Adversarial, Recursive, Generalized Uncertainty System: A falsifiable metascientific architecture for adversarial claim evaluation, causal auditing, evidence provenance, uncertainty classification, and recursive error correction
Journal of Artificial Intelligence General science (JAIGS) ISSN:3006-4023 · 2026
DOI: 10.60087/jaigs.v10i01.502 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Artificial Intelligence General science (JAIGS) ISSN:3006-4023, 2026. Verfasst von Andris Lukss. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
AbstractARGUS: An Adversarial, Recursive, Generalized Uncertainty System for Scientific Claim EvaluationScientific inference remains vulnerable to model misspecification, hidden assumptions, dependent evidence, measurement error, causal misidentification, selective analysis, evaluator bias, and premature convergence on favored explanations. This article introduces ARGUS (Adversarial, Recursive, Generalized Uncertainty System), a proposed metascientific framework developed by Andris Lukss for systematic adversarial evaluation of scientific claims. ARGUS operationalizes scientific criticism as a recursive cycle of claim decomposition, operationalization, alternative-hypothesis generation, discriminatory prediction, adversarial testing (MAX-KILL), provenance and dependency auditing, causal analysis, verification, replication, uncertainty classification, scope restriction, and subsequent retesting. A central design principle is that survival of falsification attempts does not constitute proof. ARGUS therefore distinguishes SURVIVES, REJECTED, and UNRESOLVED claims while introducing MODEL_GAP for cases in which observations are inadequately explained by the candidate model space rather than forcing selection of the least-inadequate available model. Evidence quality is represented through typed defects rather than a single scalar score, and causal claims require explicit identification assumptions and, where applicable, edge-level intervention testing. The evaluator and ARGUS itself are additionally treated as potential sources of error, preventing the framework from receiving methodological immunity. To make ARGUS empirically falsifiable, we propose prospective, blinded benchmark comparisons against simpler scientific-evaluation procedures using seeded inferential defects, hidden confounding, evidence dependencies, omitted-model cases, genuine weak effects, and adversarial tests designed to detect excessive skepticism. ARGUS is therefore presented not as an established replacement for existing scientific methodology, but as a testable error-detection architecture whose scientific value depends on demonstrating improved calibration and defect detection under independent prospective validation.Keywords: ARGUS; metascience; falsification; scientific methodology; adversarial testing; causal inference; reproducibility; replication; uncertainty; model misspecification; evidence provenance; MAX-KILL.
Zitieren
APA
Lukss, A. (2026). Adversarial, Recursive, Generalized Uncertainty System: A falsifiable metascientific architecture for adversarial claim evaluation, causal auditing, evidence provenance, uncertainty classification, and recursive error correction. Journal of Artificial Intelligence General science (JAIGS) ISSN:3006-4023. https://doi.org/10.60087/jaigs.v10i01.502
Vancouver
Lukss A. Adversarial, Recursive, Generalized Uncertainty System: A falsifiable metascientific architecture for adversarial claim evaluation, causal auditing, evidence provenance, uncertainty classification, and recursive error correction. Journal of Artificial Intelligence General science (JAIGS) ISSN:3006-4023. 2026. doi:10.60087/jaigs.v10i01.502
BibTeX
@article{Lukss2026,
author = {Lukss, Andris},
title = {Adversarial, Recursive, Generalized Uncertainty System: A falsifiable metascientific architecture for adversarial claim evaluation, causal auditing, evidence provenance, uncertainty classification, and recursive error correction},
journal = {Journal of Artificial Intelligence General science (JAIGS) ISSN:3006-4023},
year = {2026},
doi = {10.60087/jaigs.v10i01.502}
}
EUVIMED