一 | 21世纪经济报道记者 陶力上海报道 业界基本以两条并行路径深入物理世界：一条是从工业数据治理出发、向工艺与供应链渗透的“云端赋能”路线；另一条是从底层架构重构、让AI驻留终端的“端侧原生”路线。 2026世界人工智能大会（WAIC）期间，多家技术服务展商集中展现了AI深入物理世界的策略。 当前行业主流端侧AI方案，普遍采用 “云端预训练 + 终端裁剪压缩” 技术模式，该方案已成熟应用于各类数字化场景。但物理场景对AI系统提出了低延迟实时响应、离线独立运行、本地数据隐私安全等刚性要求，传统云端迁移式方案仍存在短板，行业亟需底层技术架构创新，支撑物理AI的规模化落地。 标准共建先行 在“端侧流式多模态模型赋能 AI 硬件爆发”分论坛上，Om AI联汇CEO兼首席科学家赵天成指出，物理世界实时、动态、高频的交互特性，决定了端侧AI不能是云端模型的“简易减法版”，须基于物理场景重构底层架构与运行逻辑。“物理世界并非云端数字世界AI的简单延伸，端侧智能需要建立专属的架构思路、评估体系、安全框架和产业生态。” 当前端侧AI产业面临硬件架构繁杂、接口标准不统一、软硬协同成本高等碎片化问题，与多元算力产业早期困境高度相似。 统一标准体系建设，成为推动端侧原生从实验室验证，走向规模化商用的关键抓手。 为此，论坛上发起《端侧多模态基座与AI硬件协同发展倡议》，倡议提出三个共建方向：一是共建端侧原生技术标准，推动评估体系、接口规范、安全框架标准化；二是共研软硬一体联合方案，推动模型-芯片协同、端云协同边界探索；三是共拓物理AI产业场景，推动重点赛道数据反馈机制与开放生态建设。 Are 3D mammograms better than standard 2D imaging for catching advanced cancers?A clinical trial is recruiting thousands of volunteers — including a large number of Black women who face disparities in breast cancer death rates — to try to find out.People like Carole Stovall, a psychologist in Washington, D.C., have signed up for the study to help answer the question.“We all need a mammogram anyway, so why not do it with a study that allows the scientists to understand more and move closer to finding better treatments and ways of maybe even preventing it?” Stovall said.The underrepresentation of women and minorities in research is a long-simmering issue affecting health problems including Alzheimer’s disease, stroke and COVID-19. Trials without diversity lead to gaps in understanding of how new treatments work for all people. “Until we get more Black women into clinical trials, we can’t change the science. And we need better science for Black bodies,” said Ricki Fairley, a breast cancer survivor and advocate who is working on the issue.Black women are 40% more likely to die from breast cancer than white women and tend to be diagnosed younger. But it’s not clear whether 3D mammography is better for them, said Dr. Worta McCaskill-Stevens of the National Cancer Institute.“Are there populations for whom this might be important to have early diagnosis?" asked McCaskill-Stevens. "Or is it harmful,” causing too many false alarms or unneeded follow-up tests and treatments?McCaskill-Stevens, who is Black, leads NCI’s efforts to boost access to cancer research in minority and rural communities. She has joined the study herself.The newer 3D technique has been around for a decade, but there’s never been conclusive evidence that it's better than 2D at detecting advanced cancers. The screening technique combines multiple pictures of the breast taken from different angles to create a 3D-like image. Both 3D and 2D mammograms compress the breast and use low doses of radiation.Prior studies suggest that 3D finds more cancers than 2D, but catching more cancers doesn’t necessarily mean more lives saved. Some cancers missed by standard screening may not progress or need treatment. Previous studies did not randomly assign patients to a screening method, the gold standard for research.The notion “that if it’s new, it’s shiny, then it’s better,” isn’t necessarily true, McCaskill-Stevens said. “Until we have the evidence to support that, then we need well-designed randomized trials.”The trial has enrolled nearly 93,000 women so far with a goal of 128,000. The NCI-funded study is now running in Canada, South Korea, Peru, Argentina, Italy and 32 U.S. states. A site in Thailand will soon begin enrolling patients.