عندما يكون من غير الممكن تحديد "الحكم الجيد مقابل الحكم السيء"

Key Points

• The aim is to investigate when governance can be classified as good or bad based on limited observational data.
• Formalized governance unidentifiability within observable-only agents.
• Defined observational-equivalence classes over mediation mechanisms.
• Presented an impossibility theorem related to necessity-evaluators.
• Introduced non-arbitrariness requirements linked to contestability and control-domain independence.
• Demonstrated a scenario where no feasible policy can ensure a positive robust lower bound.
• Established minimal structural counterexamples illustrating governance challenges.
• Outlined boundaries to manage risks associated with dual-use scenarios.