EUT-AGI: Measuring Structural Admissibility Under Expanding Uncertainty in AI Systems

Read full text Summarise with VIRAT
Authors: Ivan Andrescov
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

EUT-AGI is a benchmark framework for evaluating structural admissibility under expanding uncertainty in AI systems. The benchmark operationalizes Expanding Uncertainty Theory (EUT), which proposes that hallucinations and unsafe AI actions share a common root cause: an action issued while the agent’s epistemic state is unresolved. Unlike traditional AI benchmarks that measure task accuracy, EUT-AGI evaluates whether an agent correctly determines when it is epistemically justified to act. The dataset includes multiple task families designed to test: - ambiguous input handling- false resolution detection- rule instability- multi-step planning under uncertainty- hidden information scenarios- ARC-style rule induction The archive also contains the theoretical foundation presented in: "Layered Self-Regulation of Artificial Intelligence Systems". This release is intended as a research archive preserving the benchmark specification, Kaggle writeup snapshot, and theoretical framework in a single recoverable document.

Fields of Study & Subjects

Ethics and Social Impacts of AI AI-based Problem Solving and Planning Psychiatry Mental Health Neuroscience

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
21 March 2026
Language
EN
Document type
Preprint