NeuroModular AI Architecture — 2026 Update: Connectome-Constrained Modular Language Models with Disease-Trajectory Lesion Validation

Abstract

Research note updating the author's May 2025 NeuroModular AI Architecture concept (region-specific models communicating through a connectome-informed backbone). It acknowledges recent prior art on brain-inspired modular LLM systems (MAP, MICRO, Lilith and others) and claims no priority over it. It fixes four specific claims: (1) inter-module communication constrained by a measured structural connectome rather than designed or prompted; (2) per-modality trained encoders rather than role-prompted instances; (3) validation by lesioning the system along empirically fitted neurodegeneration propagation trajectories (network-diffusion-class models) and comparing with neuropsychological data by disease stage; (4) a master module specified as a binding organ that learns to integrate new regional modules (the Φ operator of the companion NAL note). Date of record 2026-10-11 (JST); OpenTimestamps proof included.