Evolving Hierarchical Modular Intelligence Networks: A Verification-Gated, Self-Expanding Architecture of Specialized Language Models

This concept paper proposes the Evolving Hierarchical Modular Intelligence Network (EHMN), an artificial-intelligence architecture in which limited specialist models operate as reusable components inside a routed hierarchy. Domain groups contain specialist generators, coordinators, and separately optimized verifiers with authority to pass, revise, escalate, or block claims. Verified claim packets preserve evidence, provenance, assumptions, contradictions, and uncertainty as information moves through cross-domain and global synthesis layers. The proposal combines sparse mixture-of-experts routing, independently specialized language models, modular continual learning, retrieval-based memory, and structured hallucination verification. It also defines system invariants, a threat model, controlled formation of provisional specialist groups, and a low-cost experiment for testing whether hierarchical specialization and verification gates improve factuality. Research status: Preprint and concept paper; not peer reviewed. No implemented system, experimental results, or evidence of artificial general intelligence are reported. Generative AI assistance disclosure: OpenAI ChatGPT was used to turn the author’s original ideas and notes into professionally structured academic prose and to assist with organization, editing, and formatting. The author originated the EHMN concept, architecture, and research direction, reviewed the final manuscript, and accepts responsibility for all claims and conclusions.