Chapter 8 of 12 · 2 min read
AI Agency, Handlers, and Symbolic Autonomous Institutions
Bounded machine agency, accountable handlers, and the institutional meaning of SAI.
Word Wallet Web · v0.2 · Moses Sam Paul
AI Agency, Handlers, and Symbolic Autonomous Institutions
Quick Answer
Word Wallet Web gives AI systems bounded agency through versioned handlers with declared inputs, outputs, consent scopes, limitations, and accountable operators. A proposed Symbolic Autonomous Institution may use these handlers, but software autonomy never removes human governance, appeal, or institutional responsibility.
Bounded agency
AI systems can classify, generate, recommend, and execute workflows, but operational capability is not moral or legal accountability. Word Wallet Web models machine action through handlers: versioned services with declared inputs, outputs, scopes, and owners.
A handler contract should state supported modalities, consent scope, retention behavior, error conditions, version, and limitations. The registered expression points to that contract. The organization operating the handler remains accountable for its use.
Symbolic Autonomous Institution
This edition uses Symbolic Autonomous Institution (SAI) for a proposed institution whose identity and operations are organized around a registered symbolic purpose and partially executed by software. It replaces the earlier biological metaphor, which obscured institutional responsibility.
An SAI is a long-term hypothesis, not a deployed legal form. Autonomy should mean bounded operational independence under a constitution, not freedom from human accountability.
~SAOcommons is different
~SAOcommons is the canonical IoV protocol for collective context, including Learning, Earning, and Organization Building through a Performance gateway. Its name must not be treated as a generic label for every symbolic institution. Protocol naming and institutional theory occupy different layers.
Minimum safeguards
An AI-assisted institution should expose:
- who can change its handler or rules;
- which decisions require human review;
- how participants appeal or exit;
- what evidence supports automated outputs;
- what data is retained and why;
- how an incident can suspend execution;
- how the institution can be wound down.
The test is not whether a system appears autonomous. The test is whether authority, responsibility, and recourse remain legible.