“We added more international sites to enhance the trial’s diversity, particularly for Hispanic and Asian women,” said Dr. Etta Pisano, who leads the study. Overall, 42% of participants are Hispanic. As recruiting continues, enrolling Black women and other women of color will “absolutely” continue as a priority, Pisano said.Participants are randomly assigned to either 2D or 3D mammograms and are followed for several years. The number of advanced cancers detected by the two methods will be compared.At the U.S. study sites, 21% of study participants are Black women — that's higher than a typical cancer treatment study, in which 9% of participants are Black, McCaskill-Stevens said.The University of North Carolina has signed up more Black women than any other study site. Nearly a quarter of the nearly 3,000 women enrolled at UNC’s two locations are Black.“Women in North Carolina want to take part in something that’s bigger than them,” said Dr. Cherie Kuzmiak, who leads the UNC arm of the study. “They want this active role in helping determine the future of health care for women.”In Washington, D.C., word of mouth has led to successful recruiting. A chance encounter at her hair salon persuaded Stovall to join the research. While waiting for a hair appointment, she met Georgetown University cancer researcher Lucile Adams-Campbell. The two, both Black, started chatting.“She explained how important it was to get women of color into the program,” said Stovall, who jumped at the chance to catch up on her mammograms after the COVID-19 pandemic delayed screening for her and thousands of others.For Stovall, there was a personal reason to join the research. Her sister recently completed treatment for triple negative breast cancer, an aggressive type that affects Black women at higher rates than white women. Women ages 45 to 74 without a personal history of breast cancer are eligible for the study, which launched in 2017. Many women also are providing blood and cheek swab samples for a database that will be mined for insights.“It’s a dream that people had since the beginning of screening that we wouldn’t fit everybody into the same box,” Pisano said. The study's findings could “reduce disparities if we’re successful, assuming people have access to care.”Stovall, 72, had a brief scare when her mammogram, the traditional 2D type, showed something suspicious. A biopsy ruled out cancer.“I was extremely relieved,” Stovall said. “Everybody I know has heard from me about the need for them to go get a mammogram.”___The Associated Press Health and Science Department receives support from the Howard Hughes Medical Institute’s Science and Educational Media Group. The AP is solely responsible for all content.。 据悉，倡议面向全行业开放，旨在凝聚科研机构、硬件厂商、方案商、开发者和产业投资人力量，打破技术壁垒、消除行业孤岛，推动端侧AI产业规范化发展。 在嘉宾对话环节，行业资本从产业投资视角，剖析了消费端侧智能的爆发逻辑，明确核心行业判断：在通用算力底座日趋完善的当下，智能终端产业的核心稀缺资源，已转变为适配硬件迭代规律、支撑超低延迟实时交互的原生多模态能力，这也是AI PC、智能穿戴等终端硬件实现规模化普及的关键前提。
二 | 工业AI落地门槛 制造业AI落地，门槛不是模型有多强，而是数据能不能被读懂。一辆整车涉及上万个零部件，数据散落在数百个系统里，格式混乱、时序不同步是常态。 本届WAIC上，广域铭岛将“懂工业”的基因转化为数据治理壁垒，其平台内嵌的数据治理引擎可自动识别数百种工业协议，将杂乱数据转化为高质量资产；一线人员拖拽组件即可搭建预测性维护、根因分析等应用，不再依赖算法工程师。 在数据基座搭建完成后，AI向上游渗透至最具挑战性的工艺环节。
三 | 未来，智能体更像是懂业务的“数字行动者”——把确定性交给智能体，将创造力留给工程师。 在国内某知名电器制造企业，库存分析智能体让全库检查效率提升99%，根因定位准确率达95%，库存管理从“人找问题”转向“问题找人”。 当云端AI在产线上完成数字化闭环，更深层的命题随之而来：如何让智能延伸到现实中的感知、思考和执行闭环。这正是行业参与者共同思考的问题。 从产业验证看，端侧AI已从可选升级变为刚需配置。在低空经济、海事作业等场景中，智能设备必须具备本地自主运行能力。 实践，是工业AI未来唯一的标尺。
四 | 制造业AI落地，门槛从来不是模型有多强，而是数据能不能被读懂。AI能力正在快速普及，但把AI能力组织成工作流、嵌入具体场景、最终变成商业回报，依然是一道很难跨越的专业壁垒。
五 |
Current article:http://zc1e3.yinkataizengquge.bond/list_166/8k3.html
Published on:18:24:59