[论文] Agentic World Modeling: Foundations, Capabilities, Laws, and Beyond

## 论文概要 **研究领域**: ML **作者**: Meng Chu, Xuan Billy Zhang, Kevin Qinghong Lin **发布时间**: 2025-04-28 **arXiv**: [2504.19771](https://arxiv.org/abs/2504.19771) ## 中文摘要 随着AI系统从文本生成转向通过持续交互完成目标，建模环境动态的能力成为核心瓶颈。我们引入了沿两个轴组织的'层级 x 法则'分类体系。第一个轴定义了三个能力层级：L1预测器，学习单步局部转移算子；L2模拟器，将其组合为多步、动作条件化的展开，并遵守领域法则；L3演化器，在预测与新证据冲突时自主修正自身模型。第二个轴识别了四种支配性法则体系：物理、数字、社会和科学。利用这一框架，我们综合了400多项工作，总结了100多个代表性系统，涵盖基于模型的强化学习、视频生成、网页和GUI智能体、多智能体社会模拟以及AI驱动的科学发现等领域。 ## 原文摘要 As AI systems move from generating text to accomplishing goals through sustained interaction, the ability to model environment dynamics becomes a central bottleneck. We introduce a 'levels x laws' taxonomy organized along two axes. The first defines three capability levels: L1 Predictor, which learns one-step local transition operators; L2 Simulator, which composes them into multi-step, action-conditioned rollouts that respect domain laws; and L3 Evolver, which autonomously revises its own model when predictions fail against new evidence. The second identifies four governing-law regimes: physical, digital, social, and scientific. Using this framework, we synthesize over 400 works and summarize more than 100 representative systems spanning model-based reinforcement learning, video generatio... --- *自动采集于 2026-04-28* #论文 #arXiv #ML #小凯