There is increasing concern about the unilateral power of the organizations involved in the development, alignment, and governance of AI. Recent pilots - such as Meta's Community Forums and Anthropic's Collective Constitutional AI - have illustrated a promising direction, where democratic processes might be used to meaningfully improve public involvement and trust in critical decisions. However, there is no standard framework for evaluating such processes. In this paper, building on insights from the theory and practice of deliberative democracy, we provide a "Democracy Levels" framework for evaluating the degree to which decisions in a given domain are made democratically. The framework can be used (i) to define milestones in a roadmap for the democratic AI, pluralistic AI, and public AI ecosystems, (ii) to guide organizations that need to increase the legitimacy of their decisions on difficult AI governance questions, and (iii) as a rubric by those aiming to evaluate AI organizations and keep them accountable.

翻译：当前，人们日益关注参与人工智能开发、对齐与治理的组织所掌握的单边权力。近期的试点项目——如Meta的社区论坛和Anthropic的集体宪法AI——揭示了一个充满希望的方向：民主进程或许能有效提升公众对关键决策的参与度和信任度。然而，目前尚缺乏评估此类进程的标准框架。本文基于协商民主的理论与实践洞见，提出一套“民主等级”框架，用于评估特定领域决策的民主化程度。该框架可用于：（i）为民主人工智能、多元人工智能及公共人工智能生态系统的发展路线图设定里程碑；（ii）指导需要提升其在困难的人工智能治理问题上决策合法性的组织；（iii）作为评估人工智能组织并确保其问责的衡量标准。