AI & The Law: Changing Practice, Claude Constitution, & New Rights, w/ Kevin & Alan of Scaling Laws

Episode Summary

Executive Summary: Kevin Frazier and Alan Rosenstein argue that AI is already transforming law as both a subject of regulation and a tool that will reshape legal work. They foresee better access to legal services, major pressure on junior work, AI-assisted judges and legislation, and new rights around compute and data, while warning that AI could also supercharge executive power and surveillance.

Main Topics: AI is transforming both the law and the legal profession (Priority: 5/5): The guests distinguish between the law of AI and AI in the law: regulation of AI systems versus AI’s direct impact on legal reasoning, research, drafting, and practice. They frame the field as moving quickly despite legacy legal structures. Frontier models are already competitive with lawyers (Priority: 5/5): They argue that modern frontier models are already better than the median lawyer in raw cognitive tasks, though still jagged and unreliable in some niche or high-stakes contexts. Legal adoption is slowed more by institutions than by capability. Demand expansion vs. displacement in legal services (Priority: 5/5): A central debate is whether AI will reduce lawyer demand or create Jevons-paradox-style expansion by making legal help cheaper and more accessible. They expect entry-level work to shrink, but total legal services may rise if latent demand is unlocked. Reimagining contracts, legislation, and adjudication (Priority: 4/5): They propose complete contingent contracts, outcome-oriented laws, AI-based simulations of proposed bills, and more principle-driven or contextual judging. AI could move law from procedural overload toward system design and better policy evaluation. Legal education and the future of junior legal work (Priority: 4/5): They expect routine paralegal and junior associate tasks to be automated, forcing law schools to adapt. Future lawyers may split between rote service delivery and higher-level 'legal architect' roles focused on regulatory design. New rights and limits in the AI era (Priority: 5/5): They discuss a right to compute, a right to share personal data, and restrictions on government AI use, especially surveillance. AI may also intensify debates over model welfare, sentience, and human attachment to AI personas.

Key Arguments: AI is already better than the median lawyer at many cognitive tasks, even if it remains jagged and unreliable in edge cases. Law is slower to change than software because it is a professional guild, but the underlying work is still symbolic and therefore highly automatable. Lower-cost legal services could trigger Jevons-paradox effects, increasing total demand rather than simply eliminating lawyers. There is substantial unmet legal need in 'legal deserts,' so AI may expand access rather than merely displace existing work. Junior legal work is the most vulnerable because discovery, summarization, and contract review are highly routinized. The legal field may bifurcate into routine service providers and higher-level 'legal architects' who design systems and regulation. Future legislation should be outcome-oriented and tested through simulation before passage. Judging and AI alignment both require a mix of rules and principles; neither pure formalism nor pure realism is sufficient. AI could increase state capacity and make government more effective, but it could also massively expand executive control and surveillance. A right to compute and a right to share personal data would help ordinary people benefit from AI, especially in education and health contexts.

Data Points: Top law firms using Harvey: 70% - Kevin says about 70% of top U.S. law firms have licensed Harvey, though day-to-day usage remains low. Legal services market concentration: about 1 lawyer per 1,000 residents - Kevin cites 'legal deserts' in many U.S. regions where people have very limited access to counsel. State courts share of litigation: 95% - Kevin notes that roughly 95% of all litigation occurs in state courts, where inefficiency is common. AI model performance against lawyers: Claude Opus 4.5 wins 1/3 and wins or ties 70% - Nathan cites a public eval where Claude Opus 4.5 outperforms or matches human lawyers in head-to-head comparisons. Technology adoption horizon in law schools: within 1-2 years, all law schools likely to have AI classes - Alan predicts AI coursework will become universal across law schools soon. AI rights legislation: Montana enacted a right to compute - Alan mentions Montana as an early adopter, with Ohio and New Hampshire considering similar bills. Industry adoption statistic: 70% of major top 100 law firms using Harvey - Kevin relays Harvey’s own reported adoption numbers, while noting actual usage is still shallow. Government data/compute policy: FERPA burdens and annual sign-offs - Kevin argues privacy rules can make it hard for parents or students to share educational data with AI providers.

Pivotal Quotes: "“The practice of law is fundamentally a cognitive activity.”" — Alan Rosenstein: Used to explain why frontier models should increasingly affect legal work, despite the profession’s guild protections. "“We need more Jillians hanging around.”" — Kevin Frazier: He argues for a new generation of legal scholars and designers who use AI to rethink regulatory design and legislation. "“What if we start to move towards outcome-based law?”" — Kevin Frazier: His vision for laws evaluated by measurable social outcomes rather than procedural complexity.

Implications: Listeners should expect AI to reshape legal services, law schools, and government faster than current institutions are ready for. The biggest stakes are access to justice, executive power, and whether society can use AI to improve law without creating new surveillance or dependency risks.