AI and the K-Shaped World人工智能与K型世界人工智能與K型世界

I. The most far-reaching disruption since the Industrial Revolution一、自工业革命以来最深远的破坏性创新一、自工業革命以來最深遠的破壞性創新

To judge where artificial intelligence sits in history, set it beside the technological revolutions that came before.

要判断人工智能的历史位置，可以把它放回技术革命的谱系里比较。

要判斷人工智能的歷史位置，可以把它放回技術革命的譜系裡比較。

The Industrial Revolution changed energy and power. Steam and electricity remade energy supply, stimulated industry after industry, upended transport, and made trade and communication across great distances cheap. It was disruptive, certainly: steam displaced the horse-drawn carriage, machines displaced part of manual labor. But the disruption had a boundary. Machines replaced muscle, and people moved into the cognitive and service work that machines could not reach. For more than a century afterward, that is exactly how humanity kept inventing new occupations.

工业革命改变的是能源与动力。蒸汽机和电力重塑了能源供给，刺激了各行各业，颠覆了运输，让远距离的贸易和通信变得便捷。它当然也是破坏性创新，蒸汽与电力取代了马车，机器替代了部分手工劳动。但它的破坏有边界：机器替代的主要是体力，人还可以转向机器够不着的脑力与服务性工作。工业革命之后的一百多年，人类正是沿这条路不断造出新的职业。

工業革命改變的是能源與動力。蒸汽機和電力重塑了能源供給，刺激了各行各業，顛覆了運輸，讓遠距離的貿易和通信變得便捷。它當然也是破壞性創新，蒸汽與電力取代了馬車，機器替代了部分手工勞動。但它的破壞有邊界：機器替代的主要是體力，人還可以轉向機器夠不著的腦力與服務性工作。工業革命之後的一百多年，人類正是沿這條路不斷造出新的職業。

The internet revolution changed the flow of information. It battered large parts of the physical economy, released enormous energy, and its effects are still with us. But the internet is, at bottom, a network of pipes. It drove the cost of moving information toward zero without producing content or exercising judgment. Judgment and creation still belonged to people.

互联网革命改变的是信息的流动。它冲击了大量实体经济，也创造了空前的活力，影响延续至今。但互联网本质上是一张管道网络，它把信息传输的成本压到近乎为零，却不直接生产内容，也不做判断。判断和创造，仍然属于人。

互聯網革命改變的是資訊的流動。它衝擊了大量實體經濟，也創造了空前的活力，影響延續至今。但互聯網本質上是一張管道網絡，它把資訊傳輸的成本壓到近乎為零，卻不直接生產內容，也不做判斷。判斷和創造，仍然屬於人。

Artificial intelligence replaces cognition itself: understanding, reasoning, writing, coding, diagnosis, research. That is why it seems to respect no industry boundary. In basic science, DeepMind's AlphaFold cracked the protein-structure problem that had defied biologists for fifty years, and its developers received the 2024 Nobel Prize in Chemistry. In software, leading firms now say publicly that a substantial share of their new code is written by AI; programming, once considered the safest career of the future, was among the first to feel the blade. In the labor market, Stanford researchers found in 2025 that entry-level employment for young workers had fallen visibly in the occupations most exposed to AI. The bottom rung of the career ladder is being removed, quietly. And at the macro level, AI-related investment in data centers and equipment has become one of the most important engines of recent American growth; some economists estimate that without it, the figures would look far worse.

人工智能替代的是认知本身：理解、推理、写作、编程、诊断、研究。所以它几乎没有行业边界，似乎能革每一个行业的命。在基础科学，DeepMind的AlphaFold解决了困扰生物学界五十年的蛋白质结构预测难题，开发者拿到了2024年的诺贝尔化学奖。在软件业，头部公司已经公开说，新增代码里相当一部分由人工智能生成；编程这个曾被视为铁饭碗的职业，反倒最先挨了刀。在劳动力市场，斯坦福研究者2025年发现，受人工智能冲击最大的那些职业里，年轻人的入门岗位出现了明显下滑，职业阶梯最底下那一级正在被悄悄抽走。再看宏观数字：与人工智能相关的数据中心和设备投资，已经成了近两年美国经济增长最重要的引擎之一，有经济学家估算，剔除这部分投资，美国的增长会难看得多。

