人工智能与大分歧

The Council of Economic Advisers January 2026 1Introduction For centuries, most of the world's economies grew at a similarly slow rate. However, a “Great Divergence”occurred with the Industrial Revolution, causing industrializing nations to accelerate their growth relative to However, we are witnessing clear leaders in AI investment, performance, and adoption metrics acrossdifferent nations. The Trump administration is laying the groundwork for American AI dominance byaccelerating innovation, infrastructure development, and deregulation while establishing global dominancethrough technology exports. If the AI revolution is as transformative as the Industrial Revolution, should we We begin by reviewing analyses of the potential for AI-led economic growth (Section 2) and then discussingestimates of AI’s impact on both GDP and the labor force. Recognizing that these impacts are uncertain andthus need constant monitoring, in Section 3 we highlight metrics for tracking the breakneck pace ofinvestment, performance, and adoption of AI. We then discuss how different countries are faring on thesemetrics (Section 4). The incredible speed of change cannot be overstated; many of these metrics aredoubling every few months and increasing manyfold each year. This means that the AI of the future will likely 2The Future Outlook The last 25 years have seen a great convergence as the world’s richest nations grew slower than manydeveloping nations. However, the advent of generative artificial intelligence based around large languagemodels (LLMs) will initiate a new wave of profound economic transformation in the United States, promising Yet, this period of innovation is not without its complexities. In this report, we focus on the long-term analysisof structural trends, as of course not every AI-related investment will be profitable, and the short-run always 2.1Background on Artificial Intelligence The last few years have seen a rapid explosion in both AI capabilities and jargon, so we begin with a review ofseveral key terms in the AI space. Artificial intelligence can refer to a wide variety of different computer systems, from chess-playingcomputers like Deep Blue to generative AI like ChatGPT. For most of AI’s history, AI was only capable ofmaking decisions among a relatively small set of options. The recent surge in AI interest has coincided withthe rise of “generative” AI, so called because they are able to “generate” text, images, or video. “Largelanguage models” are generative AI that can create text.3They are “large” because of their trillions of One framework for understanding the intelligence of an AI looks at that intelligence on two dimensions: (1)its ability to perform different tasks: from writing essays, to identifying objects in pictures, to writingcomputer code, to solving math problems and (2) how the AI’s capabilities on that task compare to human-level intelligence. Today’s artificial intelligence systems have “specialized” (or “narrow”) intelligencebecause, although they may be superhuman at a particular task (no human can multiply as fast as a calculatorcan), AI is not able to perform all the tasks a human can. Humans are capable of performing a wide variety ofdifferent tasks. Thus, we say that humans have “general” intelligence while current AI (including both Artificial general intelligence (AGI) would be a hypothetical AI that can perform all the intellectual tasks thathumans can,7but the exact definition of AGI is hotly debated, and some definitions only require that AGIperform“many but not all”human tasks.Artificial superintelligence(ASI),sometimes just called“superintelligence,” is AI with intelligence that surpasses that of humans.8The boundary between AGI andsuperintelligence is similarly contentious, partly because these terms encompass different aspects of AI:"AGI" and "specialized AI" describe the generality of tasks an AI can perform, while "superintelligence" This brings us to an important caveat to this report’s analysis: limitations of economic analysis of artificialintelligence. As noted by Hanson (2001), artificial intelligence that could perform all human tasks would leadto absolutely explosive growth and to a very different world than that seen today. Thus, the implications ofAGI (both economic and otherwise) are an important topic deserving of further study, but are generally 2.2Impact of AI on GDP Economists often think of the productive power of an economy as coming from three factors: the quantity oflabor, the quantity of capital, and total factor productivity (TFP). TFP is a measure of an economy's efficiencyand technological progress. A rising TFP indicates that an economy is producing more goods and servicesfrom the same amount of labor and capital, or the same output with fewer inputs.14This improvement in The productivity gains from TFP are eventually translated into higher overall economic output, or GDP.However, the effect of a new technology occurs with a time