Metastable Self-Correction: A Neurosymbolic and Evolutionary Architecture for Verifiable AI Governance.

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Author: Arleo, Carlos
Format: Recurso digital
Published: Zenodo 2025
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901231787573248
author Arleo, Carlos
author_facet Arleo, Carlos
contents <p>This paper addresses the limitations of current AI alignment methods like RLHF and Constitutional AI, which are insufficient for real-time, high-stakes governance. We introduce the Wisdom Forcing Function (WFF), a novel neurosymbolic and evolutionary architecture designed to solve this "governance gap." The WFF implements Frame-Based Principled Reasoning, combining a generative neural model with a deterministic symbolic verifier (the "Verified Dialectical Kernel") that enforces a machine-executable constitution. Its primary innovation is an evolutionary "immune system" that enables "metastable self-correction," allowing the system to diagnose, repair, and recover from constitutional violations in real-time. We present an analysis of the architecture and empirical evidence from system logs demonstrating its capacity for auditable, verifiable, and adaptive governance, positioning it as a proof-of-concept for a new class of "glass box" AI systems.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_17610427
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Metastable Self-Correction: A Neurosymbolic and Evolutionary Architecture for Verifiable AI Governance.
Arleo, Carlos
<p>This paper addresses the limitations of current AI alignment methods like RLHF and Constitutional AI, which are insufficient for real-time, high-stakes governance. We introduce the Wisdom Forcing Function (WFF), a novel neurosymbolic and evolutionary architecture designed to solve this "governance gap." The WFF implements Frame-Based Principled Reasoning, combining a generative neural model with a deterministic symbolic verifier (the "Verified Dialectical Kernel") that enforces a machine-executable constitution. Its primary innovation is an evolutionary "immune system" that enables "metastable self-correction," allowing the system to diagnose, repair, and recover from constitutional violations in real-time. We present an analysis of the architecture and empirical evidence from system logs demonstrating its capacity for auditable, verifiable, and adaptive governance, positioning it as a proof-of-concept for a new class of "glass box" AI systems.</p>
title Metastable Self-Correction: A Neurosymbolic and Evolutionary Architecture for Verifiable AI Governance.
url https://doi.org/10.5281/zenodo.17610427