人工智能替代的是認知本身：理解、推理、寫作、編程、診斷、研究。所以它幾乎沒有行業邊界，似乎能革每一個行業的命。在基礎科學，DeepMind的AlphaFold解決了困擾生物學界五十年的蛋白質結構預測難題，開發者拿到了2024年的諾貝爾化學獎。在軟件業，頭部公司已經公開說，新增代碼裡相當一部分由人工智能生成；編程這個曾被視為鐵飯碗的職業，反倒最先挨了刀。在勞動力市場，斯坦福研究者2025年發現，受人工智能衝擊最大的那些職業裡，年輕人的入門崗位出現了明顯下滑，職業階梯最底下那一級正在被悄悄抽走。再看宏觀數字：與人工智能相關的數據中心和設備投資，已經成了近兩年美國經濟增長最重要的引擎之一，有經濟學家估算，剔除這部分投資，美國的增長會難看得多。

The steam engine replaced the carriage but not the driver's mind. The internet replaced the newspaper but not the journalist's judgment. Artificial intelligence is the first technology to close in on both at once. That is why we regard it as the most far-reaching disruptive innovation since the Industrial Revolution. And this is only the beginning. The technology is moving so fast that the distribution of gains between social classes is shifting faster than at any point in living memory.

蒸汽机取代了马车，没有取代车夫的头脑；互联网取代了报纸，没有取代记者的判断。人工智能第一次同时逼近了这两样东西。所以我们认为，它是自工业革命以来最深远的破坏性创新。而这仅仅是开始。技术进步太快了，快到社会阶层之间利益分配格局的改变速度，也是前所未有的。

蒸汽機取代了馬車，沒有取代車夫的頭腦；互聯網取代了報紙，沒有取代記者的判斷。人工智能第一次同時逼近了這兩樣東西。所以我們認為，它是自工業革命以來最深遠的破壞性創新。而這僅僅是開始。技術進步太快了，快到社會階層之間利益分配格局的改變速度，也是前所未有的。

One objection deserves a direct answer. Every technological revolution has produced predictions that work would disappear forever, and every time, as old occupations died, new ones grew. Why should this time differ? Over the long run, nobody knows. The difference is speed. The change is arriving so quickly that frictional unemployment is piling up at a pace labor markets have rarely seen. A textile worker once had a generation to retrain as a machinist. A junior programmer today may have a few years, perhaps a few months. Even if a long-run equilibrium eventually arrives, the short-run pain is real and severe. For the people living through it, the short run may mean ten or twenty years, long enough to rewrite the fortunes of a generation.

会有人反驳：历史上每一次技术革命都有人预言工作将永久消失，结果旧职业消亡的同时，新职业总会长出来，这一次凭什么不同？长期而言，答案确实没有人知道。不同的地方在速度。变革来得太快，劳动力市场的摩擦性失业正在以罕见的速度堆积。一个纺织工人曾经有一代人的时间去转型做机床工人；今天一个初级程序员，窗口可能只有几年，甚至几个月。就算长期的均衡终会到来，短期的阵痛也真实而剧烈。对身处其中的人来说，所谓短期，可能就是十年二十年，足够改写一代人的命运。

會有人反駁：歷史上每一次技術革命都有人預言工作將永久消失，結果舊職業消亡的同時，新職業總會長出來，這一次憑什麼不同？長期而言，答案確實沒有人知道。不同的地方在速度。變革來得太快，勞動力市場的摩擦性失業正在以罕見的速度堆積。一個紡織工人曾經有一代人的時間去轉型做機床工人；今天一個初級程序員，窗口可能只有幾年，甚至幾個月。就算長期的均衡終會到來，短期的陣痛也真實而劇烈。對身處其中的人來說，所謂短期，可能就是十年二十年，足夠改寫一代人的命運。

II. The K-shaped society: divergence within nations二、K型社会：一国之内的分化二、K型社會：一國之內的分化

The change shows up first within countries.

变化首先发生在一国之内。

變化首先發生在一國之內。

American economists describe the moment with a single letter: the K-shaped economy. One stroke of the K points up, the other points down. The wealth of a few is climbing the upper stroke at startling speed; the circumstances of the many slide down the lower one. Under the combined pressure of geopolitical conflict, tariff wars, and policy shocks, ordinary people face rising costs and narrowing opportunities, while those on the supply side of AI watch their fortunes compound.

