aiqbench analysis
This research fundamentally challenges the assumption that overlapping rule sets in reinforcement learning guarantee behavioral consistency, demonstrating instead that policies may share symbols while acting randomly on new states. The study establishes a rigorous protocol for auditing RL systems by decomposing interpretability into six distinct, testable predicates rather than treating it as a monolithic property. These findings provide critical constraints for developers seeking to compose opaque neural checkpoints into auditable discrete rules without falsely claiming autonomous skill generation.
Auditability Is Not One Property: Rule Overlap, Behavioural Agreement, and Composition in Reinforcement Learning
arXiv:2609.28581v1 Announce Type: new Abstract: Reinforcement-learning (RL) policies are often distributed as opaque neural checkpoints, while training logs show that a run occurred without explaining what the policy learned. We study whether independently trained policies can be represented and composed through auditable discrete behavioral rules. We define auditability as six separately testable predicates: trace integrity, lossless coding, rule coverage, behavioral agreement, composition quality, and value-model reliability. Our protocol uses a shared frozen symbolizer, passive rule extraction, an append-only hash-bound ledger, exact environment replay, and offline confidence-ranked arbitration with an explicit blind-spot fallback. The results place strict limits on this description layer. Rule-set overlap does not imply behavioral agreement: policies may share symbolic rules while choosing near-chance-matching actions on fresh states. The fused policy therefore selects among existing rules rather than generating a new skill. On a conflict-dominated task, an apparent fusion failure is traced to an induction/deployment mismatch: rules induced from sampled actions were evaluated under argmax actions, and deployment-consistent re-induction reverses the arbitration ordering. A fitted-Q generalized-policy-improvement diagnostic also fails in both environments, limiting claims that rule fusion is superior to value-based composition. One exploratory comparison favors rule fusion, but its comparator is post hoc, the task is partly saturated, and the fused policy remains below the strongest held-out actor. We contribute an evidence-bounded audit and composition protocol, not a claim of universal interpretability or autonomous skill generation. Future work must add temporally extended skills, cross-skill interfaces, composition search, and independent novelty audits.
Abstract:Reinforcement-learning (RL) policies are often distributed as opaque neural checkpoints, while training logs show that a run occurred without explaining what the policy learned. We study whether independently trained policies can be represented and composed through auditable discrete behavioral rules. We define auditability as six separately testable predicates: trace integrity, lossless coding, rule coverage, behavioral agreement, composition quality, and value-model reliability. Our protocol uses a shared frozen symbolizer, passive rule extraction, an append-only hash-bound ledger, exact environment replay, and offline confidence-ranked arbitration with an explicit blind-spot fallback. The results place strict limits on this description layer. Rule-set overlap does not imply behavioral agreement: policies may share symbolic rules while choosing near-chance-matching actions on fresh states. The fused policy therefore selects among existing rules rather than generating a new skill. On a conflict-dominated task, an apparent fusion failure is traced to an induction/deployment mismatch: rules induced from sampled actions were evaluated under argmax actions, and deployment-consistent re-induction reverses the arbitration ordering. A fitted-Q generalized-policy-improvement diagnostic also fails in both environments, limiting claims that rule fusion is superior to value-based composition. One exploratory comparison favors rule fusion, but its comparator is post hoc, the task is partly saturated, and the fused policy remains below the strongest held-out actor. We contribute an evidence-bounded audit and composition protocol, not a claim of universal interpretability or autonomous skill generation. Future work must add temporally extended skills, cross-skill interfaces, composition search, and independent novelty audits.
35 pages, 4 figures, 14 tables. Experimental results cover eight random seeds on CartPole-v1 and Acrobot-v1. Code, data, and audit artifacts are available in the accompanying repository
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2609.28581 [cs.LG]
(or arXiv:2609.28581v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.28581 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 287 words
Related coverage
LiveLaw Supreme Court Half-Yearly Digest: January - June 2026 [Part 2]
livelaw.in · Sep 13, 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · Oct 2, 2026
Lock-In Risk Needs More Researchers; Here's Where to Start
LessWrong · Jun 17, 2026
Game Design Creativity evals
LessWrong · Oct 12, 2026