Journal of Economic Perspectives · 2026 · Charles I. Jones
中文摘要
人工智能（AI）很可能成为现代最具变革性的技术。如果机器——包括执行认知任务的AI，以及执行体力任务的AI与先进机器人——能够完成人类所能完成的每一项任务，将会怎样？本文提出三个主要观点。第一，尽管美国的增长率在过去150年间一直稳定在每年约2%，但智能自动化仍有可能推动经济增长率加速。第二，“薄弱环节”的存在很可能减缓这一加速过程。如今，我们每个人使用的台式计算机所拥有的晶体管数量都是20世纪70年代人们所用计算机的1亿倍，但我们的生产率并未提高1亿倍。计算机可以以极快的速度求矩阵的逆，但仍须由人类决定对哪个矩阵求逆、检验什么假设，等等。要使经济增长加速，就必须将绝大多数薄弱环节自动化，而这会推迟巨额收益的实现。最后，尽管薄弱环节会延缓收益的实现，却可能反而加速风险的到来。当一条链条的强度取决于其最薄弱的环节时，破坏其中一个环节便可能造成极高的代价。一个在软件工程方面能力超越人类的强大AI，可能被恶意行为者滥用，通过入侵金融系统或病毒学实验室造成重大危害。
Abstract
Artificial intelligence (AI) will likely be the most transformative technology of the modern era. What if machines—AI for cognitive tasks and AI plus advanced robots for physical tasks—can perform every task a human can? This essay makes three main points. First, even though US growth rates have been stable at roughly 2 percent per year for 150 years, it is distinctly possible that automating intelligence leads economic growth rates to accelerate. Second, this acceleration is likely to be slowed by the presence of “weak links.” While we each have access to 100 million times more transistors on our desktop computer than people in the 1970s, we are not 100 million times more productive. Computers can invert matrices at lightning speed, but we humans must still decide what matrix to invert, what hypothesis to test, and so on. Accelerating economic growth requires the vast majority of the weak links to be automated away, which delays the large gains. Finally, even though weak links slow the benefits, they may actually speed up the risks. When a chain is only as strong as its weakest link, damaging one link in the chain can be very costly. A powerful AI that is superhuman at software engineering could be misused by a bad actor to do substantial harm by hacking the financial system or a virology lab.