面向通用自主研究的假设树精炼框架
英文摘要
The paper introduces Arbor, an AI framework for autonomous scientific research that coordinates strategic exploration, isolated hypothesis testing, and knowledge accumulation. Arbor features a long-lived coordinator, short-lived executors, and a Hypothesis Tree Refinement (HTR) system that links hypotheses, artifacts, evidence, and insights over time. The framework outperforms other AI agents across diverse research tasks by enabling iterative, cumulative improvement without constant human intervention.
中文摘要
该论文提出Arbor，一个面向自主科学研究的AI框架，通过战略协调、隔离式假设检验和知识累积，将探索、实验与抽象整合为迭代过程。框架包含长期协调器、短期执行器以及假设树精炼（HTR）系统，持续连接假设、工件、证据和洞察。在多项研究任务中，Arbor的表现优于其他AI代理，无需持续人工干预即可实现自主研究的累积性提升。
关键要点
Arbor is an autonomous research framework with a long-lived coordinator, short-lived executors, and a Hypothesis Tree Refinement (HTR) system.
Arbor是一个自主研究框架，包含长期协调器、短期执行器和假设树精炼（HTR）系统。
HTR links hypotheses, artifacts, evidence, and insights, enabling cumulative and iterative research progress.
HTR连接假设、工件、证据和洞察，支持累积和迭代的研究进展。
The framework manages research strategy, updates a persistent knowledge tree, and refines the search process automatically.
框架自动管理研究策略、更新持久化知识树并优化搜索过程。
Arbor outperforms existing AI agents across a variety of research tasks without constant human intervention.
Arbor在多种研究任务上优于现有AI代理，无需持续人工干预。