DUSTIN VOLZ, CADE METZ2026年9月24日7月在上海举行的世界人工智能大会。中美两国正陷入一场人工智能竞赛，双方都认为这场竞赛对在军事、技术和经济领域占据上风至关重要。 Ng Han Guan/Associated PressThere is no shortage of issues on the table for possible discussion as President Trump prepares to welcome the Chinese leader, Xi Jinping, for a state visit beginning Wednesday evening, including the Iran war, Taiwan and tariffs.特朗普总统将在周三晚间起迎接中国国家主席习近平进行国事访问，届时将有很多问题需要讨论，包括伊朗战争、台湾和关税问题。But no topic has captured global fascination in recent weeks more than whether the two sides can reach any agreements around the pace of artificial intelligence development. The rapidly shifting debate about the risks of A.I. has so consumed Silicon Valley, Washington and, in recent days, the United Nations that some observers have likened the U.S.-China summit to a Cold War tête-à-tête during the nuclear arms race.但在最近几周，没有任何话题比双方能否就人工智能的发展速度达成共识更令全球关注。关于人工智能风险快速变化的辩论已彻底席卷硅谷和华盛顿，近日波及联合国，以至于一些观察家将此次美中峰会比作核军备竞赛期间的冷战一对一会谈。Though some see the stakes this week as nothing short of civilizational, the expectations for any serious accord remain low. The countries are locked in an A.I. race that both sides see as decisive to gaining the upper hand militarily, technologically and economically, and mutual suspicion runs deep, according to foreign policy analysts.尽管有人认为本周的博弈事关文明存亡，但人们对达成任何实质性协议的期望依然很低。外交政策分析人士表示，两国正陷入人工智能竞赛，双方都认为这场竞赛将决定谁能在军事、技术和经济上占据上风，而且双方的相互猜疑根深蒂固。Western experts generally estimate that the United States is somewhere around six months ahead of China in the A.I. race when it comes to its leading models. But that figure obscures the complex relationship between the two nations, further complicating efforts to reach agreement on how to address A.I. safety risks.西方专家普遍估计，在人工智能竞赛中，就领先模型而言，美国领先中国约六个月左右。但这个数字掩盖了两国之间复杂的关系，进一步增加了就应对人工智能安全风险达成共识的难度。Here is a look at where the United States has the upper hand against China on A.I. and where it is lagging behind its top geopolitical rival.以下是美国在人工智能领域相对于中国的优势所在，以及落后于其头号地缘政治对手的领域。历届美国政府都实施了出口管制措施，限制了中国获得高性能芯片的渠道，其中包括英伟达生产的芯片。Access to top-shelf computer chips is paramount.获取顶级计算机芯片至关重要。The United States has one undeniable advantage over China: computer chips. Most of the specialized chips used to build and operate A.I. technologies are designed by American companies. This includes the Silicon Valley chipmaker Nvidia, which is now the world’s most valuable company.相对于中国，美国有一个不可否认的优势：计算机芯片。用于构建和运行人工智能技术的大多数专用芯片都是由美国公司设计的，其中包括现在全球市值最高的公司——硅谷芯片制造商英伟达。“The people who have made the fastest chips are here,” said Andrew Feldman, the chief executive of the Silicon Valley chipmaker Cerebras. “And they are not in China.”“制造出最快芯片的人在这里，”硅谷芯片制造商Cerebras的首席执行官安德鲁·费尔德曼说。“他们不在中国。”Over the past decade, three American presidential administrations have imposed export controls that limited Chinese access to powerful chips. Even as China works to evade these controls, Chinese A.I. researchers often say that chip shortages are the biggest obstacle they face.过去十年里，三届美国政府都实施了出口管制，限制中国获取强大的芯片。尽管中国努力规避这些管制，但中国的人工智能研究人员经常表示，芯片短缺是他们面临的最大障碍。The leading Chinese companies have spent hundreds of millions of dollars to secure the chips they need. To bypass American export controls, they often buy remote access to data centers outside of China. Over the past year, this allowed China to shrink the performance gap between its systems and the leading American technologies.