DAVID WALLACE-WELLS2025年8月28日 Ibrahim RayintakathLast week, I wrote about how predictions about the near future of A.I. are changing: In place of the apocalyptic language of a year or two ago, you see a growing consensus that A.I. should be understood as a “normal technology.” Others have been more pessimistic, suggesting the A.I. bubble is set to pop, and still others have defended the underlying narrative of progress — arguing that one somewhat disappointing model release shouldn’t dampen enthusiasm for A.I. too much, or that we may have generally landed in a “Goldilocks” zone for A.I., far from doomer visions of machine takeover with remarkable gains being made nevertheless.上周，我探讨了人们对人工智能近期发展的预测正在经历变化：一两年前流行的那种末世论调正被一种日益增强的共识所取代，即人工智能应该被理解为一种“常规技术”。也有人持更悲观的态度，认为人工智能泡沫即将破裂；与此同时，还有人继续为“进步”的基本叙事辩护——他们认为，一次较为令人失望的模型发布不应过度打击人们对人工智能的热情，或者说，我们或许总体上已经进入了一个“恰到好处”的区间，既远离了机器接管的末日愿景，同时仍在不断取得显著的进步。But right-sizing our expectations for A.I. isn’t just a question of Nvidia share prices, or even short term G.D.P. growth. And it’s not a low-stakes play for the predictions markets, either. From a certain vantage it looks like one half of the central geoeconomic story of our time. The global landscape is now inarguably dominated by two rival superpowers, the United States and China, and in the post-pandemic period, one of them gambled hugely on A.I., while the other wagered hugely on green technology. The future now hangs in the balance while we wait for the bets to come in.全球格局现在无可争议地由两个竞争超级大国——美国和中国主导，而在后疫情时代，其中一方在人工智能上押下了重注，另一个则在豪赌绿色技术。未来悬而未决，我们都在等待赌局揭晓。Perhaps you have seen some version of the charts documenting the evolving shape of the world’s energy transition: Over the past few years, China has so dominated the global production and installation of clean technology that according to certain estimates, it is responsible for 74 percent of all wind and solar projects under construction. This dominance is a result of the country’s “Made in 2025” industrial policy, announced in 2015, but it really took off when the pandemic hit. After initial lockdowns, China was riding high on Zero Covid and the rest of the world was still overwhelmed by the threat of infection and the many burdens that fighting it imposed on us. Today, the dominance is up and down the supply chain: China is responsible for at least 60 percent of global production for almost all wind, solar and battery energy manufacturing capacity; for many of the most critical ones, its share is 80 percent or 90 percent.或许你已经见过某些版本的世界能源转型走势图：过去几年间，中国在全球清洁技术的生产和安装领域占据了主导地位，据某些估算，全球在建风电和太阳能项目中有74%来自中国。这种主导地位源于2015年宣布的“中国制造2025”产业政策，但真正腾飞是从新冠疫情暴发开始。最初的封锁之后，中国凭借“清零”政策高歌猛进，而世界其他地区仍深陷感染威胁及抗疫重负之中。如今这种主导地位已贯穿整个产业链：中国占全球风电、太阳能和电池制造的几乎全部产能的至少60%；许多关键组件的份额更是高达80%或90%。These charts are often presented as illustrations of America’s decline, casting a very particular imperial mood. But there are similar ones for A.I. that show the United States dominating. The United States hosts three-quarters of all global A.I. supercomputer computational performance, for instance, according to an analysis by Epoch A.I. And while that is just one estimate, with some limitations, other measures show similar patterns. Last year, American investors spent at least 11 times as much on A.I. as any other country. After tallying up totals for the 15 largest global players, the United States accounted for more than 75 percent of that investment.这些图表常常被拿出来作为美国衰落的例证，营造出一种非常特殊的帝国情绪。但也有类似的人工智能图表显示美国占据主导地位。例如，根据Epoch AI的分析，美国拥有全球四分之三的人工智能超级计算机算力。虽然这只是一个有局限的估计，但其他衡量标准也显示出类似的情况。去年，美国投资者在人工智能上的支出至少是其他任何国家的11倍。在对全球15个最大参与者的投资总额进行统计后，会发现美国的投资占比超过75%。