What if you had to write a contract without knowing which side you'd end up on? Martin Lapotka (PhD ML, Mozilla alum) applied John Rawls' veil of ignorance to…
Notes & highlights
Published: · 15:54
What if you had to write a contract without knowing which side you'd end up on?
Martin Lapotka (PhD ML, Mozilla alum) applied John Rawls' veil of ignorance to AI-driven legal contract negotiation — and it works.
The architecture: a parser extracts facts from both parties. Two reviewer agents see only the facts, NOT which party they represent. An arbiter agent iterates. 400–500 simulations per contract using open-source models on a 2013 MacBook Pro. Results stable to gender and name permutations. Three practising lawyers gave it a pass.
The intelligence isn't in the LLM. It's in the design of information flows between agents.
Peer-reviewed paper at the First Agentics Conference (October 2025, Marbella). Open-source. Fork it.
————
💥 Join the movement
Join BC + AI: https://bc-ai.ca/membership/
Next Vancouver AI meetup: https://luma.com/vancouver-ai
Smash that Subscribe & 🔔 so you don't miss future talks.
Drop a comment: Would you trust an AI to negotiate a contract for you?
Share the vid with a lawyer or legal tech enthusiast.