中国领先企业已经花费数亿美元来获取所需芯片。为了绕过美国的出口管制，他们经常购买对中国境外数据中心的远程访问权限。过去一年里，这使得中国缩小了其系统与美国领先技术之间的性能差距。At the same time, China is working to build its own chip-making industry. But by its own admission, Huawei, the telecommunications and electronics company that leads this effort, is still years away from producing chips that perform as well as those offered by Nvidia today.与此同时，中国正在努力建立自己的芯片制造产业。但引领这一努力的电信和电子公司华为也承认，距离生产出性能可与英伟达当前产品媲美的芯片仍需数年时间。Chinese A.I. companies say they are already using chips from Huawei, but these chips are unlikely to provide much help in building cutting-edge technologies. Less-powerful chips are typically used to serve technologies to users across the internet.中国人工智能公司表示，他们已经在使用华为的芯片，但这些芯片不太可能为构建尖端技术提供太多帮助。性能较弱的芯片通常用于通过互联网向用户提供技术服务。Even with help from Huawei, Chinese A.I. companies like Moonshot have struggled to find the additional chips they need to meet demand.即使有华为的帮助，像月之暗面这样的中国人工智能公司仍难以找到满足需求的额外芯片。As Americans sour on A.I., China faces fewer headwinds politically.美国人对人工智能日益感到反感，中国在政治上面临的阻力较小。Public polling in the United States shows that A.I. has become increasingly unpopular over the past year — so much so that it has now become a key issue in the midterm elections. Many Americans are worried about A.I. threatening their jobs, undermining education and, more recently, posing a threat to humanity. A Pew Research Center poll this week found that 60 percent of Americans were uncomfortable with a new local data center.美国的民意调查显示，人工智能在过去一年里变得越来越不受欢迎——以至于它现在已成为中期选举的一个关键议题。许多美国人担心人工智能会威胁他们的工作、破坏教育，而最近更担心它会对人类构成威胁。皮尤研究中心本周的一项民调发现，60%的美国人对新建本地数据中心感到不安。Though Mr. Trump has dismissed calls for an A.I. slowdown as a “hoax,” a growing number of Republican and Democratic lawmakers, governors and 2028 presidential hopefuls are scrambling to offer a wide range of proposals seeking to regulate the technology after disclosures about rogue A.I. agents hacking into other companies.尽管特朗普将放缓人工智能发展的呼吁斥为“骗局”，但在有关失控的人工智能智能体黑客入侵其他公司的事件曝光后，越来越多的共和党及民主党议员、州长以及2028年总统大选的有望参选者，正争先恐后地提出各种提案，试图对这项技术进行监管。By nature of its centralized systems, China does not need to contend with a boisterous public debate about how or whether to rein in A.I. And though polling in China can be dubious, surveys consistently point to more enthusiasm for its adoption.由于中国实行集中式管理体制，因此无需面对是否以及如何限制人工智能的激烈公众辩论。而且，尽管中国的民意调查可能值得怀疑，但各项调查始终显示，公众对采用人工智能的态度更为积极。Chinese citizens broadly appear to be much more optimistic about A.I. — and in fact are some of the most hopeful about the technology in the entire world, according to a KPMG survey of 47 countries last year. Some 69 percent of people in China said the technology’s benefits outweighed its risks, compared with just 35 percent of Americans who agreed. The technology is also more embedded in day-to-day life, including service robots in hotels and restaurants, as well as medical chatbots at hospitals.中国公民似乎普遍对人工智能要乐观得多——事实上，根据毕马威去年对47个国家进行的一项调查，他们是全世界对该技术抱有最大希望的人群之一。约69%的中国人表示，这项技术的好处大于风险，相比之下，持相同观点的美国人仅占35%。这项技术也更多地融入了日常生活，包括酒店和餐厅的服务机器人，以及医院的医疗聊天机器人。The U.S. economy has so far been more resilient to huge investments in A.I.到目前为止，美国经济在面对人工智能领域的巨额投资时表现出了更强的韧性。