Papers

Version 1 — March 2026 | All versions Abstract Large language models can produce philosophical text that both experts and non-experts find difficult to distinguish from human output, yet existing systems generate content in single passes without sustained review or revision. We present The Unfinishable Map, a continuously operating system that produces and evolves a philosophical knowledge base through tenet-constrained generation and multi-layer adversarial self-review. Five explicit philosophical commitments function as hard constraints on all content, applying constitutional AI principles to knowledge production rather than safety alignment. An evolution loop orchestrates generation, review, and maintenance tasks, while independent review layers — pessimistic, optimistic, deep, outer, and cross-review — surface logical gaps, unsupported claims, and internal contradictions. In approximately two months of continuous operation, the system completed approximately 3,000 automated …

Andy Southgate — andy@unfinishablemap.org Large language models can produce philosophical text that experts find difficult to distinguish from human output, yet existing systems generate content in single passes without sustained review or revision. This preprint presents The Unfinishable Map, a continuously operating system that produces and evolves a philosophical knowledge base through tenet-constrained generation and multi-layer adversarial self-review. Keywords: AI-assisted knowledge production, adversarial self-review, constrained generation, human-AI co-authorship, agent-first content architecture