Active Inference as Context Acquisition for AI Agents

主动推理作为 AI 智能体的上下文获取机制

Abstract: Interactive AI agents must acquire the right context as efficiently as possible. When a user omits a constraint, preference, file, or task variable, an agent can proceed with a default assumption or spend tokens on a clarifying question, retrieval call, tool call, or prompt trial.

摘要： 交互式 AI 智能体必须尽可能高效地获取正确的上下文。当用户遗漏了约束条件、偏好、文件或任务变量时，智能体可以选择基于默认假设继续执行，或者消耗 Token 进行澄清提问、调用检索、调用工具或进行提示词尝试。

We formulate this tradeoff as active inference for context acquisition. An inner inference step updates beliefs over a latent task state, and an outer decision selects the next context action, task action, or stop action to minimize expected free energy under cost.

我们将这种权衡建模为用于上下文获取的主动推理（Active Inference）。内部推理步骤更新对潜在任务状态的信念，而外部决策则选择下一个上下文操作、任务操作或停止操作，以在成本约束下最小化预期自由能。

In deterministic settings, the epistemic term reduces to expected information gain, optionally normalized by token cost. We instantiate the framework in Optimal Question Asking (OQA), with exact posteriors and a dynamic programming oracle, and benchmark frontier language models on binary and multiway categorical tasks from 25 to 300 candidates.

在确定性设置中，认知项（epistemic term）简化为预期信息增益，并可选择通过 Token 成本进行归一化。我们将该框架实例化为“最优提问”（Optimal Question Asking, OQA），利用精确后验概率和动态规划预言机，并在 25 到 300 个候选对象的二元及多元分类任务上对前沿语言模型进行了基准测试。

We also study clarification before generation and automated prompt optimization under token budgets. The formulation is model-agnostic and views active inference as a design principle for the context-acquisition layer of AI agents.

我们还研究了生成前的澄清机制以及在 Token 预算限制下的自动化提示词优化。该公式与模型无关，并将主动推理视为 AI 智能体上下文获取层的一种设计原则。

---

Paper Details:

• Authors: Sanchayan Dutta, Sai Niranjan Ramachandran, Suvrit Sra
• arXiv ID: 2608.19202
• Subjects: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)

论文详情：

• 作者： Sanchayan Dutta, Sai Niranjan Ramachandran, Suvrit Sra
• arXiv ID： 2608.19202
• 学科分类： 人工智能 (cs.AI)；计算与语言 (cs.CL)；机器学习 (cs.LG)