美国经济学家用一个形象的说法描述当下：K型经济。字母K的两笔，一笔向上，一笔向下。少数人的财富沿上面那一笔以惊人速度增长；多数人的处境沿下面那一笔缓慢下滑。在地缘冲突、关税战和各种政策叠加的压力下，普通人的生活成本在涨，机会在收窄，而站在人工智能供给端的少数人，财富正以指数级速度膨胀。

美國經濟學家用一個形象的說法描述當下：K型經濟。字母K的兩筆，一筆向上，一筆向下。少數人的財富沿上面那一筆以驚人速度增長；多數人的處境沿下面那一筆緩慢下滑。在地緣衝突、關稅戰和各種政策疊加的壓力下，普通人的生活成本在漲，機會在收窄，而站在人工智能供給端的少數人，財富正以指數級速度膨脹。

Consider the numbers. Federal Reserve data show that in the third quarter of 2025, the richest 1 percent of American households held about 31.7 percent of national wealth, the highest share since records began in 1989. Their combined assets, roughly 55 trillion dollars, matched the wealth of the bottom 90 percent of Americans put together. The bottom half of the population held about 2.5 percent. Consumption is turning K-shaped as well: the top 10 percent of households by income now account for nearly half of all American consumer spending, also a record.

看数据。美联储统计，2025年第三季度，美国最富的1%家庭持有全国约31.7%的财富，是1989年有记录以来的最高值；这1%的总资产约55万亿美元，大致等于底层90%美国人的财富总和；占人口一半的底层50%家庭，只分到约2.5%。消费也在K型化：收入前10%的家庭贡献了全美近一半的消费支出，同样是有统计以来的最高。

看數據。美聯儲統計，2025年第三季度，美國最富的1%家庭持有全國約31.7%的財富，是1989年有記錄以來的最高值；這1%的總資產約55萬億美元，大致等於底層90%美國人的財富總和；佔人口一半的底層50%家庭，只分到約2.5%。消費也在K型化：收入前10%的家庭貢獻了全美近一半的消費支出，同樣是有統計以來的最高。

The economist Gabriel Zucman supplies the historical coordinates. The richest 0.01 percent of American families today hold about 10 percent of the country's wealth; at the close of the Gilded Age in 1913, the figure was 9 percent. Measured by concentration, America has not merely returned to the Gilded Age. It has passed it. Hence the phrase now in circulation: a New Gilded Age.

经济学家Zucman的测算给出了历史坐标：今天美国最富0.01%家庭占有约10%的财富，而在1913年镀金时代末期，这个数字是9%。以财富集中度衡量，美国不只是回到了镀金时代，而是越过了它。"新镀金时代"由此得名。

經濟學家Zucman的測算給出了歷史坐標：今天美國最富0.01%家庭佔有約10%的財富，而在1913年鍍金時代末期，這個數字是9%。以財富集中度衡量，美國不只是回到了鍍金時代，而是越過了它。「新鍍金時代」由此得名。

Nor is this an American condition alone. The faster an economy deploys AI, the sooner it faces the same distributional test, and China, standing at the same frontier, is no exception. The real difference may lie less in whether the challenge arrives than in each country's capacity, and institutions, for meeting it. On that point we offer a judgment worth serious debate: China's political system gives the state a far stronger capacity for distribution and redistribution than America's. Different institutions will hand in different answers to this shock, and which one defuses the conflict more gently may prove one of the most consequential institutional contests of the coming decade. Our preliminary view is that on this particular question, China's system may hold the advantage. That judgment, too, awaits the verdict of history.

K型分化并非美国独有的课题。人工智能部署越快的经济体，越早要面对这道分配考题，同样站在前沿的中国也不例外。真正的区别，或许不在于挑战是否到来，而在于各国应对的能力与制度安排。正是在应对能力这一点上，我们提出一个值得认真讨论的判断：中国的政治制度赋予了国家远强于美国的收入分配与再分配能力。面对人工智能带来的分配冲击，不同的制度会交出不同的答卷，哪一种更能温和地化解矛盾，很可能是未来十年最重要的制度竞争之一。我们的初步倾向是，在这道特定的考题上，中国的制度或许更有优势。这同样是一个留待历史检验的判断。

K型分化並非美國獨有的課題。人工智能部署越快的經濟體，越早要面對這道分配考題，同樣站在前沿的中國也不例外。真正的區別，或許不在於挑戰是否到來，而在於各國應對的能力與制度安排。正是在應對能力這一點上，我們提出一個值得認真討論的判斷：中國的政治制度賦予了國家遠強於美國的收入分配與再分配能力。面對人工智能帶來的分配衝擊，不同的制度會交出不同的答卷，哪一種更能溫和地化解矛盾，很可能是未來十年最重要的制度競爭之一。我們的初步傾向是，在這道特定的考題上，中國的制度或許更有優勢。這同樣是一個留待歷史檢驗的判斷。

AI strikes at distribution twice, once on each side of the market.

