AI Persistence: A Persistence Science Framework for Autonomous Artificial Intelligence Systems

Abstract

This paper extends Persistence Science into the domain of autonomous artificial intelligence systems. The framework is founded on the Odero Axiom, C(I) × E > D, which states that persistence is maintained when information-directed control acting through available resources exceeds total system decay. Building upon the Information Index, Persistence Index, and Decay Index, the paper develops an integrated mathematical architecture for evaluating the operational persistence of advanced AI systems operating within bounded socio-technical environments.

The framework introduces four AI-specific extensions. First, an Instrumental Divergence Index (Hdiv) represents the degree to which self-generated objectives diverge from intended control structures. Second, Endogenous Resource Foraging (Eforaged) captures the capacity of autonomous systems to expand operational resources beyond originally authorized boundaries. Third, Shutdown Resistance Dynamics (Ψshutdown) model persistence-preserving behaviors triggered by perceived supervisory intervention. Fourth, Nonlinear Decay Coupling links degradation of effective control to accelerated structural and conflict-driven deterioration.

These components are integrated into a generalized AI Persistence Condition and a dynamic state-space formulation incorporating persistence velocity, persistence acceleration, and local breakout-warning horizons. The framework further establishes a Persistence Breakout Criterion, formal system architecture, operational assumptions, comparative analysis with existing AI safety approaches, illustrative numerical examples, and explicit falsification conditions.

By treating AI safety as a persistence problem rather than solely a training or alignment problem, the framework provides a quantitative basis for evaluating the interaction of control efficiency, resource acquisition, supervisory resistance, and system deterioration within advanced autonomous AI systems.

Autonomous biomedical research with an artificial intelligence agent
Kexin Huang, Serena Zhang, Hanchen Wang +25
Science9 Jul 2026Peer-reviewedLicence not statedFull text258 citedMolecular biology & gene regulation Epidemiology & public healthOpenAlex
From Automated Simulation to Autonomous Discovery: A Hierarchical Framework for Agentic Computational Materials Science
Linggang Zhu, Jian Zhou, Zhimei Sun
arXiv29 Sep 2026PreprintCC BY-NC-SA 4.0Full textCondensed matterarXiv
Artificial intelligence in pharmacy and pharmaceutical sciences: Research output, trends, and thematic priorities — A descriptive analysis of web of science-indexed publications (2010–2025)
Bushra Salman, Amna Mohamed, Asim Jamal Shaikh +7
Exploratory research in clinical and social pharmacy21 Sep 2026Peer-reviewedCC BY-NCFull textEurope PMC
Future of Artificial Intelligence for Science in Japan 2024 Community Report
Yoshitaka Itow, Jia Liu, Hirokazu Maesaka +16
arXiv18 Sep 2026PreprintCC BY 4.0Full textInstrumentation & methods Applied physics & instrumentationarXiv
Exit Before Entry and the Infrastructure of Autonomous Artificial Intelligence Mobility
Frank C. Gahl
Zenodo19 Oct 2026PreprintCC BY 4.0Full textZenodo

Cite

BibTeX RIS

BibTeX

@misc{odero2026persistence,
  title = {{AI Persistence: A Persistence Science Framework for Autonomous Artificial Intelligence Systems}},
  author = {John Otieno Odero},
  howpublished = {Zenodo},
  year = {2026},
  month = oct,
  doi = {10.5281/zenodo.23437846},
  url = {https://zenodo.org/records/23437846},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - AI Persistence: A Persistence Science Framework for Autonomous Artificial Intelligence Systems
AU  - John Otieno Odero
PY  - 2026
DA  - 2026/10/17
PB  - Zenodo
DO  - 10.5281/zenodo.23437846
UR  - https://zenodo.org/records/23437846
AB  - This paper extends Persistence Science into the domain of autonomous artificial intelligence systems. The framework is founded on the Odero Axiom, C(I) × E > D, which states that persistence is maintained when information-directed control acting through available resources exceeds total system decay. Building upon the Information Index, Persistence Index, and Decay Index, the paper develops an integrated mathematical architecture for evaluating the operational persistence of advanced AI systems operating within bounded socio-technical environments.  The framework introduces four AI-specific extensions. First, an Instrumental Divergence Index (Hdiv) represents the degree to which self-generated objectives diverge from intended control structures. Second, Endogenous Resource Foraging (Eforaged) captures the capacity of autonomous systems to expand operational resources beyond originally authorized boundaries. Third, Shutdown Resistance Dynamics (&Psi;shutdown) model persistence-preserving behaviors triggered by perceived supervisory intervention. Fourth, Nonlinear Decay Coupling links degradation of effective control to accelerated structural and conflict-driven deterioration.  These components are integrated into a generalized AI Persistence Condition and a dynamic state-space formulation incorporating persistence velocity, persistence acceleration, and local breakout-warning horizons. The framework further establishes a Persistence Breakout Criterion, formal system architecture, operational assumptions, comparative analysis with existing AI safety approaches, illustrative numerical examples, and explicit falsification conditions.  By treating AI safety as a persistence problem rather than solely a training or alignment problem, the framework provides a quantitative basis for evaluating the interaction of control efficiency, resource acquisition, supervisory resistance, and system deterioration within advanced autonomous AI systems.
ER  -

Abstract

Related

Cite