GrowLoop: Self-Evolving Conversation Evaluation Seeded by Human

2026-05-26

AI Summary

This work addresses three core challenges in evaluating human-likeness in open-domain dialogue: implicit evaluation criteria, annotator disagreement, and the dynamic evolution of conversational norms. To overcome these issues, the authors propose a self-evolving evaluation paradigm that bootstraps from minimal human annotations and leverages large language model agents to iteratively refine assessment standards through heuristic learning. The framework integrates a human–AI consensus mechanism with a rubric-case co-evolution strategy, enabling continuous adaptation and expansion of the evaluation system. Unlike conventional static or merely difficulty-scaled benchmarks, this approach significantly outperforms existing methods in alignment with human judgments, effectively discriminates between model capability tiers, uncovers issues overlooked by human evaluators, and demonstrates strong generalization to novel conversational scenarios.

Abstract

With the rapid advancement of large language models, evaluating human-likeness in open-ended conversation has become increasingly important. However, human-likeness is a form of tacit knowledge that humans perceive intuitively, yet the underlying criteria resist explicit formulation. Human judgments vary widely, with strong agreement on some cases and legitimate disagreement on others. Meanwhile, the criteria behind human judgments remain implicit, leaving no clear basis for constructing cases. Further, what counts as human-like is not static, but evolving with model capability and human expectations. Despite progress in evaluation methods such as expert-authored benchmarks, Reward Models, and self-evolving benchmarks, none addresses all three challenges simultaneously. Therefore, we propose GrowLoop, a self-evolving conversation evaluation system that continuously adapts as models advance and scenarios shift. With minimal human seed annotations as the first mover, LLM agents iteratively extract and refine evaluation rubrics through Heuristic Learning. Human-AI agreement is required where annotators converge, while only plausibility is expected where they diverge. Moreover, the Rubric-Case co-evolution mechanism enables continuous evolution, expanded through new seeds when the evaluation target moves. Applied to human-likeness evaluation in open-ended conversation, the generated rubrics not only substantially outperform existing methods in alignment with human judgments, but also uncover issues that annotators overlook. The resulting benchmark effectively discriminates models across capability tiers and reveals where they fall short, while generalizing to new scenarios and adapting as models advance. Our work shifts the benchmarking paradigm from manual updates or difficulty scaling to comprehensive, continuous self-evolution.

Problem

• human-likeness
• conversation evaluation
• tacit knowledge
• evolving benchmarks
• open-ended dialogue

Innovation

• self-evolving evaluation
• heuristic learning
• rubric-case co-evolution
• human-likeness assessment
• LLM-based benchmarking