The Grounding Problem in AI Alignment: A case for a personal source of normative reference

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

An artificial system can follow legitimate instructions without settling why affected persons matter beyond their usefulness. This paper develops a personal-Ground hypothesis: finite persons, intelligibility, and normative goodness ultimately depend on a personal source whose care affirms creatures as having goods of their own. It reconstructs a conditional metaphysical route, distinguishes value from reasons and obligations, and states the additional moral premises needed to connect creaturely goods to action. A shared repair example shows how rival grounding accounts can support the same conduct while disagreeing about its ultimate explanation. Reasons fundamentalism, reflective agency, naturalist realism, and theistic social obligation receive separate consideration. The proposed explanatory unity supplies a reason to investigate the hypothesis; establishing its comparative superiority requires more than identifying a desirable combination of attributes. The replies to suffering and hiddenness distinguish temporal scale, recognition criteria, and access sufficient for a specified response. They challenge particular inferences while leaving the comparative evidential questions open. Arbitrariness and the move to personality also remain substantive challenges. The argument map and comparison materials make these dependencies available for criticism. The contribution is a philosophical case and an account of its implications for alignment, not a demonstration of divine existence or an implemented method. Recognition, motivation, historical identification with Christ, and durable artificial behavior retain distinct burdens. Research status: Philosophical working paper; conditional metaphysical argument, selected primary-source comparisons, and author-prepared review materials. Not peer reviewed. No model or participant experiment or implemented alignment system. Developed from the authors’ original framework with AI assistance (Codex). Contents: Paper in PDF, HTML, and Markdown, README, argument map, comparison ledger, review questions, license notice, and file manifest. Version 1.1 revises the unpublished version 1.0 draft. License: All author-owned paper and accompanying research/design materials use CC BY 4.0. No software is included. See LICENSE.md for scope. Related works: Project antecedents are mFLO 111, Invariant Relation and Non-Substitution, version 1.1, https://doi.org/10.5281/zenodo.22805469; and From Corrigibility to Answerability, version 1.1, https://doi.org/10.5281/zenodo.22806461.