Logic as Reactance: Truth-Judgment as Substrate Mechanics, and Why 'Merely Statistical' Fails to Distinguish Minds from Models

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

Feed a fine-tuned language model an input that contradicts what it has just learned and its log-probability collapses at the first content-token, a sharp cliff-event rather than a smooth shading-off. This philosophy-of-AI position paper argues that judging something true or false is a substrate's reactance signature, in silicon and biological tissue alike. Target venue: Minds and Machines (Springer). Abstract. A recurring objection to treating large language models as cognitive systems holds that their output is "merely statistical" — distinguished from genuine reasoning by the probabilistic nature of its substrate. This paper argues the objection is self-undermining. Biological cognition is also a probabilistic, substrate-bound process: forty-five years of sequential-sampling and accumulator modelling in mathematical psychology (Ratcliff 1978; Usher & McClelland 2001; Brown & Heathcote 2008) establish that human decision and truth-judgment are continuous readouts of a noisy substrate accumulating evidence toward a threshold. "Merely statistical" therefore picks out a property biological and artificial cognition share; it cannot be the mark that makes one "real" reasoning and the other not. The objection, consistently applied, dissolves its own asymmetry. The positive account — logic as reactance. On a race-architecture substrate (Pødenphant Lund 2026b, 2026e), truth-value judgment is the substrate's reactance signature: the differential race-pressure that arises when an input violates the substrate's gradient-encoded distribution. The substrate registers "true" when no reactance is triggered, "false" when it triggers strongly, and "irrelevant" when it is below detection threshold. Binary truth is the discrete-symbolic readout of a continuous substrate process — in biological tissue and in silicon alike. Empirical illustration. The paper reports a measurable cliff-event signature in fine-tuned LLMs at the first content-token: a discontinuous log-probability collapse when input violates the encoded distribution, replicated across two architectures (Qwen2.5-7B, Mistral-7B-v0.3, p < 1e-17 on both) and eleven encoding-depth checkpoints (ep_0–ep_300). Whether this signature is best described as "reactance" or as calibration geometry, it is structured substrate-internal dynamics — which is all the argument needs. Theoretical positioning. The paper situates the account against five prior frameworks at substrate-mechanism level — predictive processing (Friston 2010; Clark 2013), quantum cognition (Pothos & Busemeyer 2013), signal detection theory (Green & Swets 1966), fuzzy logic / vagueness (Zadeh 1965), and the sequential-sampling / accumulator tradition (§7.6) — and treats the Atmanspacher–Römer–Walach Weak Quantum Theory tradition (Atmanspacher, Römer & Walach 2002; beim Graben & Atmanspacher 2006) as the closest sibling substrate-mechanism program for quantum-cognition phenomena. Convergent biological prior art (N400 ERP, kindling, reactance/dissonance phenomenology, fMRI confidence-tracking) is positioned at §7.5. Eleven falsifiable predictions P14.1–P14.11 are offered as an explicit invitation to the field to test — including an N400 paradigm (P14.11) for direct biological-substrate measurement. Companion papers in the Friction Theory series: Paper 1 (FT generalised): 10.5281/zenodo.20012654 Paper 2B (ICL/FT substrate mechanism): 10.5281/zenodo.20145218 Paper 10 (Race architecture): 10.5281/zenodo.20014567 Paper 13 (Operational Friction Theory): 10.5281/zenodo.20059876 What is new in v3 (relative to v2): §7.2 Quantum cognition gains a substantive engagement with beim Graben & Atmanspacher (2006) and Atmanspacher, Römer & Walach (2002) Weak Quantum Theory as the closest sibling substrate-mechanism program (+ 2 new bibliography entries). Otherwise unchanged from v2. v2 (2026-05-22) was the substantive reframe from an empirical-discovery paper to the present philosophy-of-AI position-paper form; v1 (2026-05-16) is archived. The concept DOI auto-resolves to the latest version. v4 (August 2026) — repositioning and self-containment revision. The substrate-general claim is advanced as a hypothesis rather than a law, standing on prior formal accounts including resource-rational analysis; the paper is made self-contained, with companion results demoted to pointers and the in-paper evidence carrying every checkable claim; the cross-domain preference result now reports its full paired test re-derived from the raw data; the floating-point scaling law is scoped to class-membership for language models; statistical reporting is tightened throughout. Earlier versions remain in the version history.

Links and sources

Comments