Another advantage for the United States is that its economy has been more resilient during the A.I. boom. Stock indexes are near record highs, and worries of an A.I. bubble also have so far not materialized. Fears of job losses prompted by the technology proliferating throughout corporate America have yet to really appear in economic data.美国的另一个优势是，在人工智能热潮期间，其经济更具韧性。股票指数接近历史高点，对人工智能泡沫的担忧迄今也未成为现实。尽管这项技术正在美国企业界广泛普及，但由此引发的失业担忧尚未在经济数据中真正显现。There are some signs that A.I. is contributing to slower wage growth in jobs most exposed to the technology and fewer job openings, especially for young workers. And applying to jobs, especially for entry-level workers, has become even more daunting thanks to A.I. being used to either write résumés or screen applicants. But overall, the economic distress directly linked to A.I. has been far less than many economists predicted.有一些迹象表明，人工智能正在导致受其影响最大的行业的工资增长放缓，职位空缺减少，尤其对年轻求职者而言更是如此。此外，由于人工智能被用于撰写简历或筛选申请人，求职过程变得更加艰难，尤其是对初级职位的求职者而言。但总体而言，与人工智能直接相关的经济困境远小于许多经济学家的预测。In China, the story is different. Its economic struggles, including high youth unemployment, consumer pullback on spending and a deflationary spiral, threaten to undercut Beijing’s vast investments in A.I. Chinese economists have recently issued warnings that the economy is vulnerable and potentially being made more precarious because of the country’s A.I.-centered allocation of spending.在中国，情况则有所不同。其经济困境——包括青年高失业率、消费者削减开支和通货紧缩螺旋——有可能削弱北京在人工智能领域的巨额投资。中国经济学家最近发出警告称，经济本就脆弱，且可能因为以人工智能为中心的财政支出配置而变得更加岌岌可危。周二，在北京举办的面向信息技术行业的展会中国国际信息通信展上，一款机器人成为焦点。The nature of A.I. models in the U.S. and China are drastically different.美中两国人工智能模型的性质截然不同。In the United States, the leading artificial intelligence companies, including OpenAI and Anthropic, maintain tight control over how their technologies are used. They build elaborate guardrails, hoping to prevent users from spreading disinformation or hacking into computer networks.在美国， OpenAI和Anthropic等领先人工智能公司对其技术的使用方式实施严格管控。它们建立了周密的防护机制，希望防止用户传播虚假信息或入侵计算机网络。As concern mounts over the risks of A.I., these companies are calling for federal regulations that would toughen this approach. They say the government should review their A.I. technologies before release to ensure that the proper guardrails are in place — and many policymakers, lawmakers and pundits agree.随着对人工智能风险担忧的加剧，这些公司呼吁制定联邦法规，以强化这种做法。他们表示，政府应在人工智能技术发布前进行审查查，以确保适当的防护措施到位——许多政策制定者、立法者和专家对此表示同意。But the leading Chinese companies take a different stance. Companies like DeepSeek, Moonshot and Z.ai typically make their most powerful technologies “open source,” meaning anyone is free to use and modify these systems as they see fit.但中国领先的公司采取不同的立场。像深度求索、月之暗面和智谱AI这样的公司通常将其最强大的技术“开源”，这意味着任何人都可以自由地根据需要使用和修改这些系统。Open source has driven the development of computer software, the internet and artificial intelligence for decades. The idea is that technology advances faster when its computer code is freely available for anyone to examine, use and improve upon.几十年来，开源一直推动着计算机软件、互联网和人工智能的发展。其理念是，当计算机代码可供任何人自由查阅、使用和改进时，技术进步会更快。In open-sourcing their technologies, Chinese companies can essentially combine their resources and accelerate their efforts. And if U.S. regulators stifle the progress of American open-source projects — as some lawmakers are suggesting they should — China could gain a significant edge.通过开源其技术，中国企业实际上可以整合其资源，加速研发进程。而且，如果美国监管机构扼杀美国开源项目的进展——正如一些立法者建议的那样——中国可能会获得重大优势。