ScienceOne: AI Empowers New Paradigm in Research

On April 29, the Institute of Automation, Chinese Academy of Sciences (IACAS) unveiled ScienceOne, an intelligent research platform based on scientific foundational large models, at the Eighth Digital China Construction Summit. Deputy Director of IACAS, Dr. Zeng Dajun, demonstrated two initial products: S1-Literature, the literature assistant, and S1-ToolChain, the scientific tool scheduling platform. The integration of AI into scientific research to accelerate discoveries and reshape methodologies is a critical frontier in global technological development. By leading a collaborative effort involving several prestigious research institutes, including the Institute of Computer Network Information, the National Science Library, the Institute of Mathematics and Systems Science, and the Institute of High Energy Physics, as well as industrial partners such as ZK WenGe and ZK Zidian TaiChu, IACAS has pioneered an architecture-integrated scientific foundational model. This model serves as the backbone of the ScienceOne platform, which aims to support interdisciplinary innovation and promote a new paradigm of intelligent scientific research. The ScienceOne platform addresses the current limitations of AI in scientific research, where most applications involve fine-tuning general large models or reinventing the wheel within individual disciplines. These approaches often fail to overcome challenges such as the strong hallucination tendency of general models, weak domain-specific knowledge, and poor logical reasoning capabilities. ScienceOne, however, leverages its scientific foundational model to meet universal research needs across various fields. It achieves significant advancements in data understanding, computational optimization, and reasoning evaluation, thereby enabling scalable and systematic support throughout the entire research process—from hypothesis formulation to simulation and experimental validation. S1-Literature: A Comprehensive Literature Assistant S1-Literature is a joint effort between IACAS and the National Science Library. It utilizes China's largest database of scientific literature and real-time open-source materials to generate high-quality literature reviews and summaries. With capabilities that include deep comprehension of common scientific data types and precise interpretation of formulas and specialized language, S1-Literature simplifies the task of writing comprehensive literature reviews. Users can issue brief commands like "Write a literature review on..." and the assistant will automatically organize the review framework, sift through thousands of papers, and generate detailed content. For in-depth reading, it offers tools such as mind maps, citation tracing, research mapping, and extraction of key technical pathways. Additionally, the assistant supports multidisciplinary knowledge Q&A and data interpretation. Currently, S1-Literature is adapted for use in mathematics, physics, and materials science, with plans to expand dynamically to cover all scientific disciplines. This versatility ensures that researchers from various fields can benefit from its advanced features and capabilities. S1-ToolChain: An Autonomous Scientific Tool Scheduler S1-ToolChain is designed to autonomously coordinate cross-disciplinary data understanding, scientific computation, and simulation tools. Using standardized scientific model protocols, it integrates a wide range of general and specialized models and tools, orchestrating them through intelligent agents to perform tasks sequentially. The platform currently encompasses nearly 300 multi-modal scientific data analysis tools, differential equation solvers, discrete optimizers, and multiscale simulators for fields such as mathematics, physics, and engineering. An example of its functionality is the completion of protein sequence predictions. When a researcher inputs a command like "Complete protein sequence alignment," S1-ToolChain uses its foundational model to identify the research intent, employs a custom sequence understanding model to analyze the sequence structure, and then plans and executes the sequence completion task. It further utilizes scientific computing tools and the ESM3 professional model to complete the alignment, ensuring a seamless and efficient research workflow. Leveraging Institutional Strengths for AI-Driven Research Innovation The integration of AI and scientific research has already produced breakthrough results in areas such as biology, materials science, and pharmaceuticals, fostering the emergence of a new generation of multidisciplinary research paradigms. IACAS, recognizing the pivotal role of this transformation, is leveraging its institutional strengths in systematic discipline organization and a pool of top-tier multidisciplinary researchers. The scientific foundational model and the ScienceOne platform together provide robust AI technologies and support across all scientific disciplines. Looking ahead, the research team plans to open-source the scientific foundational model, designated S1-Base, and release the ScienceOne agent factory, S1-Agent. These initiatives will form a comprehensive suite of platform-based tools, accelerating the development of the AI4Science paradigm. By doing so, IACAS aims to empower scientists with cutting-edge AI solutions, facilitating more effective and innovative research practices.