在2026年10月15日于上海举行的2026未来科学·创想季跨界对话中，来自中国科学院计算技术研究所、北京大学、上海交通大学等机构的专家就人工智能的能力边界进行了深入探讨。1中国科学院计算技术研究所副所长包云岗指出，AI能力的深度取决于使用者的知识深度，虽然AI的随机性可用于探索多样性，但需要通过严格流程加以约束以保障可靠性，最终可验证性决定了AI的适用边界。1
与会专家认为，AI正在深刻改变科学研究的方式。北京大学的董彬表示，AI正在改变数学研究，将科研人员从繁重的证明工作中解放出来，数学家应主动学会驾驭智能体、甄别模型幻觉，实现与AI的协同进化。1然而，董彬也指出AI目前擅长知识储备与推理，但提出高价值、有意义的数学猜想仍是人类数学家的核心优势。1OpenAI公布的大量数学证明结果中仍存在不少错误。1对于AI的未来地位，上海交通大学的张娅表示，AI可能像数学一样，成为每个人从小就开始学习的基础内容。1
Leading researchers gathered for an interdisciplinary dialogue to examine the limits and potential of artificial intelligence in shaping next-generation digital infrastructure. The discussion, held in Shanghai on October 15, 2026, brought together scholars from China's top institutions to debate AI's role in advancing science and mathematics.1
The depth of AI capabilities fundamentally depends on the knowledge depth of its users, according to Bao Yungang, Vice Director of the Institute of Computing Technology at the Chinese Academy of Sciences.1 While AI's randomness can enable exploration of diverse possibilities, this quality must be constrained through rigorous processes to ensure reliability, and verifiability ultimately determines the boundaries within which AI can be applied.1
Dong Bin from Peking University emphasized that mathematicians must actively embrace AI and learn to manage intelligent systems while identifying model hallucinations, enabling human-AI co-evolution.1 However, Dong noted that despite AI's growing proficiency in knowledge storage and reasoning, formulating high-value and meaningful mathematical conjectures remains a core human advantage in mathematics.1 This observation gains significance given that OpenAI has disclosed numerous errors persisting in its published mathematical proofs.1
Looking forward, Zhang Ya from Shanghai Jiao Tong University suggested that AI may eventually become a foundational subject taught to everyone from childhood, much like mathematics itself.1 The broader consensus among participants reflected a recognition that AI is transitioning from foundational infrastructure to edge intelligence applications across various domains.1
评论
还没有评论，欢迎留下第一条。