Knowledge Fingerprint (KFP) — Algebraic Epistemological Identity, Reconstructable Hashes, and Self-Healing Knowledge Bases

Key Points

• The aim is to characterize the epistemological identity of knowledge units and AI agents through a novel deterministic system.
• Developed a 7-dimensional taxonomy for knowledge characterization.
• Utilized dual-fingerprinting for detecting metadata inconsistencies with an identified misclassification rate.
• Implemented Reconstructable Knowledge Hashes for knowledge management and profiling.
• Applied epistemological RAG filtering to reduce unsafe content.
• Conducted testing using a benchmark inspired by PubMedQA.
• Achieved 82% accuracy in detecting metadata inconsistencies through dual-fingerprinting.
• Reduced unsafe content by 86% using epistemological RAG filtering.
• Demonstrated 66% rule-based and 82% LLM accuracy on the benchmark for agent behavior.