文章背景与核心概要
本文是一篇发表于 ICML 2026 的前沿立场论文，直面了当前人工智能领域关于“推理（Reasoning）”概念的定义危机。随着 AI 从传统的符号逻辑转向深度概率生成模型，学界始终缺乏对推理的统一操作性定义，这导致当前的 AI 评估缺乏构念效度，阻碍了迈向真正值得信赖的自主系统的步伐。
为了解决这一模糊性，作者提出了一套将有效且可靠的推理视为“基于规则的可学习过程”的理论框架，并综合文献给出了明确的操作定义。同时，论文还提供了一份标准化检查清单，旨在规范 AI 推理研究的学术交流与最佳实践，为该领域的科学发展奠定可验证的基础。
Position: Reasoning is a Learnable Rule-Based Process
Position: Reasoning is a Learnable Rule-Based Process
Authors: Rachel Lawrence, Jacqueline Maasch
Published: 29 May 2026
Venue: Proceedings of the 43rd International Conference on Machine Learning (ICML), 2026
arXiv: 2608.12325 [cs.AI]
Authors: Rachel Lawrence, Jacqueline Maasch
Published: 29 May 2026
Venue: Proceedings of the 43rd International Conference on Machine Learning (ICML), 2026
arXiv: 2608.12325 [cs.AI]
Summary
本立场论文探讨了当前人工智能领域关于“推理”概念的定义危机。尽管现代 AI 已从符号逻辑转向深度概率生成模型，但该领域仍缺乏对推理的统一操作性定义。作者认为，这种模糊性使得当前 AI 评估的构念效度无法得到验证，从而阻碍了朝着真正值得信赖的自主系统迈进的步伐。为了解决这一问题，论文提出了一个将有效且可靠的推理视为基于规则的可学习过程的框架，并为交流 AI 推理研究提供了标准化的检查清单。
Summary
This position paper addresses the current crisis of definition in the field of Artificial Intelligence regarding "reasoning." While modern AI has shifted from symbolic logic to deep probabilistic generative models, the field lacks a unified operational definition of reasoning. The authors argue that this ambiguity makes it impossible to verify the construct validity of current AI evaluations, thereby hindering progress toward truly trustworthy autonomous systems. To resolve this, the paper proposes a framework that treats valid and sound reasoning as a learnable rule-based process and provides a standardized checklist for communicating AI reasoning research.
Abstract
自主推理是当今 AI 领域最具科学和经济驱动力的课题之一。历史上，这一直是符号式 AI 的专属领域，而近期的进展主要源自深度概率生成模型。尽管人们对此表现出极大的兴趣并取得了快速进展，但生成式 AI 社区尚未就推理的操作性定义达成明确共识，并且常常隐含地排斥逻辑学和可验证自动化推理在历史上对这一课题的研究。
本立场认为，定义的模糊性导致推理评估的构念效度无法验证，从而破坏了朝着值得信赖的自主推理取得可量化进展的努力。我们同时认为，这种模糊性是可以解决的。为此，我们提供了： 1. 操作性定义：基于对文献的综合，将有效且可靠的推理定位为一个基于规则的可学习过程。 2. 一份检查清单：用于规范 AI 推理研究交流的最佳实践。
Abstract
Autonomous reasoning is among the most scientifically and economically motivating topics in AI today. Historically the purview of symbolic AI, recent advances have mainly emerged from deep probabilistic generative models. Despite immense interest and rapid progress, the generative AI community has not clearly converged on operational definitions for reasoning and often implicitly rejects the historical treatment of this topic in logic and verifiable automated reasoning.
This position contends that definitional ambiguity leaves the construct validity of reasoning evaluation unverifiable, undermining quantifiable progress toward trustworthy autonomous reasoning. We also contend that this ambiguity is addressable. To that end, we provide: 1. Operational definitions based on a synthesis of the literature, positioning valid and sound reasoning as a learnable rule-based process. 2. A checklist for best practices in the communication of AI reasoning research.
Access & Resources
Access & Resources
Metadata
| 类别 | 详情 |
|---|---|
| 学科分类 | 人工智能 (cs.AI); 计算与语言 (cs.CL); 机器学习 (cs.LG) |
| DOI | 10.48550/arXiv.2608.12325 |
| 引用格式 | arXiv:2608.12325 [cs.AI] |
Metadata
Category Details Subjects Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) DOI 10.48550/arXiv.2608.12325 Cite As arXiv:2608.12325 [cs.AI]