play out in a society like ancient China, which denigrated martial values, as opposed to a post-Roman Europe ruled by illiterate warriors?  

How do we know when to trust the experts? On January 23rd, 2020, Vox published a piece titled The evidence on travel bans for diseases like coronavirus is clear: They don’t work. Journalists are largely limited to reporting what experts tell them, and in this case, it seems Vox's experts misled them. By December 2020 The New York Times could reflect that “interviews with more than two dozen experts show the policy of unobstructed travel was never based on hard science. It was a political decision, recast as health advice, which emerged after a plague outbreak in India in the 1990s.” The coronavirus pandemic has highlighted for many that expertise and specialized knowledge are not so straightforward, and “trusting the science” isn’t always straightforward, and hasty decisions can have global consequences. More narrowly, the political scientist Philip Tetlock’s 2005 Expert Political Judgment: How Good is it? How can We Know? reported that the most confident pundits often prove the least accurate. To get around the biases and limitations of individuals, there has been a recent vogue for “prediction markets” using distributed knowledge and baking “skin in the game” into the process. On this episode of Unsupervised Learning Richard Hanania joins Razib to discuss his think tank’s collaboration with UT Austin’s Salem Center for Policy and Manifold Markets on a forecasting tournament. What’s their goal? What are the limitations of these sorts of markets? Why do they not care about the contestants’ credentials? Razib pushes Hanania on the idea there is no expertise, and they discuss domains where the application of specialized knowledge has concrete consequences (civil engineering) as opposed to those where it does not (political and foreign policy forecasting). Hanania also addresses his decision to leave Twitter after his latest banning.

“Yankee go home!” has often been hurled at Americans indiscriminately. But the reality is that Yankee as a category initially meant the people of New England and its colonies across the northern fringe United States, from upstate New York to Minnesota. Yankees were a minority of Northerners during the American Civil War. Nevertheless, Yankee spearheading the Northern cause meant that Southerners disparaged all their occupiers with that label. This reflects the core insight that Yankees were, and arguably still are, far more influential in American culture and history than the raw weight of their numbers would indicate.  On this episode of The Unsupervised Learning podcast, Razib talks to Kerry of the Mary Lincolniana Substack about the role New England culture has played in shaping America and the world. A native New Englander, she does not flinch from asserting that in many fundamental ways, being American is a product of the norms and values of New England culture. Kerry argues the formative history of the colony of Massachusetts set the template for the later United States of America. Razib and Kerry discuss the possibility that the rise of a Southern elite counterculture was mainly a reaction to the preeminence of New England as an intellectual superpower in the early 19th century. They also explore the idea that America’s middle-class egalitarianism today reflects the aspirations of the founders of the New England colonies specifically, where the early focus was on literacy, communal debate, respect and rank accrued by those who attained erudition and learning. Kerry believes that the New England model of acculturating immigrants through a path of ascendance up the class hierarchy, starting with the Irish of the 1830’s, informed America’s later success with the mass migration of the late 19th and early 20th century. She also argues that 21st-century America could still learn much from the New England model of a well-educated and socially engaged populace.