我国也在加快构建AI科学家系统。2024年10月，上海人工智能实验室联合多家单位开源了多智能体科 学社群模拟系统——"虚拟科学家"，可以模拟人类科学家的合作过程来研究科技创新规律；2025年7 月，联合崖州湾国家实验室和中国农业大学等单位，发布了首个生物育种领域的自主科学发现系统"丰 登·基因科学家"。在该系统的辅助下，科研人员在主粮作物中发现了数十个此前未被报道的基因功能， 并获得实验证实。 尽管AI科学家在技术上已取得阶段性成果，但在可靠可控、人机关系、科研伦理、合规监管等诸多方 面仍有大量问题亟待解决。科学界长久以来关于"AI是否会取代人类""AI是否会弱化人类独立思考能 力"等问题的争论也在持续。不过可以确定的是，人类社会正不断朝着通用人工智能的时代迈进。未 来，AI科学家技术有望与机器人和传感器深度结合，通过传感器感知实验环境并借助机器人执行实 验，从而完成从假设提出到实验验证的全自主科学发现闭环。 可以预见，未来将有更多具备实体形态的"AI科学家"辅助人类在复杂环境下开展科学研究："AI作物遗 传学家"在盛夏的稻田中检查水稻生长状况，"AI土壤学家"在月球基地的无氧环境下分析月壤，"AI核物 理学 ...

Group 1 - The core idea of the article revolves around the emergence of AI-driven virtual scientists that collaborate to develop treatment strategies for diseases like Alzheimer's, showcasing a new trend in scientific research [1][2] - Multiple institutions, including Google's DeepMind and Stanford University, are developing virtual laboratory systems powered by AI agents to enhance research efficiency and creativity [2][3] - These AI systems utilize large language models (LLMs) that can autonomously perform tasks such as information retrieval and code execution, representing a shift towards "agent-based AI" systems [3] Group 2 - Concerns exist regarding the reliability of AI-generated ideas, with current systems facing issues like "hallucinations" or generating incorrect information, though the introduction of reviewer roles can enhance reliability [4] - Research indicates that multi-agent collaboration outperforms single AI systems, with the inclusion of a reviewer improving the accuracy of responses in scientific applications [4][5] - The effectiveness of AI agents in generating novel ideas is debated, with some researchers finding AI suggestions to be innovative while others view them as lacking originality [6][7] Group 3 - AI systems are currently seen as research assistants that help summarize information, inspire new ideas, and improve efficiency, but their potential to generate groundbreaking concepts remains to be validated over time [7] - The widespread adoption of AI collaborative systems in research is anticipated, similar to the integration of search engines, although they are not expected to replace human researchers [7]