Scaling Laws

Alan Rozenshtein, Research Director at Lawfare, spoke with Robert Wright—author of Nonzero, The Moral Animal, The Evolution of God, and Why Buddhism Is True, and the writer behind the NonZero Newsletter and podcast—about his new book, The God Test: Artificial Intelligence and Our Coming Cosmic Reckoning, which argues that AI is an evolutionary threshold on the scale of the entire history of life, that we are collectively failing to grasp its magnitude, and that rising to the challenge will require both new forms of international governance and an expansion of human moral and cognitive perspective. The conversation covered the multiple meanings of the book's title and what it means to view AI from a "cosmic" perspective; whether the public is finally starting to "feel the AGI" and where skepticism about AI's capabilities now comes from; how large language models are trained and Wright's claim that we have built "machines that create machines that think"; whether these systems genuinely understand, what Searle's Chinese Room and Nagel's "what is it like to be a bat?" have to do with it, and the open question of AI moral patienthood; the two families of AI risk—bad actors empowered by AI versus AI itself going rogue—and why the near-term disruption to jobs, relationships, and security may matter most; the "But China!" argument against AI regulation, China hawkishness, and why Wright thinks racing toward superintelligence is dangerously destabilizing; the case for "global governance" over "world government" and the perils of concentrating AI power at home; and why a book about AI and geopolitics closes with a call for mindfulness, cognitive empathy, and transcending the psychology of tribalism.

In this cross-pod episode, Alan and Kevin join Seth Benzell and Andrey Fradkin of Justified Posteriors to explore a big question: what should AI be for?The conversation begins with Pope Leo XIV’s recent encyclical. The group discusses how economists should think about the Church’s role in AI debates, what counts as an AI-related market failure, whether moral and religious institutions can help address social harms, and whether such interventions risk crowding out private action or local experimentation.The episode then turns to the emerging idea of positive alignment. A recent paper, Positive Alignment: Artificial Intelligence for Human Flourishing, argues that AI alignment has focused too heavily on negative alignment—preventing harms such as manipulation, bias, dangerous outputs, and misuse—and should also ask how AI systems can actively support autonomy, wisdom, truth-seeking, pluralism, and human flourishing.

Shane Tews, host of Explain to Shane and nonresident senior fellow at AEI, joins Kevin Frazier, director of the AI Innovation and Law Program at the University of Texas School of Law and a senior fellow at the Abundance Institute, for a cross-post conversation about the AI and cyber executive order, workforce disruption, and the future of education. They also share their respective research agendas for the summer.

Janel Thamkul, former frontier counsel team member at Anthropic, joins Kevin Frazier to discuss what it means to practice law at the frontier of AI.This episode starts with a review of Janel’s fascinating and varied background. Next, she walks through her initial exploration of a career in art before eventually pivoting to the law based on some very formative experiences. Kevin and Janel then investigate some of the most pressing and open questions related to transformative AI.

Alan Rozenshtein, Research Director at Lawfare and Visiting Senior Fellow at the Institute for Law & AI (LawAI), spoke with Christoph Winter, LawAI Founding Director and Assistant Professor of Law and AI at the University of Cambridge, and LawAI Senior Research Fellow Charlie Bullock, about their new paper "Radical Optionality: Governing Transformative AI Under Uncertainty," which argues that, given the possibility of transformative AI within the next decade and deep uncertainty about its capabilities and risks, governments should aggressively build the institutional capacity to regulate competently when needed, rather than either deferring to the market or locking in premature substantive rules. The conversation covered the four foundational assumptions underlying the paper and what makes the optionality "radical"; the difficulty of regulating an exponentially improving and poorly understood technology and what it means to "feel the AGI"; why a pure permissionless-innovation approach breaks down once the national-security implications of transformative AI come into view; why the European precautionary approach risks regulating without the expertise to enforce; the centrality of hiring and talent and what an adequately funded U.S. counterpart to the UK AI Security Institute would look like; the concrete work that such an agency would do, including evaluations, standard-setting, and procurement-side cybersecurity requirements modeled on CMMC; the importance of building international information-sharing channels among liberal democracies before they are urgently needed; and the case against broad federal preemption of state AI laws before any federal regulatory framework exists.

Tom Davidson joins Kevin Frazier, AI Innovation and Law Fellow at the University of Texas School of Law and a Senior Editor at Lawfare, to discuss AI timelines, explosive economic growth, and the increasingly urgent debate over “AI character” — the behavioral traits and decision-making tendencies embedded into advanced AI systems.Drawing on Davidson’s recent paper, “The Importance of AI Character,” which he co-authored with Will MacAskill, their conversation explores how the character of future AI systems may influence democratic governance, military conflict, institutional trust, and even the long-run trajectory of civilization. The discussion also examines the key influences on character development and which actors should ultimately play a part in dictating the default values and behaviors of AI models. You may also enjoy Tom's article on AI as advisors.

Owen Larter, Senior Director and Head of Frontier Policy and Public Affairs at Google DeepMind, joins Kevin Frazier, AI Innovation and Law Fellow at the University of Texas School of Law and a Senior Editor at Lawfare, to provide an inside look at how DeepMind approaches frontier governance. The conversation moves beyond the familiar U.S.-EU-China framing of AI policy to examine international coordination after the recent U.S.-China summit, Google DeepMind’s national AI partnerships, the role of the Frontier Model Forum, and the challenge of expanding AI adoption. Kevin and Owen also discuss policy formation inside frontier AI companies. They close with an examination of the need to build a deeper AI policy talent pipeline.

Melissa Hutchins, founder and CEO of Certifi AI, joins Kevin Frazier, AI Innovation and Law Fellow at the University of Texas School of Law and a Senior Editor at Lawfare, to discuss the rise of deepfakes, non-consensual sexually explicit imagery, and the growing policy fight over AI-generated harms online.Drawing from both her professional work and her personal experience as a victim of cyberstalking, Melissa explains how synthetic media is changing the threat landscape for individuals, platforms, and policymakers alike. The duo also unpack proposals like the Take It Down Act, the challenges posed by a fragmented patchwork of state AI laws, and what it’s like building an AI company from Seattle rather than Silicon Valley.

Adeel Khan of MagicSchool AI and Ryan Trattner of StudyFetch join Kevin Frazier, AI Innovation and Law Fellow at the University of Texas School of Law and Senior Editor at Lawfare, to discuss the rapid integration of artificial intelligence into education.The conversation explores how AI tools are being used by both teachers and students, from automating lesson planning to providing personalized tutoring and study support. The group examines claims about improved learning outcomes and time savings, while probing what counts as meaningful evidence versus early-stage or self-reported metrics.They also discuss the regulatory and operational challenges of building AI systems in education, including constraints imposed by student data protections, limited access to high-quality training data, and the growing impact of regulations and public scrutiny on product development and deployment.