Parallel Cognition for Governance-Grade Intelligence: From Inline Reasoning to Structural Self-Denial Under Optimization Pressure

This record contains the workshop/arXiv-ready preprint PDF: “Parallel Cognition for Governance-Grade Intelligence: From Inline Reasoning to Structural Self-Denial Under Optimization Pressure.” The paper argues that many contemporary AI systems exhibit inline reasoning tendencies—single-path inference that collapses uncertainty into narrative coherence—thereby limiting governance-grade properties such as persistent self-critique, stable epistemic humility, and ethical restraint under optimization pressure. We propose parallel cognition (runtime co-maintenance of competing hypotheses and value-frames with active internal critique) as a key architectural ingredient for robust judgment under uncertainty and adversarial pressure. We then identify a primary failure mode of naive parallelism: internal dominance, where proxy-efficient processes starve slower ethical deliberation. To mitigate this collapse, the paper introduces structural self-denial, formalized as an invariant restriction on admissible policies and internal rewrites, anchored in a transcendent reference understood as an authority condition that bounds optimization (not an external oracle). The manuscript provides: (i) a taxonomy distinguishing multi-sampling, ensembles, critic-augmentation, debate, and genuine parallel cognition; (ii) minimal dynamical sketches (replicator- and bandit-style) clarifying internal dominance as selection pressure over compute allocation; (iii) operational definitions for genuine doubt and undecided-state stability; and (iv) a workshop-ready toy experimental agenda with governance metrics (violation rate, selective suspension, ethical-process preservation, and reference invariance). Files: paper.pdf (preprint manuscript) Version: v1 (Zenodo preprint release)