Scaling Laws

Hélène Landemore, Professor of Political Science at Yale University and a fellow at Oxford’s Institute for Ethics in AI, joins Kevin Frazier, director of the AI Innovation and Law Program at the University of Texas School of Law and a Senior Editor at Lawfare, to examine who should decide what AI is for. Starting with Landemore’s recent Noema essay, the pair discuss why decisions about AI’s purpose, limits, and governance should not be left solely to elected officials, technical experts, and the companies building the systems.

Landemore explains some key ideas--covering the fundamentals of deliberative democracy, sortition, and citizens’ assemblies. They then explore her proposal for global public consultation and deliberation. The duo conclude with hard questions about representation, expertise, and platform power.
Hosted on Acast. See acast.com/privacy for more information.

Dave Willner, co-founder of Zentropi, and Vaishnavi J., founder and principal of Vys, join 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, to examine how AI is affecting younger users and why existing child-safety responses often default to age gates, parental consent, and outright bans.

The conversation then turns to their open-source teen self-harm and suicidal ideation policies, which use age-banded rules and a four-zone framework: remove, suppress with a warning, suppress from recommendations, and allow. Dave and Vaishnavi explain how the policies were developed using more than 1,200 hand-labeled examples, how Zentropi’s CoPE classifier translates written policy into enforcement, and how platforms should evaluate false positives, false negatives, over-refusal, under-refusal, human review, and appeals.

Dave's Tech Policy Press piece.
More on their classifer and research.

Explainer blog - Overall: https://quire.substack.com/p/what-it-takes-for-llms-to-make-age
Explainer blog - Part 1: https://quire.substack.com/p/f4b99ab0-0127-4992-a31d-67b989f2cf58
Explainer blog - Part 2: https://quire.substack.com/p/ae6a5436-21e5-4fec-81fe-8025004f3b16
Full set of policies: https://zentropi.ai/u/vys
Hosted on Acast. See acast.com/privacy for more information.

Chloé Bakalar, ASI Ethics Lead at Responsible Superintelligence Institute (RESI) and a Senior Research Associate at University College London, joins Kevin Frazier, Director of the AI Innovation and Law Program at the University of Texas School of Law and a Senior Editor at Lawfare, for a Scaling Laws Roadshow conversation recorded at Columbia Law School. Big thanks to Eliza Steffen for organizing a fantastic visit.

Recorded on the eve of Constitution Day, Chloé and Kevin explore the purposes of political constitutions, the American experience with constitutional change, and debate the extent to which the U.S. Constitution constrains private power. They then turn to AI constitutions: who writes them, whose values they reflect, how those principles are interpreted and revised, and how we can tell whether they meaningfully constrain an AI system.

The conversation closes by testing a broader proposal: whether governing advanced AI may require a dual constitutional structure—one governing the model itself and another governing the companies that develop and deploy it.
Hosted on Acast. See acast.com/privacy for more information.

Kenji Yoshino, Chief Justice Earl Warren Professor of Constitutional Law at New York University School of Law and a member of Meta’s Oversight Board, joins Kevin Frazier, director of the AI Innovation and Law Program at the University of Texas School of Law, and guest co-host Kate Klonick to discuss what frontier AI companies can learn from the Oversight Board. Yoshino recently co-authored a Tech Policy Press piece with fellow Board member Ronaldo Lemos arguing that meaningful AI oversight should feature independent and representative overseers, external standards, and transparent, reasoned decisions.

The trio tests that proposal against some of the hardest questions raised by the Oversight Board’s own experience. Can a selective, precedent-setting institution provide meaningful oversight at AI scale? What gives a privately created board legitimacy to constrain decisions with global consequences? What powers would an AI oversight body actually need? And when does private oversight complement democratic regulation—and when might it merely give corporate power a new source of legitimacy?

Here’s the Tech Policy Press piece.
Hosted on Acast. See acast.com/privacy for more information.

Christina Lee, Visiting Associate Professor of Law and Privacy and Technology Law Fellow at George Washington University Law School, 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, from the Institute of Law and AI’s Workshop on Law-Following AI in Cambridge, UK.

They discuss her forthcoming article, Beyond Algorithmic Disgorgement: Remedying Algorithmic Harms.

The conversation begins with the basics: what is algorithmic disgorgement, why did the FTC first use it in the Cambridge Analytica case, and how has the remedy evolved from data-deletion-plus-model-deletion orders into a much broader candidate remedy for AI harms? Christina explains why the remedy’s expansion matters, especially after the FTC’s Rite Aid settlement, where the alleged wrong was not unlawful data collection but the allegedly unfair use of facial recognition technology.

They also discuss alternative remedies, including injunctions, post-processing, fine-tuning, machine unlearning, and training-data attribution. Finally, they wrap up with Christina’s professional journey and the future of legal education in the age of AI.
Hosted on Acast. See acast.com/privacy for more information.