Symbolic Governance for Probabilistic Intelligence: Why Governable AI Requires Symbolic Structure Without Reverting to Symbolic AI

This publication presents a canonical architectural framework for governing probabilistic AI systems in regulated environments. Modern AI systems—large language models, autonomous agents, and tool-calling workflows—operate through statistical inference and cannot provide the deterministic guarantees required for authorization, auditability, and legal accountability. As a result, existing AI “governance” approaches based on alignment, guardrails, human-in-the-loop review, or provenance tracking fail to provide enforceable control over consequential actions. The paper argues that governable AI requires a strict architectural separation between intelligence and governance. Neural systems generate proposals (recommendations, hypotheses, candidate actions), while authorization must occur through an independent, deterministic verification layer. This layer evaluates proposed actions against explicit policy constraints and permits execution only when authorization can be mechanically proven. Probabilistic compliance is rejected as a governance strategy; authorization is treated as a binary, fail-closed decision. The core contribution is the introduction of Proof-Carrying Decisions (PCDs): structured decision artifacts that carry all evidence necessary to independently verify that an action was authorized at execution time. PCDs enable replayable verification, allowing auditors, regulators, and investigators to mechanically reconstruct authorization decisions without relying on post-hoc narratives or probabilistic model behavior. The paper defines the deterministic envelope that enforces “no proof, no action,” specifies decision compilation requirements, establishes a replay contract, and formalizes escalation and bounded human discretion. The work clarifies the relationship between governance and provenance, demonstrating that provenance systems provide evidentiary inputs but cannot substitute for authorization. Provenance enables forensic reconstruction; governance enforces permission prior to execution. The paper rejects claims that traceability alone constitutes governance and provides a formal dependency model in which provenance feeds evidence, evidence feeds constraint evaluation, and verification produces authorization. This publication is not a return to symbolic AI as a theory of intelligence. Instead, it demonstrates that symbolic structure—determinism, inspectability, and verifiability—is essential for governance even as probabilistic models remain indispensable for intelligence. The framework is model-agnostic, vendor-neutral, and applicable across regulated domains including healthcare, finance, and public-sector systems. This document is intended as a stable, citable technical publication and canonical reference for deterministic AI governance architectures.

Ҳали таржима қилинмаган