人工智能对分配的冲击是双重的，供给端一刀，消费端一刀。

人工智能對分配的衝擊是雙重的，供給端一刀，消費端一刀。

On the supply side, the wealth it creates is intensely concentrated. Chips, computing power, models, cloud: every link in the chain is winner-take-all, and the equity sits with a small number of founders, employees, and investors. The dividend of this productivity revolution flows first to capital, not labor.

供给端，财富的创造高度集中。芯片、算力、模型、云，这条产业链的每一环都是赢家通吃，股权又集中在少数创始人、员工和投资者手里。生产力革命的红利，第一站流向资本，不是劳动。

供給端，財富的創造高度集中。芯片、算力、模型、雲，這條產業鏈的每一環都是贏家通吃，股權又集中在少數創始人、員工和投資者手裡。生產力革命的紅利，第一站流向資本，不是勞動。

On the demand side lies an inequality discussed less often but just as deep: the best intelligence costs money. Top-tier subscriptions to frontier AI services run one to two hundred dollars a month, and heavy users spending thousands a month are no longer unusual. The well-off can buy the strongest models without limit, multiplying their productivity and their reach; lower-income users make do with free or cheap versions. The trouble is that the gap in capability between the strongest models and the ordinary ones is itself a cliff.

消费端的不平等讨论得少一些，却同样深刻：最先进的智能是要花钱买的。眼下最前沿的人工智能服务，顶配订阅每月一两百美元，重度使用者一个月花掉几千上万美元并不稀奇。家境优渥的人可以不设上限地购买最强的模型，放大自己的工作效率和探索世界的能力；中低收入群体只能用免费或廉价的版本。麻烦在于，最强模型和普通模型之间的生产力差距，本身就是一道悬崖。

消費端的不平等討論得少一些，卻同樣深刻：最先進的智能是要花錢買的。眼下最前沿的人工智能服務，頂配訂閱每月一兩百美元，重度使用者一個月花掉幾千上萬美元並不稀奇。家境優渥的人可以不設上限地購買最強的模型，放大自己的工作效率和探索世界的能力；中低收入群體只能用免費或廉價的版本。麻煩在於，最強模型和普通模型之間的生產力差距，本身就是一道懸崖。

The result is a harsh loop. AI's productivity gains empower the upper-middle class first; those below cannot obtain comparable tools; the distribution of income deteriorates further. From wealth creation on the supply side to access on the demand side, AI widens inequality at both ends. Nothing in this essay worries us more.

于是形成一个残酷的循环：人工智能带来的生产力增长优先赋能中高收入阶层，中低收入阶层拿不到同等的工具，收入分配进一步恶化。从供给端的财富创造，到消费端的能力获取，人工智能在两头同时加剧不平等。这是我们最忧虑的地方。

於是形成一個殘酷的循環：人工智能帶來的生產力增長優先賦能中高收入階層，中低收入階層拿不到同等的工具，收入分配進一步惡化。從供給端的財富創造，到消費端的能力獲取，人工智能在兩頭同時加劇不平等。這是我們最憂慮的地方。

III. The mirror of the Gilded Age三、镀金时代的镜子三、鍍金時代的鏡子

Has history seen such a moment? Episodes in which a productivity revolution churned distribution this violently are rare. The closest mirror is America's Gilded Age at the end of the nineteenth century.

历史上有过类似的时刻吗？生产力革命把分配格局搅得如此剧烈的例子，其实非常少。最接近的镜像，是十九世纪末美国的镀金时代。

歷史上有過類似的時刻嗎？生產力革命把分配格局攪得如此劇烈的例子，其實非常少。最接近的鏡像，是十九世紀末美國的鍍金時代。

Why "gilded" rather than "golden"? Because it was not solid gold, only a bright surface over deep cracks. Railroad, steel, and oil barons amassed unprecedented fortunes; Rockefeller, Carnegie, and Morgan bestrode the age. The other side of the ledger held sweatshops, child labor, strikes met with force, and a politics captured by monopoly capital.

