行为系统需要行为测试：AI智能体评估新视角

AI 摘要

本文提出，随着AI智能体系统日益成为与动态环境交互的行为系统，现有评估方法仅关注性能结果而忽视行为过程，因此需要借鉴行为科学的方法，通过系统观察、扰动和解释来评估AI行为。作者建议开发严格的行为测试，包括从动作序列中恢复决策策略、构建隔离行为差异的环境以及探测多智能体系统中的涌现动态，并呼吁建立AI行为科学。

正文节选

Position: Behavioral Systems Require Behavioral Tests Abstract Artificial agentic systems increasingly operate as behavioral systems by interacting with dynamic environments, pursuing goals, and adapting over time. Yet, current evaluation methods largely focus on performance outcomes, not the underlying behavioral processes that produce them. This paper argues that AI agents must be evaluated like other behavioral systems: through systematic observation, perturbation, and interpretation of their

发布时间：2026-08-21 12:00

抓取时间：2026-08-20 12:13

来源机构：arXiv