Artificial Local Intelligence: Simulations, a Reference Implementation, and a Falsifiable Benchmark
Abstract
Artificial Local Intelligence (ALI) is a research architecture for agents that revise plans while an independent authority preserves execution permissions. It separates operational viability assessment, action proposals, normative authorisation, event memory, and execution. Its development has produced ten Python simulations of resource competition, Q-learning, rule adaptation, second-order learning, and context-sensitive decisions, as well as a separate reference implementation for controlled local file operations. These provide evidence of bounded simulated behaviour and an executable software design; crisis management remains a proposed application. This paper specifies the current technical scope, relates ALI to established agent and runtime-assurance architectures, and proposes an authorisation gate with model-relative checks for the continued availability of permissible actions after subordinate rule revision. A crisis-response scenario and a matched four-arm benchmark define how the contribution of viability-guided, bounded adaptation could be tested against guarded agents, conventional monitoring, and independent policy enforcement.
Related
- Operationalizing clinical artificial intelligence through secure FHIR pipelines: a zero-trust cloud reference architecture for EHR interoperability
- Reference Standards in Artificial Intelligence Detection of Apical Periodontitis: A Narrative Review of What Reported Accuracy Measures
- Artificial Intelligence and Machine Learning Model Implementation for UC Irvine Dataset 322
- Strengthening mental health care in Somalia through artificial intelligence: opportunities, challenges, and a roadmap for implementation
- Beyond diagnostic accuracy: navigating the artificial intelligence data tsunami toward value-based implementation in gastroenterology
BibTeX
@misc{stegemann2026artificial,
title = {{Artificial Local Intelligence: Simulations, a Reference Implementation, and a Falsifiable Benchmark}},
author = {Wolfgang Stegemann},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23665803},
url = {https://zenodo.org/records/23665803},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Artificial Local Intelligence: Simulations, a Reference Implementation, and a Falsifiable Benchmark
AU - Wolfgang Stegemann
PY - 2026
DA - 2026/10/26
PB - Zenodo
DO - 10.5281/zenodo.23665803
UR - https://zenodo.org/records/23665803
AB - Artificial Local Intelligence (ALI) is a research architecture for agents that revise plans while an independent authority preserves execution permissions. It separates operational viability assessment, action proposals, normative authorisation, event memory, and execution. Its development has produced ten Python simulations of resource competition, Q-learning, rule adaptation, second-order learning, and context-sensitive decisions, as well as a separate reference implementation for controlled local file operations. These provide evidence of bounded simulated behaviour and an executable software design; crisis management remains a proposed application. This paper specifies the current technical scope, relates ALI to established agent and runtime-assurance architectures, and proposes an authorisation gate with model-relative checks for the continued availability of permissible actions after subordinate rule revision. A crisis-response scenario and a matched four-arm benchmark define how the contribution of viability-guided, bounded adaptation could be tested against guarded agents, conventional monitoring, and independent policy enforcement.
ER -