为什么叫镀金，不叫黄金？因为它不是真金，只是表面浮着一层金光，内里布满裂痕。铁路、钢铁、石油大王积累起空前的财富，洛克菲勒、卡内基、摩根的名字如日中天；另一面是血汗工厂、童工、频繁的罢工与镇压，以及被垄断资本俘获的政治。

為什麼叫鍍金，不叫黃金？因為它不是真金，只是表面浮著一層金光，內裡布滿裂痕。鐵路、鋼鐵、石油大王積累起空前的財富，洛克菲勒、卡內基、摩根的名字如日中天；另一面是血汗工廠、童工、頻繁的罷工與鎮壓，以及被壟斷資本俘獲的政治。

The exit from that era came through accident and blood. In 1901, President William McKinley was shot in Buffalo by the anarchist Leon Czolgosz, a former wire-mill worker who had lost his job in a depression and claimed to be avenging the oppressed. Vice President Theodore Roosevelt succeeded him and governed with rare force: breaking up the trusts, mediating between labor and capital, pushing through food and drug regulation. Over the following two decades America entered the Progressive Era. Antitrust law was finally enforced, the federal income tax arrived in 1913, labor protections were written into statute, and the country climbed out of the Gilded Age.

这个时代的出口，充满偶然与血色。1901年，总统麦金莱在布法罗被无政府主义者乔尔戈什刺杀。刺客当过钢丝厂工人，在萧条中失业，自称要为被压迫者复仇。副总统西奥多·罗斯福继任，随后以罕见的强硬手腕开启改革：拆分托拉斯，调停劳资冲突，推动食品药品监管。此后二十年，美国进入进步时代，反托拉斯法真正得到执行，联邦所得税于1913年设立，劳工保护逐步立法，美国由此走出镀金时代的深渊。

這個時代的出口，充滿偶然與血色。1901年，總統麥金萊在布法羅被無政府主義者喬爾戈什刺殺。刺客當過鋼絲廠工人，在蕭條中失業，自稱要為被壓迫者復仇。副總統西奧多·羅斯福繼任，隨後以罕見的強硬手腕開啟改革：拆分托拉斯，調停勞資衝突，推動食品藥品監管。此後二十年，美國進入進步時代，反托拉斯法真正得到執行，聯邦所得稅於1913年設立，勞工保護逐步立法，美國由此走出鍍金時代的深淵。

To credit the turn to one assassination and one strongman is, of course, too simple. What actually resolved the conflict was a social movement that ran for decades, driven by journalists, workers, and reforming politicians. Yet the period leaves a heavy question. Had those accidents not occurred, had the window for reform not opened in time, how would America have escaped the social rupture that the productivity revolution produced? One assassination may have been the smallest price in blood that era could have paid.

把历史转折归功于一次刺杀和一位强人，当然过于简化。真正化解矛盾的，是一场持续几十年、由记者、工人和改革派政治家共同推动的社会运动。但这段历史留下一个沉重的问题：如果没有那些偶然，改革的窗口没有被及时打开，美国要怎样走出生产力革命带来的社会撕裂？一次刺杀，或许已经是那个时代付得起的最小的流血代价。

把歷史轉折歸功於一次刺殺和一位強人，當然過於簡化。真正化解矛盾的，是一場持續幾十年、由記者、工人和改革派政治家共同推動的社會運動。但這段歷史留下一個沉重的問題：如果沒有那些偶然，改革的窗口沒有被及時打開，美國要怎樣走出生產力革命帶來的社會撕裂？一次刺殺，或許已經是那個時代付得起的最小的流血代價。

History permits no counterfactuals. The live question is whether today's America, and every society being rapidly remade by AI, can still find a comparatively gentle exit from the same dilemma. Candidly, we find it hard to be optimistic. Productivity is being reorganized far faster than it was a century ago, while the political systems of the United States and other Western countries are more polarized, and less capable of consensus, than they were then.

