Abstract

This paper specifies a written constitution for the human oversight body of an autonomous AI system: the board that holds a human override, an authority to stop or correct the system that narrows over time and is never removed. Proposals for corrigible AI commonly reserve such an override and leave the body that holds it unconstituted. The constitution answers five ways an oversight body can be captured (selection, funding, membership conflicts, emergency powers and founder succession), and most of its procedure is taken from the Vinaya Piṭaka, the Buddhist monastic code. Four chambers (monastic men, renunciant women, laymen and laywomen) must all be occupied for the gravest acts; lay chambers hold the cryptographic key-shares and renunciant chambers the doctrinal veto. A candidate enters through a sponsor, a fixed public interrogation whose first question is a proof-of-personhood check that also bars AI agents from seats, a seating vote of the body, and probation. Source communities nominate; oversubscribed pools are decided by lot under publicly verifiable randomness and an appointed draw-officer; and the overseen AI never nominates, draws or seats. Support to overseers' communities cannot be modulated by any oversight outcome; membership ends automatically on predefined acts; and every member, and the AI, faces an annual invited-criticism audit whose agenda reports what is provided, never what is attained. The founder seats the first members once, after which admission authority passes to the body permanently. The body, here called the Aquarian Sangha, does not yet exist.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS