Why AI Isn't Actually Boosting Productivity

以下是这段内容的中文翻译： 在《特朗普经济学》(Trumponomics) 的一期节目中，主持人斯蒂芬妮·弗兰德斯采访了《进步何以终结》(How Progress Ends) 一书的作者卡尔-贝内迪克特·弗雷，深入探讨了全球经济中两种看似矛盾的论调：一种是非凡的创新（尤其是人工智能）正在涌现，另一种则是停滞不前和竞争日益减弱。 弗兰德斯介绍了这种二元对立：一方面，日常的人工智能进步预示着一个创新丰富的时代，美国和中国都在争夺主导地位。这里的挑战在于如何在不造成剧烈社会动荡的前提下，消化如此多的创新。另一方面，尤其是自全球金融危机以来，持续存在的论调是经济停滞、垄断势力上升、政府干预增多以及贸易关税——这些主题都与特朗普时代的美国经济息息相关，并在欧洲普遍存在。本次讨论旨在调和这些观点，并判断世界是正走向前所未有的增长，还是潜在的衰退。 卡尔-贝内迪克特·弗雷在他的书中阐述的核心论点是，持续的进步需要制度适应技术变革。他解释说，国家可以通过简单地采纳并推广现有技术来实现显著增长，并以苏联在大规模生产方面的成功为例。然而，当技术成熟并需要新的创新时，一个能够承担风险和进行探索的去中心化系统就变得至关重要。弗雷断言，进步并非必然发生；它是一种“非自然”的状态，需要特定的条件，历史上的长期停滞和现代的生产力放缓都证明了这一点。 弗雷指出了一个悖论：尽管“发明产出”（专利、科学出版物）激增，但“转型性产出”和生产力却在下降。他认为，计算机革命比当前的人工智能更具变革性，主要是因为互联网实现了“停机时间”的自动化，提供了即时信息访问。人工智能虽然自动化了知识工作，但仍需要人工验证，这限制了其相对于前代技术所能带来的整体生产力提升。 关于中美创新竞赛，弗雷对中国是一个纯粹的集中式模仿者的说法提出了质疑。他指出，中国存在去中心化的省级竞争，这可以促进亲竞争的产业政策。然而，中国缺乏法治意味着政治关系对企业而言至关重要。颇为有趣的是，弗雷指出美国正日益趋同于中国的“政治资本主义”模式，即企业能否与政府保持一致（例如OpenAI寻求政府投资）变得至关重要。 弗雷将“既得利益者”视为历史进步的障碍。现有企业常常通过“扼杀式收购”有前途的初创公司，或者利用与专利局的密切关系制造进入壁垒来抵制颠覆。这解释了尽管技术使得创办公司成本更低，商业活力却在下降的原因。在中国，中共将重心从经济目标转向国家安全和自给自足，导致对国有企业的依赖增加，而国有企业通常创新能力和生产力都较低。 针对人工智能是否代表着一个“这次不一样”的局面，即纯粹的规模（计算力、数据、能源）驱动进步而无需太多新颖创新，弗雷表示不同意。他认为世界是动态的，而非静态的，“蛮力”不足以解决问题。他引用了AlphaGo的例子，业余人类选手通过引入训练中未遇到的新棋局暴露出其弱点，这说明了人工智能目前缺乏人类般的“数据效率”。弗雷认为，人工智能仍在等待其“独立冷凝器时刻”——一项基础性创新，就像使蒸汽机真正实现高效能的那项发明一样——以达到变革性的潜力，而非仅仅是渐进式的规模扩张。 对于那些尚未处于技术前沿的国家，弗雷建议专注于技术采纳。历史上，战后的马歇尔援助促进了技术转移。然而，出于国家安全考虑，例如美国近期对人工智能模型出口的限制，仅仅依赖外部技术存在风险。弗雷建议，这些国家可能会转向中国或欧洲技术，或努力构建国内的开源权重人工智能生态系统，以确保技术主权。他还警告说，学术界和研究领域目前的激励机制往往促使个人“钻更多的孔”（做更多的项目），而非“挖得更深”以寻求真正的突破，这阻碍了变革性进展。

In an episode of "Trumponomics," host Stephanie Flanders interviews Carl-Benedict Frey, author of "How Progress Ends," to delve into two seemingly conflicting narratives about the global economy: one of extraordinary innovation (particularly AI) and another of stagnation and declining competition. Flanders introduces the dichotomy: on one hand, daily AI advancements suggest an era of abundant innovation, with the US and China vying for dominance. The challenge here is absorbing too much innovation without disruption. On the other hand, the persistent narrative, especially since the global financial crisis, is one of stagnation, rising monopoly power, increased government intervention, and trade tariffs – themes relevant to Trump's American economy and prevalent in Europe. The discussion aims to reconcile these views and determine if the world is headed for unprecedented growth or potential decline. Carl-Benedict Frey's core thesis, as outlined in his book, is that sustained progress requires institutions to adapt to technological shifts. He explains that countries can grow significantly by merely adopting and scaling existing technologies, citing the Soviet Union's success with mass production. However, when technologies mature and new innovation is needed, a more decentralized system capable of risk-taking and exploration becomes crucial. Progress, Frey asserts, is not inevitable; it's an "unnatural" state requiring specific conditions, as evidenced by long periods of stagnation in history and modern productivity slowdowns. Frey highlights a paradox: despite a surge in "inventive output" (patents, scientific publications), there's a decline in "transformational output" and productivity. He argues that the computer revolution was more transformative than current AI, primarily because the internet automated "downtime," offering instantaneous access to information. AI, while automating knowledge work, still requires human verification, limiting its overall productivity boost compared to its predecessors. Regarding the US-China innovation race, Frey challenges the notion of China as a purely centralized copycat. He notes China's decentralized provincial competition, which can foster pro-competitive industrial policies. However, China's lack of rule of law means political connections are paramount for businesses. Intriguingly, Frey suggests the US is increasingly resembling China's "political capitalism," where alignment with the administration (e.g., OpenAI seeking government investment) is crucial. Frey points to "vested interests" as a historical barrier to progress. Incumbent firms often resist disruption through "killer acquisitions" of promising startups or by leveraging close ties with patent offices to create barriers to entry. This explains the decline in business dynamism despite technologies making it cheaper to start companies. In China, the CCP's shift from economic targets to national security and self-sufficiency has led to greater reliance on state-owned enterprises, which are generally less innovative and productive. Addressing whether AI represents a "this time is different" scenario, where sheer scale (compute, data, energy) drives progress without needing much novel innovation, Frey disagrees. He contends that the world is dynamic, not static, and "brute force" isn't enough. He references the AlphaGo example, where human amateurs exposed weaknesses by introducing novel positions not encountered during training, illustrating AI's current lack of human-like "data efficiency." Frey argues AI still awaits its "separate condenser moment"—a fundamental innovation, akin to the invention that made steam engines truly energy-efficient—to achieve transformative potential, rather than just incremental scaling. For countries not at the technological frontier, Frey advises focusing on adoption. Historically, post-war Marshall Aid facilitated technology transfer. However, national security concerns, exemplified by recent US restrictions on AI model exports, make relying solely on external tech risky. Frey suggests that such countries might pivot towards Chinese or European technology, or strive to build domestic open-weights AI ecosystems to ensure technological sovereignty. He also warns that current incentives in academia and research often lead individuals to "drill more holes" (do more projects) rather than "dig deeper" for true breakthroughs, hindering transformative progress.