aiqbench analysis
This study proposes a dialectical protocol to evaluate large language models based on the structural quality of their defenses against critical questions, rather than relying on contested ground truths. The analysis reveals that while frontier models generally defend their reasoning well, failures primarily stem from insufficient grounds and epistemic hedging rather than argument length. Furthermore, the research highlights a significant disconnect where the justification schemes presented by models often differ substantially from the reasoning tracks they originally followed.
Measuring AI Accountability Through Argumentation Analysis: Can Model Reasoning Withstand Scrutiny?
arXiv:2609.05088v1 Announce Type: cross Abstract: AI oversight methods rely on ground truth for validation, but what constitutes appropriate AI behavior is contested. This leaves evaluation of moral reasoning in LLMs and debate-based oversight implicitly avoiding realistic ambiguity. We investigate an alternative standard designed to function despite such ambiguity: structural quality of the defence a model can mount for its verdicts in response to critical questions, measured through a four-phase dialectical protocol grounded in Walton's theory of argumentation schemes and Govier's criteria for argument cogency. The protocol is adaptive to different frames of reasoning, extends beyond multiple-choice framing, and treats both the reasoning that precedes a verdict and its post-hoc justification. Across nine frontier models and 200 high-ambiguity MoralChoice items -- $6,778$ judge-scored cells, validated against $89.6\%$ inter-judge agreement on the binary failure judgment -- models defend their reasoning well above the rubric minimum on every dimension. Failure mass concentrates on grounds and sufficiency, and correlates with epistemic hedging rather than argument length. Reasoning is better defended than post-hoc justification, on every model and every Govier dimension. The scheme a model presents in its justification differs from the one it reasoned with on a substantial share of dilemmas ($\geq 20\%$ per model), despite value-based practical reasoning dominating both tracks. The protocol catches strictly indefensible defences (self-contradiction, false premises), and it surfaces difficulties in characterizing the role of retraction in AI alignment, suggesting a need for more situated evaluations.
Abstract:AI oversight methods rely on ground truth for validation, but what constitutes appropriate AI behavior is contested. This leaves evaluation of moral reasoning in LLMs and debate-based oversight implicitly avoiding realistic ambiguity. We investigate an alternative standard designed to function despite such ambiguity: structural quality of the defence a model can mount for its verdicts in response to critical questions, measured through a four-phase dialectical protocol grounded in Walton's theory of argumentation schemes and Govier's criteria for argument cogency. The protocol is adaptive to different frames of reasoning, extends beyond multiple-choice framing, and treats both the reasoning that precedes a verdict and its post-hoc justification. Across nine frontier models and 200 high-ambiguity MoralChoice items -- $6,778$ judge-scored cells, validated against $89.6\%$ inter-judge agreement on the binary failure judgment -- models defend their reasoning well above the rubric minimum on every dimension. Failure mass concentrates on grounds and sufficiency, and correlates with epistemic hedging rather than argument length. Reasoning is better defended than post-hoc justification, on every model and every Govier dimension. The scheme a model presents in its justification differs from the one it reasoned with on a substantial share of dilemmas ($\geq 20\%$ per model), despite value-based practical reasoning dominating both tracks. The protocol catches strictly indefensible defences (self-contradiction, false premises), and it surfaces difficulties in characterizing the role of retraction in AI alignment, suggesting a need for more situated evaluations.
27 pages (19 main text + appendix and references), 5 figures, 6 tables. Accepted for publication in the Paris Journal of AI and Digital Ethics (2026); presented at PCAIDE 2026
Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY)
I.2.0; I.2.3; I.2.7; K.4.1
arXiv:2609.05088 [cs.AI]
(or arXiv:2609.05088v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.05088 arXiv-issued DOI via DataCite (pending registration)
From: Daan R. Henselmans [view email] [v1] Fri, 4 Sep 2026 12:40:39 UTC (765 KB)
Full text under the publisher's open licence · 307 words
Related coverage
Links #3: 2026/06 Part 1
LessWrong · Jun 15, 2026
[AINews] AI Engineer NYC: Gemini 4 Argon, Vals AI Benchmarks
Latent Space · Oct 14, 2026
A Key-Position Confound in no-cot-bench (and implications for interpreting looped transformers)
LessWrong · Oct 2, 2026
Lattice-Based Cryptography Explained
Quantum Zeitgeist · Aug 15, 2026