This summer, my colleagues in the newsroom published an eye-opening comparison of the energy exports of the two countries, showing that China was obliterating the United States in terms of green tech and America was racing ahead with fossil fuel products. The shorthand implication was that the United States was becoming a petrostate, or at least much more of one, while the Chinese were well on their way to becoming what futurists in Silicon Valley like to call “the world’s first electrostate.” (Sometimes the phrase they invoke is “Type I Civilization,” a term borrowed from a starry-eyed Russian astrophysicist to mean a technological culture capable of harnessing all of a planet’s naturally available energy.)今年夏天，我的新闻编辑部同事发表了一篇令人大开眼界的文章，对两国的能源出口进行了比较，结果显示，中国在绿色技术领域碾压美国，美国则在化石燃料产品方面遥遥领先。简而言之，美国正在成为一个石油国家，或至少愈发倾向于此；而中国正稳步成为硅谷的未来学家所称的“全球首个电动化国家”。（他们有时会援引“Ⅰ型文明”的说法，这个术语源自一个俄罗斯天体物理学家的浪漫构想，意指一种能够利用行星全部自然能源的科技文明。）The comparison is striking, and illuminating, though in several ways the full picture is a bit more complex — these are, after all, not single-industry economies but enormously complex economic hegemons. However miraculous China’s green-tech boom has been, for instance, in aggregate it has hardly made up for the self-imposed contraction of its real estate sector. There are signs the country’s breakneck green energy buildout is slowing, thanks to pricing reforms. And however impressive the rollout of clean energy looks from the perspective of the rest of the world, the country is still burning through gobsmacking amounts of fossil fuel — four times as much coal last year than the world’s second-largest consumer, even as the country as a whole appears to to be reaching a peak in carbon emissions.这种比较引人注目，也富有启发，尽管其全貌在某些层面要更为复杂一些——毕竟，美中不是单一产业经济体，而是极其复杂的经济霸主。例如，尽管中国的绿色技术蓬勃发展令人惊叹，但总的来说仍难以弥补房地产行业因为政策调控导致的收缩。有迹象表明，由于定价改革，中国势不可挡的绿色能源建设正在放缓。而且，尽管从世界其他地区的角度来看，中国的清洁能源部署令人印象深刻，该国化石燃料的消耗仍然数量惊人——去年其煤炭消耗量是世界第二大消费国的四倍，尽管整体碳排放似乎已达峰。In the same way, the United States has doubled down on fossil fuels in recent years, making it the world’s largest producer of oil and natural gas. And while it has acquired some cultural features of a petrostate along the way, it is ultimately responsible for less than one-fifth of oil production and just one-quarter of gas production.同样，美国近年来也加大了对化石燃料的投入，成为世界上最大的石油和天然气生产国。虽然在此过程中沾染了石油国家的某些文化特征，但其石油产量实际占比不足全球五分之一，天然气仅占四分之一。Total investment doesn’t tell the full story of A.I., since Chinese models have been delivering impressive results despite much smaller investments and, thanks to American export controls, many fewer high-end chips. China may already be ahead of the United States in what is called embodied A.I., particularly in the form of humanoid robots. And the country undertook a frenzied data-center buildout of its own — though recent reporting suggests the boom was somewhat wasteful. It can be a little hard to assess the exact state of the intelligence arms race, given the lack of transparency, but conventional wisdom isn’t exactly that the United States is 11 times as far ahead, as the private investment figures imply. Survey the people who spend a lot of time thinking about all this and the basic view you get of the American position is something like “a little bit ahead.”总投资并不能说明人工智能的全部情况，之所以这么说，是因为中国的人工智能模型虽投资规模小得多，却持续交付令人印象深刻的成果，当然这也是由于美国的出口管制缘故，它能获得的高端芯片数量也要少得多。在所谓的具身人工智能方面，尤其人形机器人方面，中国可能已领先美国。中国还在疯狂建设数据中心——尽管近期有报道暗示这种繁荣背后存在浪费现象。由于缺乏透明度，要准确评估智能军备竞赛的实际状况略有困难，但普遍的看法并非如私人投资数据所暗示的“美国领先11倍”。如果去问那些长期研究这一领域的人，你会得到的基本结论大致是美国“略微领先”。The deeper irony for the United States is that one of these bets is tied up in the other, since at least as it has been pursued in this recent phase of large language models and hyperscaling, A.I. really, really needs abundant and cheap electricity. And although the fastest and cheapest way to provide that today is through solar power, the new administration has not just kneecapped the I.R.A. but also undertaken a war on renewables.