When Do Causal World Models Help Modular LLM Agents

因果世界模型何时能辅助模块化 LLM 智能体？

Abstract: LLM agents increasingly act through modular systems, such as order, payment, inventory, and shipment services, where actions in one module change which transitions are valid in another. Standard world models usually fit observational traces, but this is not the quantity needed for intervention-time planning: a trace may show that payment precedes shipment without identifying whether payment authorizes shipment, inventory mediates the effect, or a hidden trigger explains both.

摘要： 大语言模型（LLM）智能体正越来越多地通过模块化系统（如订单、支付、库存和物流服务）进行操作，其中一个模块的操作会改变另一个模块中有效转换的状态。标准的“世界模型”通常基于观测轨迹进行拟合，但这并非干预时规划（intervention-time planning）所需的关键信息：一条轨迹可能显示“支付”发生在“物流”之前，但无法识别究竟是支付授权了物流、库存中介了这一效应，还是存在某种隐藏的触发因素同时导致了两者。

We study this gap through FedCausalCompose, a causal world-model framework for modular LLM agents in which local actions provide intervention-response evidence for cross-module interfaces. We first show that observational world models incur an irreducible interventional error under unblocked back-door paths, that interface recovery improves with intervention-response coverage, and that an oracle causal composition can beat the non-causal lower bound when coverage and local mechanism errors are controlled.

我们通过 FedCausalCompose 研究了这一差距。这是一个专为模块化 LLM 智能体设计的因果世界模型框架，其中局部操作为跨模块接口提供了干预响应证据。我们首先证明，在存在未阻断的后门路径（back-door paths）时，观测型世界模型会产生不可约的干预误差；同时，接口恢复能力会随着干预响应覆盖率的提高而改善；此外，当覆盖率和局部机制误差得到控制时，预言机（oracle）因果组合的表现可以优于非因果模型的下界。

We then test the resulting prediction in diagnostic agent settings. Causal interfaces help most in structured tool environments, where API signatures expose preconditions and downstream effects. In contrast, dialogue and narrative environments often ignore raw edge lists unless a short attention anchor makes the causal information decision-relevant.

随后，我们在诊断智能体场景中测试了上述预测。因果接口在结构化工具环境中帮助最大，因为 API 签名能够明确暴露前提条件和下游影响。相比之下，在对话和叙事环境中，除非存在简短的注意力锚点（attention anchor）使因果信息与决策相关，否则模型往往会忽略原始的边列表（edge lists）。

These results identify a concrete condition for causal world models in LLM agents: causal structure helps when cross-module interfaces are both statistically identifiable and presented in a form the agent can use at action time.

这些结果为 LLM 智能体中的因果世界模型确定了一个具体条件：当跨模块接口既能在统计上被识别，又以智能体在行动时可用的形式呈现时，因果结构才能发挥作用。