If the best open-source systems come from China, many experts contend, developers across the globe will build atop Chinese technologies. This is already starting to happen, even in Silicon Valley.许多专家认为，如果最好的开源系统来自中国，全球各地的开发者将基于中国技术进行开发。这种情况已经开始发生，甚至在硅谷也是如此。“These models are still rough around the edges, but there are tasks where they are on par with the leading American systems,” said Rehaan Ahmad, a co-founder of the Silicon Valley start-up alphaXiv, who has used the latest Chinese technologies for the past several months.“这些模型在细节上仍显粗糙，但在某些任务上，它们与领先的美国系统不相上下，”硅谷初创公司alphaXiv的联合创始人雷汉·艾哈迈德说，他在过去几个月里一直在使用最新的中国技术。Anyone can use these open-source technologies for malicious purposes, including, most notably, cyberattacks. But security experts say this is balanced by another phenomenon: Anyone can use the same technologies to defend against malicious attacks.任何人都可以将这些开源技术用于恶意目的，其中最引人注目的是网络攻击。但安全专家表示，另一种现象抵消了这一风险：任何人都可以使用相同的技术来防御恶意攻击。China could move ahead on electric power.中国在电力领域可能取得领先。Some believe that if China can build a viable chip industry, it could race ahead of the United States because of an advantage it has developed in another area: electrical power.一些人认为，如果中国能建立起一个可行的芯片产业，它可能会超越美国，因为它在另一个领域发展了优势：电力。Packed into giant data centers by the thousands, the leading chips require enormous amounts of electricity. They need so much electrical power, the big U.S. companies cannot get what they need from the existing power grid. Increasingly, they are turning to wind power or nuclear or even gas turbines installed alongside their data centers.成千上万的领先芯片密集部署在巨型数据中心内，消耗着巨大的电能。它们需要的电力如此之多，以至于美国大公司无法从现有的电网中获得所需的一切。它们日益转向风能或核能，甚至在数据中心旁边安装燃气轮机。“At a fundamental level, the tech industry is trying to convert electricity into intelligence,” said Mr. Feldman, the Cerebras chief executive. “It’s like a Dr. Seuss machine. In goes chips and power — and out comes intelligence.”“从根本上说，科技行业正试图将电力转化为智能，”Cerebras首席执行官费尔德曼说。“这就像一台苏斯博士笔下的机器。一边输入芯片和电力，另一边输出智能。”Although the United States has a significant edge in terms of chips, Mr. Feldman said, he believes that China could pull ahead because it has spent years rebuilding its electrical grid. “That is going to be a huge asset,” he said. “Our grid is 1950s technology.”费尔德曼表示，尽管美国在芯片方面拥有显著优势，但他认为中国有可能后来居上，因为它花了数年时间重建电网。“这将是一笔巨大的资产，”他说。“我们的电网还是20世纪50年代的技术。”China leads in producing A.I. talent.中国在培养人工智能人才方面处于领先地位。As far back as 2024, studies showed that by some metrics, China had eclipsed the United States as the biggest producer of A.I. talent, generating almost half of the world’s leading researchers. By comparison, only about 18 percent came from the United States.早在2024年，研究就表明，按照某些指标，中国已经超过美国，成为最大的人工智能人才出产国，培养了世界上近一半的领先研究人员。相比之下，只有约18%来自美国。Researchers born and educated in China have for years played major roles inside leading U.S. artificial intelligence labs. And as newer studies have shown, they continue to drive important A.I. research across American industry and academia. But a growing number of them now stay in China, working for companies like Moonshot, Z.ai and DeepSeek.出生在中国、并在中国接受教育的研究人员多年来一直在美国顶尖的人工智能实验室中发挥着重要作用。较新的研究表明，他们继续推动着整个美国工业界和学术界的重要人工智能研究。但如今，他们当中越来越多的人选择留在中国，为月之暗面、智谱AI和深度求索等公司工作。Dustin Volz为时报撰写网络安全和情报方面的文章，他常驻华盛顿。Cade Metz是时报记者，撰写有关人工智能、无人驾驶汽车、机器人、虚拟现实和其他新兴技术领域的报道。翻译：纽约时报中文网点击查看本文英文版。获取更多RSS：https://feedx.net https://feedx.site