对美国来说，更深层的讽刺在于，这两项赌注是相互交织的，因为至少在近期大语言模型和超大规模化的发展阶段，人工智能确实极度依赖充足而廉价的电力。尽管今天提供这种电力最快、最便宜的方式是太阳能，但新政府不仅削弱了《通胀削减法案》，还对可再生能源发起了围剿。This is one reason electricity prices, which have recently spiked, are likely to continue to rise — demand is almost universally expected to explode in the years to come, and we aren’t doing nearly enough to add capacity to supply it. In fact, the opposite: Though Trump loyalists talk often about the need for “energy dominance,” the supposedly future-forward administration is wrapping everything green in endless red tape and torpedoing existing clean-infrastructure projects for sport. “A.I. experts return from China stunned,” ran one hyperbolic Fortune headline. “The U.S. grid is so weak, the race may already be over.”这就是电价最近飙升的原因之一，而且未来几年电价很可能还会持续上涨——几乎普遍预计，电力需求将呈现爆炸式增长，而我们远未做好增加电力供应的准备。事实上，情况恰恰相反：尽管特朗普的忠实追随者经常谈论“能源主导”的必要性，但这个号称面向未来的政府却用无尽的繁文缛节束缚了所有绿色项目，并如儿戏般破坏现有的清洁基础设施项目。《财富》杂志一条夸张的标题写道：“AI专家中国归来震惊不已：美国电网如此脆弱，竞赛或已终结。”Just on Friday, for instance, the Trump administration ordered a stop to construction on an offshore wind farm, mostly built already, scheduled to come online as soon as next year, when it would provide a meaningful chunk of electricity to Rhode Island and Connecticut. And investment in fossil-fuel production isn’t exactly picking up the slack.例如就在上周五，特朗普政府下令停建一个海上风电场（该项目已基本建成，原定明年并网，为罗得岛州和康涅狄格州提供可观电力）。而对化石燃料生产的投资也并没有真正弥补这一缺口。This is all the more remarkable because the most conspicuous new faction in the MAGA coalition, the tech right, is not just obsessed with A.I. but also chose to join team Trump largely out of frustration with the regulatory and permitting obstacles imposed by the previous administration on its vision of the future. Now out of government, Elon Musk has taken to posting about China’s expanding electricity lead, and in fact it’s on this question, perhaps more than any other, that you can see the beginnings of a cleavage between the tech right and Trump’s base.这一切之所以更显荒谬，是因为MAGA联盟中最引人注目的新派系——科技右翼——不仅痴迷AI，而且他们之所以选择加入特朗普阵营，很大程度上是源于对前政府在其未来愿景上设置的监管与许可障碍的不满。退出政府的埃隆·马斯克开始频繁发帖，谈论中国不断扩大的电力领先优势，事实上，通过这个问题也许最能看到科技右翼与特朗普票仓之间初现的裂痕。But factional politics is one thing, geopolitics another. And if one path for artificial intelligence is that it reveals itself as merely a normal technology, one related possibility could be that it brings about more normality in the realm of imperial competition, too. In this future, perhaps, artificial general intelligence isn’t the contemporary equivalent of the atomic bomb, toward which any self-respecting superpower must sprint, and each party can therefore start to relax some of the hardball tactics recently imposed in the spirit of winner-take-all competition. Pollyannas may even find reasons to believe they are seeing some signs of such a shift, with Trump making an otherwise hard-to-explain concession allowing Nvidia to export its H20 chips to China, amid ongoing trade negotiations between the two countries, and longtime A.I. hawk Eric Schmidt now urging the country to focus away from national-security-focused superintelligence and toward concrete domestic applications for current A.I. A few years ago, China hawks might have bristled at the slightest hint of rapprochement. But perhaps it signals a return to the status quo ante, however strange that was — with great-power rivalry unfolding even as the deep integration of the two economic systems mitigated the friction. Here’s hoping.但派系政治是一回事，地缘政治又是另一回事。如果人工智能的一个发展路径是最终揭晓它只是一种常规技术，那么一个相关的可能性是，它也将为帝国竞争领域带来更多常态。在这种未来中，或许通用人工智能不再是当代等同于原子弹的存在，每个自重的超级大国都必须为此冲刺，因此各方可以开始放松最近为追求赢家通吃式竞争而施加的一些强硬策略。乐观主义者甚至可能找到理由相信，他们正看到这种转变的一些迹象，比如在两国正在进行的贸易谈判中，特朗普做出了一项难以解释的让步，允许英伟达向中国出口其H20芯片；长期以来的人工智能鹰派人物埃里克·施密特现在也敦促美国将重心从国家安全导向的超级智能转移到当前人工智能的具体国内应用上。几年前，对华鹰派人士可能会对哪怕最轻微的缓和迹象都勃然大怒。但这或许预示着一种回归常态，无论那曾经多么奇怪——大国竞争在展开，而两国经济体系的深度融合又缓解了摩擦。希望如此。翻译：杜然点击查看本文英文版。