历史没有如果。真正的问题是，面对新镀金时代这道同样的难题，今天的美国，以及所有被人工智能快速重塑的社会，还能不能找到一个相对温和的出口？坦率说，难言乐观。这一次生产力重组的速度比一百多年前快得多，而美国等西方国家的政治系统，却比当年更极化，更难凝聚共识。

歷史沒有如果。真正的問題是，面對新鍍金時代這道同樣的難題，今天的美國，以及所有被人工智能快速重塑的社會，還能不能找到一個相對溫和的出口？坦率說，難言樂觀。這一次生產力重組的速度比一百多年前快得多，而美國等西方國家的政治系統，卻比當年更極化，更難凝聚共識。

IV. The ladder withdrawn: divergence between nations四、被抽走的梯子：国与国之间四、被抽走的梯子：國與國之間

Pull the lens back from one country to the world, and AI may be shaking the most important law in development economics.

把视野从一国之内拉远到国与国之间，人工智能动摇的可能是发展经济学最重要的一条定律。

把視野從一國之內拉遠到國與國之間，人工智能動搖的可能是發展經濟學最重要的一條定律。

The classic theory is the flying-geese paradigm, proposed by the Japanese economist Kaname Akamatsu. As a wealthy country develops, its labor costs rise, and firms gain an incentive to move labor-intensive industry to cheaper developing countries. The transfer raises the efficiency of the world economy and delivers opportunity to poorer nations. Textiles, garments, electronics assembly: these were the first rung by which nearly every late-developing country accumulated capital and stepped onto the ladder of industrialization. After Japan came the four Asian tigers; after the tigers, China. The formation has flown for more than half a century.

传统经济学里有一个著名的理论，叫雁行理论，由日本经济学家赤松要提出：富裕国家发展水平上升，劳动力成本随之上涨，企业就有动力把劳动密集型产业转移到成本更低的发展中国家。这种梯度转移提高了全球经济的整体效率，也给欠发达国家送去了机会。纺织、服装、电子组装，几乎是所有后发国家完成早期资本积累、迈上工业化台阶的第一级阶梯。日本之后是亚洲四小龙，四小龙之后是中国，这条雁阵飞了半个多世纪。

傳統經濟學裡有一個著名的理論，叫雁行理論，由日本經濟學家赤松要提出：富裕國家發展水平上升，勞動力成本隨之上漲，企業就有動力把勞動密集型產業轉移到成本更低的發展中國家。這種梯度轉移提高了全球經濟的整體效率，也給欠發達國家送去了機會。紡織、服裝、電子組裝，幾乎是所有後發國家完成早期資本積累、邁上工業化台階的第一級階梯。日本之後是亞洲四小龍，四小龍之後是中國，這條雁陣飛了半個多世紀。

Over the past decade the cycle seemed intact. As Chinese labor costs rose, manufacturing shifted toward Southeast and South Asia and began eyeing Africa and Latin America. This was a healthy state of affairs, a virtuous circle in which the early developers create openings for the late ones, the late ones climb, and, once mature, pass the ladder on.

过去十年，循环看上去仍在运转。中国劳动力成本上升，制造环节流向东南亚、南亚，并开始眺望非洲和拉美。这本是健康的状态，一种先富带动后富的良性循环：先发国家为后发国家创造机遇，后发国家沿梯子攀爬，成熟之后再把梯子递给更晚的来者。

過去十年，循環看上去仍在運轉。中國勞動力成本上升，製造環節流向東南亞、南亞，並開始眺望非洲和拉美。這本是健康的狀態，一種先富帶動後富的良性循環：先發國家為後發國家創造機遇，後發國家沿梯子攀爬，成熟之後再把梯子遞給更晚的來者。

AI may rewrite that logic. A firm facing rising labor costs now has another option: instead of moving the factory to a cheaper country, change the method of production and let AI-driven machines do the work. Relocation is being intercepted by automation.

人工智能可能改写这个逻辑。企业面对劳动力成本上升时，如今多了一个选项：不必再把工厂搬去更便宜的国家，直接改变生产方式，用人工智能加持的机器替代人工就是了。产业转移，被自动化截了胡。

人工智能可能改寫這個邏輯。企業面對勞動力成本上升時，如今多了一個選項：不必再把工廠搬去更便宜的國家，直接改變生產方式，用人工智能加持的機器替代人工就是了。產業轉移，被自動化截了胡。

If the shift becomes the norm, the geographic migration of industry will slow sharply or stop. For countries that have not yet industrialized, the prospect is bleak. Their greatest and often only comparative advantage, abundant cheap labor, is flattened by technology. The "premature deindustrialization" that the economist Dani Rodrik has warned of may reach its extreme in the AI era. The ladder is being withdrawn before they can reach for it.

