Guardado en:
| Autor principal: | |
|---|---|
| Formato: | Recurso digital |
| Lenguaje: | inglés |
| Publicado: |
Zenodo
2026
|
| Materias: | |
| Acceso en línea: | https://doi.org/10.5281/zenodo.18121500 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
Tabla de Contenidos:
- <p>We present Structural Self-Governance (SSG): a foundations-level architecture for adaptive<br>and self-modifying intelligent systems operating under uncertainty and adversarial pressure. Intelligence scales not through eliminating error, but through relocating error into reversible internal cognition while irreversible state transitions are governed by a non-bypassable Safety Kernel. We formalize the Constrained Commitment Principle, introduce Risk Debt as a persistent authorization resource, and prove a reachability theorem showing unsafe states are unreachable regardless of internal cognition accuracy, calibration, or truthfulness—explicitly tolerating hallucinations so long as commitment remains constrained. We extend the theory with missing fundamentals required for future AGI: action conservatism under uncertainty, epistemic asymmetry against self-reference exploits, evolutionary stability, composable governance, and irreducible uncertainty limits. Finally, we provide a completeness layer and RFC-style normative specification defining SAM (Sensitive Asset Map), invariant coverage obligations, token semantics with execution-time predicate binding, anti-laundering policies, estimator constraints, audit/replay requirements, and conformance criteria.</p>