Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics

Article Source: in The Economics of Artificial Intelligence: An Agenda, Ajay K. Agrawal, Joshua Gans, and Avi Goldfarb, eds., University of Chicago Press, 2019
Publication Date:
Time to Read: 2 minute read
Written By:

 Chad Syverson

Chad Syverson

 Daniel Rock

Daniel Rock



Artificial intelligence (AI) is advancing rapidly, but productivity growth has been falling for a decade, and real income has stagnated. The most plausible explanation is that it will take considerable time for AI-related technologies to be deployed throughout the economy.


Policy Relevance:

Intangible AI-related gains are hard to measure using traditional metrics such as Gross Domestic Product. Economists should develop new methods of measurement.


Key Takeaways:
  • Advances in AI have the potential to transform the economy and yield benefits for all.
    • Error rates in machine learning systems such as speech recognition systems have fallen to about 5%, about the same as human error rates.
    • Global investment in companies focused on AI has grown from $259 million in 2012 to over $5 billion in 2016.
  • Productivity growth in the United States has declined from 2.8% to 1.3% over the past decade; the decline is widespread, affecting many developed nations and emerging economies.
  • Four explanations for the clash between technology and investment gains and productivity declines are offered:
    • Like fusion power or flying cars, AI has raised false hopes.
    • AI has led to gains, but these have not been measured; however, this is unlikely, because studies using different measures yield the same results.
    • AI gains are concentrated in a few firms, leading to inequality and welfare losses.
    • AI gains will be realized slowly because of lags in implementation and restructuring; this is the most plausible explanation.
  • Historically, past productivity growth is not a good predictor of future productivity growth.
  • AI-based technologies such as autonomous cars will increase the productivity of labor, as about 45% of all tasks can be automated; AI will also improve the productivity of materials and energy usage (for example, AI reduced the energy usage of a data center cooling system by 40%).
  • AI, like the steam engine, is a “general purpose technology,” and it will increase productivity directly and indirectly by spurring many complementary technologies.
  • Historical evidence shows that productivity gains lag decades behind the development of a technology; e-commerce took two decades to reach 10% of total sales.
  • In the last wave of computerization, the gains from the new technology were linked to investments in intangible assets about ten times as valuable as the computer hardware itself; intangible assets associated with AI could require even more investment.
  • AI is a type of intangible capital, and AI-related assets such as data sets, human knowledge, and business processes will not appear “on the books;” traditional measures of GDP will fail to capture the effect of AI diffusion.



Erik Brynjolfsson

About Erik Brynjolfsson

Erik Brynjolfsson is the Jerry Yang and Akiko Yamazaki Professor and Senior Fellow at the Stanford Institute for Human-Centered AI (HAI), and Director of the Stanford Digital Economy Lab. He also is the Ralph Landau Senior Fellow at the Stanford Institute for Economic Policy Research (SIEPR), Professor by Courtesy at the Stanford Graduate School of Business and Stanford Department of Economics, and a Research Associate at the National Bureau of Economic Research (NBER).