MIT paper uses LLMs to measure ambiguity in laws and AI constitutions, EMNLP 2026 Oral
sebkrier · x · 2026-10-26
A paper co-led by Sally Dong and Luke Thorburn (MIT / AI & Democracy Foundation), accepted as an EMNLP 2026 Oral, proposes using LLMs to measure ambiguity in prescriptive texts like laws, public proposals, and even AI constitutions.
The motivation comes from the Habermas Machine project: LLMs often maximize group consensus simply by making statements more ambiguous — e.g. "peaceful protests should be allowed" scores high agreement only because disagreement hides in different readings of "peaceful".
The classic way to measure this is asking many people how they interpret a statement, which doesn't scale. The paper shows LLMs can approximate that process, enabling scalable ambiguity measurement.
More from AGI Musings
- Blogger claims tipping point crossed: next 6 months to beat past 60 years — SydSteyerhart · 2026-10-26
- tszzl on holding awe and dread at once: 'I don't know how else to live' — tszzl · 2026-10-26
- Mario creator Shigeru Miyamoto embraces generative AI: everyone can make games now — RachelVT42 · 2026-10-26
- Nassim Taleb: AI's impact hits hiring first, not layoffs — andreisavu · 2026-10-26
- Microsoft's Baym: computation is nowhere near brute-forcing biology — KevinKaichuang · 2026-10-26
- NYT op-ed: The AI boom looks a lot like the 2008 housing bubble — nordicinst · 2026-10-26