[Purpose/Significance] In the face of the transformative wave of science agents reshaping research productivity, S&T documentation and information service (DIS) institutes urgently need to explore how to deeply integrate science agent technology with DIS works to accelerate the enhancement of the capabilities and upgrading the service quality of DIS works. [Method/Process] Based on 518 papers related to scientific intelligence agents from arXiv, this study revealed the current development and characteristics of science agents from two dimensions: domain distribution and research themes. The findings suggested that the development of specialized agents based on multi-agent collaboration is an inevitable choice for various disciplines to embrace science agent technologies. In response to this insight, and by integrating the evolving needs of DIS work in the digital and AI era with a universal agent framework, this paper designed a general framework for DIS agent. It further deconstructed the key tasks in building DIS agent, considering both the technological adaptability required during the development phase and the system security challenges in the application phase. Additionally, the study examined the profound impact of DIS agent on the transformation of DIS work paradigms. [Result/Conclusion] The multi-agent collaborative DIS agent consists of six core components: DIS cognition models, general foundational resources, planning and reasoning, configuration files, large models, and memory. When constructing DIS agent, attention should be focused on six key areas: construction of reliable knowledge bases with multimodal standard alignment, research and development of tool technologies corresponding to information method models, development of modular agents for information contexts, risk testing of action plans for DIS agent, comprehensive evaluation of system performance of DIS agent, and supervision, regulation, and security governance of DIS agent. In the future, driven by DIS agent, the paradigm of DIS work will shift from a human-centric model to one of human-multi-agent collaboration, enhancing the efficiency and intelligence level of intelligence work.
[1] OPENAI. Hello GPT-4o[EB/OL]. [2024-09-01]. https://openai.com/index/hello-gpt-4o/.
[2] TOUVRON H, LAVRIL T, IZACARD G, et al. LLaMA: open and efficient foundation language models [J/OL]. [2024-09-05].ArXiv, 2023, https://arxiv.org/abs/2302.13971.
[3] GUO T, CHEN X, WANG Y, et al. Large language model based multi-agents: a survey of progress and challenges [J/OL]. [2024-09-05]. ArXiv, 2024, https://arxiv.org/abs/2402.01680.
[4] SANTACROCE M, LU Y, YU H, et al. Efficient RLHF: reducing the memory usage of PPO [J/OL]. [2024-09-05]. ArXiv, 2023, https://arxiv.org/abs/2309.00754.
[5] LEE H, PHATALE S, MANSOOR H, et al. RLAIF vs. RLHF: scaling reinforcement learning from human feedback with AI feedback[C]//Proceedings of the international conference on machine learning. Hawaii: International Machine Learning Society, 2023.
[6] YANG J, LI A, FARAJTABAR M, et al. Learning to incentivize other learning agents [J/OL]. [2024-09-05]. ArXiv, 2020, https://arxiv.org/abs/2006.06051.
[7] SUN W, YAN L, MA X, et al. Is ChatGPT good at search? Investigating large language models as re-ranking agent [J/OL]. [2024-09-05]. ArXiv, 2023, https://arxiv.org/abs/2304.09542.
[8] HUANG Q, WAKE N, SARKAR B, et al. Position paper: agent AI towards a holistic intelligence [J/OL]. [2024-09-05]. ArXiv, 2024, https://arxiv.org/abs/2403.00833.
[9] OUYANG S Q, LI L. AutoPlan: automatic planning of interactive decision-making tasks with large language models[C]//Findings of the association for computational linguistics: EMNLP 2023. Stroudsburg: Association for Computational Linguistics, 2023.
[10] XUE S, ZHOU F, XU Y B, et al. WeaverBird: empowering financial decision-making with large language model, knowledge base, and search engine[J/OL]. [2024-09-05]. ArXiv, 2023, https://arxiv.org/abs/2308.05361.
[11] GUAN Y, WANG D, CHU Z, et al. Intelligent virtual assistants with LLM-based process automation[J/OL]. [2024-09-05]. ArXiv, 2023, https://arxiv.org/abs/2312.06677.
[12] YAO S, YU D, ZHAO J, et al. Tree of thoughts: deliberate problem solving with large language models[J/OL]. [2024-09-05]. ArXiv, 2023, https://arxiv.org/abs/2305.10601.
[13] SHINN N, CASSANO F, LABASH B, et al. Reflexion: language agents with verbal reinforcement learning[C]// Proceedings of the neural information processing systems. New Orleans: The NIPS Foundation, 2023.
[14] LI M H, ZHAO Y X, YU B W, et al. API-bank: a comprehensive benchmark for tool-augmented LLMs[C]//Proceedings of the 2023 conference on empirical methods in natural language processing. Stroudsburg: Association for Computational Linguistics, 2023.
[15] ZHANG Y, GE F, LI F, et al. Prediction of multiple types of RNA modifications via biological language model [J]. IEEE/ACM transactions on computational biology and bioinformatics, 2023, 20(5): 3205-3214.
[16] ZHOU G, GAO Z, DING Q, et al. Uni-Mol: a universal 3D molecular representation learning framework[C]//Proceedings of the international conference on learning representations. Kigali: The International Conference on Learning Representations, 2023.
[17] NIJKAMP E, RUFFOLO J A, WEINSTEIN E N, et al. ProGen2: exploring the boundaries of protein language models [J]. Cell systems, 2023, 14(11): 968-978.e3.
