In this paper, we propose a semantic framework to describe the outputs of AI systems. The output of an AI system is not the fact or world state it appears to describe, but rather an engineered representation.
By distinguishing between accepted domain knowledge, the content of reference sources, and what the system can currently utilize, we can examine the correctness of these representations. This framework allows us to precisely define common failures such as extrapolation, refuted or unsupported assertions, source versus knowledge mismatch, stale or refuted sources, added hypotheses, and unsupported usage.
We hope our framework provides a useful vocabulary for specifying and checking AI systems whose outputs, citations, tool calls, and world-changing actions must be justified by reliable claims and explicit authority rather than apparent fluency.
Blogger's Review: This study emphasizes the semantic accuracy of AI system outputs, highlighting the importance of establishing a reliable knowledge base in AI applications. By defining common errors, the research offers a fresh perspective on the reliability of AI systems and encourages a deeper reflection on the mechanisms for reviewing system outputs.