Paper measures how AI debates express and organize values
Original title: Value-Rich, Differently Organized: Measuring Value Expression and Organization in AI-Generated Debates
Generated debates often invoke as many values as human references, but spread them more diffusely across turns.
An arXiv paper introduces a diagnostic framework that separates value expression from value organization in human-reference and AI-generated debates. Grounded in Schwartz's Basic-10 theory, it scores coverage, alignment, diffusion, and cross-turn dynamics over utterances, stances, and full debates. Across several generation systems, AI debates were broadly value-expressive but more locally value-diffuse than matched human references. Code and supplementary materials are on GitHub.
- The framework represents debates as multi-level value distributions.
- Source-blind judges distinguish value clarity from value focus.
- The authors say evaluation should weigh organization, not only presence.
- Code is at github.com/Maoyuan-li/Value-Expression-and-Organization.
From the original article
1 min read at arxiv.orgCurrent 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.
By Maoyuan Li · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.