Neural Noise as a Substrate of Cognitive Novelty

This paper is a companion meta-analysis to: lajjadred (2026). Beyond Deductive Machinery: Why LLMs Cannot Generate Truly Novel Conclusions. Zenodo. https://doi.org/10.5281/zenodo.20324204 Background. Large language models (LLMs) are widely described as systems capable of creative output, yet the theoretical basis for this claim has not been rigorously examined against the neuroscientific evidence on what biological creativity actually requires. A companion paper (lajjadred, 2026) argues that LLMs are structurally incapable of generating genuinely novel conclusions because they lack physical stochasticity at the substrate level. The present paper provides the empirical foundation for that argument by systematically reviewing and meta-analyzing the evidence on whether neural stochasticity is a functionally relevant substrate of cognitive novelty in biological systems. Objectives. Four research questions were addressed: (RQ1) Is there a reliable empirical association between neural stochasticity and cognitive novelty? (RQ2) Does evidence support a causal direction from neural noise to creativity? (RQ3) Which types of neural noise matter most, and at which level of the neural hierarchy? (RQ4) Does the relationship conform to an inverted-U function consistent with stochastic resonance? Results. Neural variability reliably predicts creative performance across modalities. The largest study (Chen et al., 2025; N = 2,433 across 10 datasets from 5 countries) found DMN-ECN switching frequency predicted divergent thinking (g = 0.174, p < .001) but not general intelligence (g = 0.023, p = .485). tRNS studies demonstrated that externally induced cortical noise produces large improvements in verbal creativity (pooled d = 1.57). Pharmacological suppression of dopaminergic tone reduced creative output (ρ = −0.47). The noise-creativity relationship conforms to an inverted-U consistent with stochastic resonance. Conclusions. Neural stochasticity is a reliable, specific, and causally relevant substrate of cognitive novelty. LLM temperature sampling is empirically dissociable from biological stochasticity: it is weakly correlated with novelty, convergent toward repetition under sustained generation, and structurally incapable of redirecting computation in flight. The development of AI systems capable of genuine cognitive novelty requires hardware-level physical indeterminacy that no current architecture instantiates. Keywords: neural noise, stochastic resonance, cognitive novelty, creativity, divergent thinking, Default Mode Network, dopamine, transcranial random noise stimulation, large language models, meta-analysis