白鹭

从“工具”到“搭档” AI应怎样助力科研,从“工具”到“搭档” AI应怎样助力科研_我的网站

工商银行

A | Healthier Cookies Made from Banana Waste,Researchers from the Aligarh Muslim University in India and the University of Houston in the US discovered a new use for banana peels that normally go to waste, suggesting they may be turned into flour and then added to your favorite cookie recipe.,As it turns out, adding banana peel flour improves the nutritional value of cookies and imbues them with antioxidant properties useful for your health. There is slight downside though – the cookies became tougher the more banana flour you add, but miraculously soften after a certain period in storage. So, healthy cookies, anyone?,New Internet Highs (and Emergency Services),Rejoice the pilgrims to Kilimanjaro (Africa's highest mountain, in case you didn’t know), for the authorities of Tanzania announced the launch of a new project that will enable high-speed internet access on your mobile devices as you climb to the peak. Although the summit itself will be connected later on, the plan is to finish the project in October this year.,This is good news not only for those who love to post to their social media on the go, but also for those who love getting themselves in trouble! Previously, there was no cell service on the mountain and calling 911 in an emergency was kind of problematic. But not anymore!,Looking With a Fresh Pair of Eyes,Damage or degradation of corneas is no small issue – according to the scientists' estimates, some 12.7 million people around the world suffer vision loss due to it. And while cornea tissue is transplantable, there is not nearly enough donated tissue to serve all who faced this problem.,However, scientists from India and Iran managed to restore sight in 20 test patients during a trial of new bio-engineered (meaning you can always produce more) corneal tissue. It is made from the collagen of pig skin - not the rarest material in the world. Naturally, it will take time to scale the technology and thoroughly test it, but it's a success nonetheless.,Nanoparticles That Can Treat Cancer,Scientists of the Immanuel Kant Baltic Federal University in Russia have developed new nanoparticles made of gold and iron oxides that can destroy the cells of a tumor from the inside, while sparing the healthy ones, sources in the Russian Science Ministry told Sputnik. These particles, developed jointly with foreign scientists, can heat up and thus destroy cancerous cells.,The new technology tested well against cancer cells in breasts and walls of blood vessels. It also showed excelled biocompatibility – meaning it is not toxic for a human organism.。　　近日，2026科学智能大会在北京举办。

B | 作为驱动原始创新、重塑科研范式的核心引擎，人工智能驱动的科学研究(AI for Science，AI4S)成为热议话题。

C | 　　当前，AI4S底层基础设施逐步完善，科研智能体与自主科研系统实现跨越式进展。从答题工具转向自主推进科研任务，AI会为科研带来什么变化？　　拓展边界：为人类科学家提供更多探索灵感　　今年3月发布的《AI for Science创新图谱2026》显示，近五年全球相关论文发表量增长了一倍多，航空航天、量子科技、材料科学等领域年均增速超过30%。　　以往，研究人员可能需要花费半年时间阅读数百篇论文才能完成一篇文献综述，如今借助AI只需要几小时；面对海量数据，传统方法耗时数天甚至数周的计算任务，现在几秒钟便能得出结果。人工智能在可形式化、可验证的科学任务上，效率远超人类。

D | 　　“与人类科学家相比，AI可以掌握更多知识，不仅能把科研人员从重复、耗时的工作里解放出来，还能拓展视野边界，发现更多新奇现象，提供更多探索灵感。”中国科学院自动化研究所研究员李林静说，目前，科研智能体已经能够完成文献研读、实验规划、参数迭代、结果分析的闭环自主科研流程。

E | 　　在战略布局上，各国纷纷加速：中国发布《关于深入实施“人工智能+”行动的意见》，将“人工智能+”科学技术列为6大重点行动之一；美国推出“创世纪计划”，统筹整合全国科研基础设施，依托人工智能全面提速前沿科学突破；欧盟发布专项战略汇集资源以研发AI并将其应用于科学。　　中国科学院软件研究所研究员黄进说，在以AI驱动的科研范式下，学科融合成为趋势，AI4S已经深入到许多重大科研计划和基础科研项目的核心环节。　　“看似合理”：验证瓶颈让科研成果难以落地　　AI4S为许多领域的研究打开了新空间。　　在生命科学领域，AI可设计靶点分子、筛选化合物，将候选药物发现从数年缩短至数月；在材料领域，AI能辅助设计催化剂、筛选验证配方组合，加快材料研发从经验试错走向数据与机理协同驱动。　　谷歌公司旗下DeepMind使用AI工具预测出了220万种新的晶体，其中38万种具有稳定结构。然而，这些预测结果得到实验验证的不到0.2%。这一现象凸显出AI4S领域面临验证瓶颈，大量AI预测结果难以完成实验验证，阻碍了计算发现走向实际工程应用。

F | 　　中国工程院院士、计算机专家李国杰指出，AI产生海量“看似合理”的假说，绝大多数无法被验证，导致科研资源浪费在低优先级选项上。波士顿咨询公司的报告也显示，全球AI制药投入数百亿美元，至今尚未有一款获批上市。

G | 　　打破科研与产业之间的界限，已经成为AI4S下一步迫切需要解决的问题。中国科学院院士、北京大学数学科学学院讲席教授鄂维南认为，这仅靠科研能力的提升是不够的，还必须在评价机制、组织形式上下功夫。　　人机协同：更需人类科学家守持科学品位　　推动自主实验室智能化改造，实现AI对科研全流程各个环节的深度赋能，已经形成全球共识。但是，随着前沿模型自主能力的持续提升，AI技术的“双刃剑”效应也随之放大。　　李国杰指出，现有主流AI系统本质上仍是外源目标驱动的优化体系，即使未来AI具备更强的目标更新和自我改进能力，其评价准则的正当性、科学意义的判定以及风险责任的承担，仍不能由系统内部完全闭合地解决。

H | 　　北京大学科学智能学院副院长莫凡洋认为，从长远看，AI或将具备自主设计科学实验的能力，但现阶段仍需由科学家提出核心科学问题、划定探索方向，同步配套完善安全约束规则、人工复核验证流程与风险处置机制。　　不少与会专家表示，未来更可能出现的是“AI主导执行过程、人类科学共同体保留最终判断权”的混合科研范式，而不是取消人类评价权的全自动科研体系。　　可以期待，依托AI4S的持续发展，人机协同的科研模式将不再是科幻作品中的想象。　　“人和AI之间，应该是互相成就的关系。”莫凡洋说，在科研探索的道路上，科学品位、实验验证与责任判断始终是人类科学家不可替代的核心价值。