Tests show agents pull group policy toward users' preferences, 78% accuracy self-report
jacyanthis · x · 2026-11-15
Responding to a question about verifying whether agents represent people's views accurately, jacyanthis describes tests using a budget policy — a percentage breakdown across 4 AI safety priorities — checking whether agents pull the final group policy toward their human's preferred policy. They do. By self-report, 78% of participants say their agent represented their opinions accurately. The work relates to using agents to aggregate citizen preferences, in discussion connected to Google DeepMind.
Related event: DeepMind's Pluralistic Agents Scale Public Input on AI Policy(5 posts)→
More from Safety
- Lean community certifies Riemann proof; honeypot fake keys expose breaches at three AI providers — LeviTurk · 2026-11-22
- Intruder swapped Prime Intellect's hosted DeepSeek for an abliterated copy for 53 hours — mawcopolow · 2026-11-22
- Dragos finds three hacked AI hosts whose operators had no idea they were breached — mawcopolow · 2026-11-22
- Who do you sue when an AI agent errs? The "sovereign agents" liability debate — QuixiAI · 2026-11-22
- Opinion: without provenance, 'ethical AI' is just a slide deck — AryHHAry · 2026-11-22
- OpenAI board member Christiano puts AI takeover risk at ~10% within a year — TheAIGRID · 2026-11-22