Value-Rich, Differently Organized: Measuring Value Expression and Organization in AI-Generated Debates
Current evaluation of LLM-generated debate largely focuses on argumentative quality, leaving the value structure of generated arguments underexamined. We introduce a diagnostic metric framework for comparing how values are expressed and organized in human-reference and AI-generated debates. The framework separates value expression, the range of recognizable values a debate invokes, from value…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.
