Human-AI Co-Interpretation for Responsible AI: A Hermeneutic Perspective

LLMs should present interpretations as "candidate readings" rather than settled truths: The paper identifies "interpretive misplacement" as a key failure mode where AI-generated interpretations are treated as definitive meanings without explicit justification, sources, or acknowledgment of alternatives. This risks losing accountability in fields like law, education, and policy where decisions depend on defensible interpretations.

AI outputs need transparent interpretive frames to enable accountability: For responsible AI use in high-stakes domains, outputs should include the sources consulted, scope constraints applied, normative commitments made, and provenance information that allows readers to verify supporting passages. Without this transparency, institutions cannot reliably assess what they're committing to when they act on AI interpretations.

Human interpreters must retain final authority over meaning-making: The paper argues that genuine hermeneutic understanding—the kind required for justified interpretation in law, policy, and education—must remain the responsibility of accountable human experts situated within their disciplinary traditions. AI can support this process but cannot replace human judgment in contexts requiring normative accountability.

Current LLM design conflates text generation with interpretation: The paper distinguishes between token prediction (what LLMs actually do) and hermeneutic understanding (what responsible interpretation requires). This distinction is crucial because treating predictive text generation as genuine interpretation obscures the need for human oversight and justification.

Responsible AI requires redesigning human-AI interaction as a structured "interpretive loop": Rather than accepting AI outputs directly, the paper proposes structuring AI-human collaboration so that humans actively engage with AI-generated candidates, evaluate them against evidence and standards, and maintain control over final interpretive judgments.