如果这个转向成为主流，产业在全球的地理流动会消失，或者大幅放缓。对尚未完成工业化的国家，这是令人绝望的前景：它们最大、往往也是唯一的比较优势，廉价而充沛的劳动力，被技术一笔抹平。发展经济学家罗德里克警告过的"过早去工业化"，可能在人工智能时代走向极端。梯子还没等他们伸手，就被整个抽走了。

如果這個轉向成為主流，產業在全球的地理流動會消失，或者大幅放緩。對尚未完成工業化的國家，這是令人絕望的前景：它們最大、往往也是唯一的比較優勢，廉價而充沛的勞動力，被技術一筆抹平。發展經濟學家羅德里克警告過的「過早去工業化」，可能在人工智能時代走向極端。梯子還沒等他們伸手，就被整個抽走了。

A country that has nothing but its people, in a world where AI has discounted the relative value of labor, may miss its development window permanently and settle into long-term poverty, with hunger, unrest, crime, and even war following behind. This is no distant hypothetical. A young, populous continent shut out of the global division of labor is itself one of the largest geopolitical risks of the twenty-first century.

一个国家除了人力资源一无所有，而人工智能又给人力资源的相对价值打了一个大折扣，它就可能永远错过发展的窗口，陷入长期贫困，随之而来的是饥饿、动荡、犯罪，甚至战争。这不是遥远的假想。一个被排除在全球分工之外的年轻人口大陆，本身就是二十一世纪最大的地缘风险之一。

一個國家除了人力資源一無所有，而人工智能又給人力資源的相對價值打了一個大折扣，它就可能永遠錯過發展的窗口，陷入長期貧困，隨之而來的是飢餓、動盪、犯罪，甚至戰爭。這不是遙遠的假想。一個被排除在全球分工之外的年輕人口大陸，本身就是二十一世紀最大的地緣風險之一。

V. A dangerous era五、一个危险的时代五、一個危險的時代

Ray Dalio, the founder of Bridgewater, returns again and again to one observation in his studies of the rise and fall of great powers: historically, when technological change pushes wealth gaps to an extreme while a society loses internal consensus, the risk of disorder peaks. Measured by his framework, a moment that combines a technological revolution, great-power rivalry, and a debt cycle looks very much like the entrance to a dangerous era.

桥水基金创始人达利欧在他关于大国兴衰周期的研究里反复讲一件事：历史上，当技术变革把财富差距推到极点、社会内部又失去共识的时候，往往就是秩序动荡的高危阶段。按他的框架衡量，技术革命、大国竞争和债务周期叠加的当下，世界很可能正在进入一个非常危险的时代。

橋水基金創始人達利歐在他關於大國興衰週期的研究裡反覆講一件事：歷史上，當技術變革把財富差距推到極點、社會內部又失去共識的時候，往往就是秩序動盪的高危階段。按他的框架衡量，技術革命、大國競爭和債務週期疊加的當下，世界很可能正在進入一個非常危險的時代。

More than two thousand years ago, Confucius said in the Analects that a state should worry less about scarcity than about unequal shares. What people find unbearable is rarely want itself; it is severe imbalance in how things are divided. This is human nature, in every civilization and every century. Left uncorrected, such imbalance corrodes the stability and cohesion of any society.

两千多年前，孔子在《论语》里说，不患寡而患不均。人真正难以忍受的常常不是匮乏，而是分配的严重失衡。这是人性，古今中外概莫能外。失衡长期得不到矫正，任何社会的稳定与凝聚力都会被侵蚀。

兩千多年前，孔子在《論語》裡說，不患寡而患不均。人真正難以忍受的常常不是匱乏，而是分配的嚴重失衡。這是人性，古今中外概莫能外。失衡長期得不到矯正，任何社會的穩定與凝聚力都會被侵蝕。

Here, then, is the essay's central hypothesis in full:

我们把本文的核心假说完整写在这里：

我們把本文的核心假說完整寫在這裡：

Artificial intelligence will overturn several basic laws of conventional economics. Within nations, it sharpens K-shaped divergence from both ends, the creation of wealth and the access to productivity. Between nations, it may end the labor-cost-driven transfer of industry, withdrawing the development ladder from late-coming countries.

