ARCHIVE-GROUNDED AI FOR INTERDISCIPLINARY EDUCATION:

Abstract

Archive-grounded generative AI systems are increasingly important for science, education, culture, and public-interest knowledge work, yet their reliability depends on more than model performance alone. This paper examines three related systems - BloomBot, VelikovskyBot, and a family of Evolutionary Culturology GPTs and EthiSizer AI GPTs - as case studies in transparent, socially responsible human-AI-archive interaction. We argue that such systems should be understood not merely as chatbots, but as governed epistemic interfaces: systems that mediate between users and curated bodies of expert knowledge, while making sources, limits, and interpretive steps visible. BloomBot, developed by Ryan T. Dean, supports engagement with Howard Bloom's interdisciplinary archive. VelikovskyBot extends this approach to J. T. Velikovsky's archive and the meta-meta-science of Evolutionary Culturology. The Ev Cult GPTs further explore how archive-grounded AI can assist interdisciplinary education and public reasoning by mapping concepts across domains through the universal unit of the fractal HOLON/parton. The paper also considers the international portability of such systems, comparing their potential relevance for Australian, North American/Canadian, Balkan/EU, and MENA/Global South educational and cultural contexts. We propose a practical design framework for public-interest AI: source grounding, inference labelling, persona restraint, epistemic humility, and user-facing explainability. These principles help reduce hallucination, preserve intellectual context, and build public trust in AI-assisted inquiry.