科研场景下的智能体技术与应用研究

doi:10.13998/j.cnki.issn1002-1248.25-0386

摘要/Abstract

摘要：

【目的/意义】 目前，大语言模型与智能体已成为人工智能领域的核心技术范式，探索智能体在科研场景中的应用发展对推动科研范式的变革具有重要意义。 【方法/过程】 本研究采用客观归纳法，阐述智能体的核心技术模块及多智能体协作方式，结合科研生命周期下的各场景包括文献综述、实验规划设计、数据处理、实验执行及结果分析发现的应用案例，分析其应用价值与现存问题。 【结果/结论】 以大语言模型驱动的智能体核心技术支撑其从基础任务执行者向科学发现者演进，在科研全流程展现革新潜力，但面临推理局限、跨学科整合、生态不完善和伦理安全等挑战。未来需通过领域专属框架开发、可解释性技术突破与伦理治理完善，推动其成为科研创新核心伙伴，助力科研范式向智能化、协同化跃迁。

关键词: 智能体, 大语言模型, 科研服务, AI4S, 科研范式

Abstract:

[Purpose/Significance] Currently, large language models (LLMs) and agents have emerged as core technical paradigms in artificial intelligence, with their integration into scientific research scenarios holding profound significance for transforming research paradigms. Traditional scientific research is facing an increasing number of challenges such as inefficient literature searches, the processing of massive amounts of data, repetitive experimental tasks, and barriers to collaborative innovation. Agents, empowered by LLMs, offer a promising solution to these bottlenecks by enabling intelligent automation and adaptive collaboration across research workflows. Beyond basic task assistance, they play a pivotal role in facilitating knowledge fusion, accelerating breakthroughs in frontier areas, and reshaping traditional research models. This study aims to clarify the core techniques and applications of agents in scientific research, highlighting their transition from auxiliary tools to integral innovation partners, which is crucial for accelerating knowledge discovery, enhancing research efficiency, and promoting the shift toward intelligent and collaborative research models. [Method/Process] Employing an objective, inductive approach, this study thoroughly explains the core technical modules of agents including planning, perception, action, and memory, as well as the operational mechanisms of multi-agent collaboration. It also integrates an analysis of agent applications throughout the entire scientific research lifecycle. This analysis covers key scenarios including literature review and idea formulation, experimental planning and design, data processing and execution, result analysis and knowledge discovery, and research report composition. By analyzing the application value and existing limitations of agents, this study proposes prospects and recommendations for the application and development of agents in scientific research scenarios. [Results/Conclusions The findings reveal that LLM-driven agents are evolving from basic task executors to active participants in scientific discovery, demonstrating significant transformative potential throughout the entire research workflow. They facilitate more efficient information processing, smarter experimental design, and deeper knowledge integration, thereby redefining traditional research patterns. However, several challenges persist, including limitations in long-range reasoning capabilities, and underdeveloped ecosystem support. There are also ethical and security concerns, such as data privacy and academic integrity. To address these, future efforts should focus on strengthening intelligent computing infrastructure for scientific data, deepening collaborative development of domain-specific agents, establishing a unified open collaboration framework with standardized interfaces, and building "human-in-the-loop" hybrid systems and multiple evaluation mechanisms. These measures will enable agents to become core partners in scientific innovation, driving the transition of research paradigms toward greater intelligence and collaboration.

Key words: agents, large language model, scientific research service, AI4S, scientific research paradigm

中图分类号: G350

钱力, 王茜颖, 刘熠, 张元哲, 常志军. 科研场景下的智能体技术与应用研究[J]. 农业图书情报学报, 2025, 37(5): 5-14.

QIAN Li, WANG Qianying, LIU Yi, ZHANG Yuanzhe, CHANG Zhijun. Agent Technology and Its Applications in Scientific Research[J]. Journal of library and information science in agriculture, 2025, 37(5): 5-14.