[18] KING R, ZENIL H. A framework for evaluating the AI-driven automation of science [M]. Paris: OECD Publishing, 2023.
[19] HUANG Y. Levels of AI Agents: from rules to large language models[J/OL]. [2024-09-05]. ArXiv, 2024, https://arxiv.org/abs/2405.06643.
[20] LU C, LU C, LANGE R T, et al. The AI scientist: towards fully automated open-ended scientific discovery[J/OL]. [2024-09-05]. ArXiv, 2024, https://arxiv.org/abs/2408.06292.
[21] CHU Z, WANG Y, ZHU F, et al. Professional agents - evolving large language models into autonomous experts with human-level competencies[J/OL]. [2024-09-05]. ArXiv, 2024, https://arxiv.org/abs/2402.03628.
[22] 李国杰. 智能化科研（AI4R）：第五科研范式[J]. 中国科学院院刊, 2024, 39(1): 1-9. (LI G J. AI4R: the fifth scientific research paradigm[J]. Bulletin of Chinese Academy of Sciences, 2024, 39(1): 1-9.)
[23] 刘细文, 付芸. 数智赋能背景下情报学研究进展[J]. 情报学进展, 2024(15): 131-171. (LIU X W, FU Y. The research advancements in information science amidst data & AI empowerment: data-driven, model-driven and knowledge discovery [J]. Advances in information science, 2024(15): 131-171.)
[24] 刘细文, 孙蒙鸽, 王茜, 等. DIKIW逻辑链下GPT大模型对文献情报工作的潜在影响分析[J]. 图书情报工作, 2023, 67(21): 3-12. (LIU X W, SUN M G, WANG X, et al. Analysis of the potential impact of GPT large model under DIKIW logic chain on documentation and information services[J]. Library and information service, 2023, 67(21): 3-12.)
[25] 刘细文. 贯彻落实二十大精神，开创文献情报工作的高质量发展道路[J]. 图书情报工作, 2023, 67(1): 4-8. (LIU X W. Implementing the spirit of the 20th national congress of CPC and planning the road of high-quality development of library and information service[J]. Library and information service, 2023, 67(1): 4-8.)
[26] 孙蒙鸽, 韩涛, 王燕鹏, 等. GPT技术变革对基础科学研究的影响分析[J]. 中国科学院院刊, 2023, 38(8): 1212-1224. (SUN M G, HAN T, WANG Y P, et al. Impact analysis of GPT technology revolution on fundamental scientific research[J]. Bulletin of Chinese Academy of Sciences, 2023, 38(8): 1212-1224.
[27] 刘细文. 情报学范式变革与数据驱动型情报工作发展趋势[J]. 图书情报工作, 2021, 65(1): 4-11. (LIU X W. Paradigm transformation of library and information science and trends of data-driven information services[J]. Library and information service, 2021, 65(1): 4-11.)
[28] 刘细文, 付芸, 韦华楠. 情报智能体——面向“十五五”的科技情报工作新范式[J]. 农业图书情报学报, 2024, 36(12): 22-36. (LIU X, FU Y, WEI H. DIS agent: new paradigm of S&T documentation and information service for fifteenth five-year-plan [J]. Agricultural library and information, 2024, 36(12): 22-36.)
[29] GHAFAROLLAHI A, BUEHLER M J. SciAgents: automating scientific discovery through multi-agent intelligent graph reasoning [J]. Advanced materials, 2024: 2413523.
[30] SHAO W, ZHANG R, JI P, et al. Astronomical knowledge entity extraction in astrophysics journal articles via large language models [J]. Research in astronomy and astrophysics, 2023, 24: 065012.
[31] OPENAI. GPT-4 Technical Report[J/OL]. [2024-09-05]. ArXiv, 2023, https://arxiv.org/abs/2303.08774.
[32] WANG L, MA C, FENG X, et al. A survey on large language model based autonomous agents[J]. Frontiers of computer science, 2024, 18(6): 186345.
[33] XI Z, CHEN W, GUO X, et al. The rise and potential of large language model based agents: a survey[J/OL]. [2024-09-05]. ArXiv, 2023, https://arxiv.org/abs/2309.07864.
[34] TANG X, JIN Q, ZHU K, et al. Prioritizing safeguarding over autonomy: risks of LLM agents for science[J/OL]. [2024-09-05]. ArXiv, 2024, https://arxiv.org/abs/2402.04247.
[35] KUHN T S. The structure of scientific revolutions [M]. Chicago: University of Chicago Press, 1962.
[36] CROWDFLOWER. Data science report [EB/OL]. [2024-09-01]. https://www2.cs.uh.edu/~ceick/UDM/CFDS16.pdf.
[37] BORNMANN L. The sound of science [J]. EMBO reports, 2024, 25(9): 3743-3747.
[38] USHER W, CAPLES A, KURATA K, et al. The future of intelligence analysis: US-Australia project on AI and human machine teaming [R]. Australia: Australian Strategic Policy Institute, 2024.
[39] 刘细文, 郭世杰. 情报认知模型库构建研究[J]. 农业图书情报学报, 2021, 33(1): 32-40. (LIU X W, GUO S J. A database construction of S&T intelligence cognition models [J]. Agricultural library and information, 2021, 33(1): 32-40.)
