Scaling Laws Lawfare Daily: Why AI Won’t Revolutionize Law (At Least Not Yet), with Arvind Narayanan and Justin Curl
May 5, 2026
Justin Curl, a Harvard J.D. student focused on legal policy and access to justice, and Arvind Narayanan, Princeton CS professor who studies privacy and societal impacts of computing, discuss why AI’s technical prowess may not cut legal costs fast. They cover the slow diffusion of tech, law as a credence good, regulatory and adversarial bottlenecks, unauthorized practice rules, arms‑race dynamics, and proposed regulatory sandboxes.
AI Snips
Chapters
Books
Transcript
Episode notes
AI Diffuses Slowly Through Four Stages
- AI effects unfold through four stages: capability gains, product translation, worker adoption, and deep organizational/legal change.
- Narayanan compares AI to electricity, noting real benefits required decades of factory redesign and institutional shifts to appear.
Three Structural Reasons Law Is Costly
- Law is expensive because it's a credence good, often relative in value, and tightly regulated by professional rules.
- Justin cites Jillian Hadfield: clients often can't evaluate quality and rely on reputation or credentials instead.
UPL Rules Create A Legal Barrier For AI
- Unauthorized practice of law (UPL) rules broadly forbid applying legal knowledge without a license and can criminalize non-lawyer assistance.
- Justin warns chatbots offering tailored legal advice risk running into UPL liability in many jurisdictions.