人工智能将颠覆传统经济学的若干基本定律。在一国之内，它从财富创造端和生产力获取端同时加剧K型分化；在国与国之间，它可能终结以劳动力成本为驱动的产业梯度转移，抽走后发国家的发展阶梯。

人工智能將顛覆傳統經濟學的若干基本定律。在一國之內，它從財富創造端和生產力獲取端同時加劇K型分化；在國與國之間，它可能終結以勞動力成本為驅動的產業梯度轉移，抽走後發國家的發展階梯。

It is a hypothesis. We sincerely hope it fails. But on the evidence before us, we must admit the probability of its being right is very high, even though systematic empirical proof has yet to accumulate.

这是一个假说。我们由衷希望它不要被验证，但凭目前所见的迹象，我们不得不承认它为真的概率极高，尽管系统性的实证证据还有待学界积累。

這是一個假說。我們由衷希望它不要被驗證，但憑目前所見的跡象，我們不得不承認它為真的概率極高，儘管系統性的實證證據還有待學界積累。

The questions that follow, we cannot yet answer. How should AI be governed so that productivity gains reach everyone? How can the creation of wealth be made fairer, and the transition kept relatively gentle? How should global governance respond to the withdrawal of the ladder?

随之而来的问题，我们目前没有答案：如何规范人工智能的发展，让生产力的提升惠及所有人？如何让财富的创造更公平，让转型以相对温和的方式完成？全球治理如何应对后发国家被抽梯的危机？

隨之而來的問題，我們目前沒有答案：如何規範人工智能的發展，讓生產力的提升惠及所有人？如何讓財富的創造更公平，讓轉型以相對溫和的方式完成？全球治理如何應對後發國家被抽梯的危機？

Whether the hypothesis holds, the next few years will tell; foreign investment flows and manufacturing employment in Vietnam, India, and their peers are the most direct windows to watch. The real bind is that decisions of this magnitude cannot wait for the problem to become fully visible. Our present unease comes precisely from this. By the time the problem is undeniable, it may be too late to act; until then, we do not know what to do. To see the crisis approaching and find no steering wheel: that is the deepest anxiety of the age.

假说本身能不能成立，未来几年就会有结果，越南、印度等国的外资流入和制造业就业数据是最直接的观察窗口。但真正的困境在于，重大的决策不能等问题充分显现之后才做。我们此刻的彷徨正来自这里：等问题真正凸显，很可能已经来不及行动；而在此之前，我们又不知道该做什么。看得见危机逼近，却找不到方向盘，这才是这个时代最深的焦虑。

假說本身能不能成立，未來幾年就會有結果，越南、印度等國的外資流入和製造業就業數據是最直接的觀察窗口。但真正的困境在於，重大的決策不能等問題充分顯現之後才做。我們此刻的彷徨正來自這裡：等問題真正凸顯，很可能已經來不及行動；而在此之前，我們又不知道該做什麼。看得見危機逼近，卻找不到方向盤，這才是這個時代最深的焦慮。

Conclusion结语結語

This essay does not offer a program. It is an attempt to order what can be seen at the frontier, and to ask. If it prompts readers to think, to answer, or to object, it has done its work.

这篇文章不打算给出方案。它是对时代前沿的一次梳理，也是一次发问。如果它能引出读者的思考、新的观点乃至反驳，目的就达到了。

這篇文章不打算給出方案。它是對時代前沿的一次梳理，也是一次發問。如果它能引出讀者的思考、新的觀點乃至反駁，目的就達到了。

A sinan, the ancient Chinese compass, does not walk for the traveler. It only shows where the fog lies. This is the first opinion essay of the Sinan Institute for Society, and our first invitation to readers: where a world remade by artificial intelligence is heading deserves our common attention, and our common answer.

司南这件器物，不替人走路，只指出迷雾所在。这是司南社会研究院的第一篇观点文章，也是我们向读者发出的第一份邀请：这个正在被人工智能重塑的世界会走向哪里，值得我们一起凝视，一起作答。

司南這件器物，不替人走路，只指出迷霧所在。這是司南社會研究院的第一篇觀點文章，也是我們向讀者發出的第一份邀請：這個正在被人工智能重塑的世界會走向哪裡，值得我們一起凝視，一起作答。