图/表 2

参考文献 31

[1]	YANG J, LI A, FARAJTABAR M, et al. Learning to incentivize other learning agents[J/OL]. ArXiv, 2020. .
[2]	LAIRD J E, NEWELL A, ROSENBLOOM P S. SOAR: An architecture for general intelligence[J]. Artificial intelligence, 1987, 33(1): 1-64.
[3]	WOOLDRIDGE M, JENNINGS N R. Intelligent agents: Theory and practice[J]. The knowledge engineering review, 1995, 10(2): 115-152.
[4]	WANG L, MA C, FENG X Y, et al. A survey on large language model based autonomous agents[J]. Frontiers of computer science, 2024, 18(6): 186345.
[5]	KHOT T, TRIVEDI H, FINLAYSON M, et al. Decomposed prompting: A modular approach for solving complex tasks[J/OL]. ArXiv, 2022. .
[6]	YAO S, ZHAO J, YU D, et al. ReAct: Synergizing reasoning and acting in language models[J/OL]. ArXiv, 2022. .
[7]	HOU X, YANG M, JIAO W, et al. CoAct: A global-local hierarchy for autonomous agent collaboration[J/OL]. ArXiv, 2024.
[8]	DONGRE V, YANG X C, ACIKGOZ E C, et al. ReSpAct: Harmonizing reasoning, speaking, and acting towards building large language model-based conversational AI agents[J]. arXiv e-prints, 2024: arXiv: .
[9]	PAN J B, ZHANG Y, ZHANG C, et al. DynaThink: Fast or slow? A dynamic decision-making framework for large language models[C]//Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Miami, Florida, USA. Stroudsburg, PA, USA: ACL, 2024: 14686-14695.
[10]	LEANG J O J, GEMA A P, COHEN S B. CoMAT: Chain of mathematically annotated thought improves mathematical reasoning[J/OL]. ArXiv, 2024.
[11]	SCHICK T, DWIVEDI-YU J, DESSÌ, et al. Toolformer: Language models can teach themselves to use tools[J]. arXiv e-prints, 2023: arXiv: .
[12]	DU Y, WEI F, ZHANG H. AnyTool: Self-reflective, hierarchical agents for large-scale api calls[J/OL]. ArXiv, 2024. .
[13]	LIU X, PENG Z, YI X, et al. ToolNet: Connecting large language models with massive tools via tool graph[J/OL]. ArXiv, 2024. .
[14]	MODARRESSI A, IMANI A, FAYYAZ M, et al. RET-LLM: Towards a general read-write memory for large language models[J/OL]. ArXiv, 2023. .
[15]	HONG S, ZHUGE M, CHEN J, et al. MetaGPT: Meta programming for a multi-agent collaborative framework[J/OL]. ArXiv, 2023. .
[16]	ANOKHIN P, SEMENOV N, SOROKIN A, et al. AriGraph: Learning knowledge graph world models with episodic memory for LLM agents[J]. arXiv e-prints, 2024: arXiv: .
[17]	TRAN K, DAO D, NGUYEN M, et al. Multi-agent collaboration mechanisms: A survey of LLMs[J/OL]. ArXiv, 2025. .
[18]	WU Q Y, BANSAL G, ZHANG J Y, et al. AutoGen: Enabling next-gen LLM applications via multi-agent conversation framework[J/OL]. ArXiv, 2023.
[19]	HE Y, HUANG G, FENG P, et al. PaSa: An LLM agent for comprehensive academic paper search[J/OL]. ArXiv, 2025. .
[20]	CHEN Z, LIU K, WANG Q, et al. MindSearch: Mimicking human minds elicits deep AI searcher[J/OL]. ArXiv, 2024. .
[21]	SAMI M A, RASHEED Z, KEMELL K, et al. System for systematic literature review using multiple AI agents: Concept and an empirical evaluation[J/OL]. ArXiv, 2024. .
[22]	BAEK J, JAUHAR S K, CUCERZAN S, et al. ResearchAgent: Iterative research idea generation over scientific literature with large language models[J/OL]. ArXiv, 2024. .
[23]	GOTTWEIS J, WENG W, DARYIN A, et al. Towards an AI co-scientist[J/OL]. ArXiv, 2025. .
[24]	QU Y H, HUANG K X, COUSINS H, et al. CRISPR-GPT An LLM agent for automated design of gene-editing experiments[J]. bioRxiv, 2024: 2024.04.25.591003.
[25]	NI B, BUEHLER M J. MechAgents: Large language model multi-agent collaborations can solve mechanics problems, generate new data, and integrate knowledge[J]. Extreme mechanics letters, 2024, 67: 102131.
[26]	CHRIS L, CONG L, ROBERT T L, et al. The AI scientist: Towards fully automated open-ended scientific discovery[J/OL]. arXiv preprint arXiv:, 2024.
[27]	XIN Q, KONG Q, JI H, et al. BioInformatics Agent (BIA): Unleashing the power of large language models to reshape bioinformatics workflow[J/OL]. ArXiv, 2024. .
[28]	BRAN A M, COX S, SCHILTER O, et al. Augmenting large language models with chemistry tools[J]. Nature machine intelligence, 2024, 6(5): 525-535.
[29]	GHAFAROLLAHI A, BUEHLER M J. ProtAgents: Protein discovery via large language model multi-agent collaborations combining physics and machine learning[J]. Digital discovery, 2024, 3(7): 1389-1409.
[30]	XIAO Y, LIU J, ZHENG Y, et al. CellAgent: An LLM-driven multi-agent framework for automated single-cell data analysis[J/OL]. ArXiv, 2024. .
[31]	SCHMIDGALL S, SU Y, WANG Z, et al. Agent laboratory: Using LLM agents as research assistants[J/OL]. ArXiv, 2025. .