AI has potential to balance both productivity and growth
Nobel Prize-winning economist Robert Solow once quipped that one could see the computer age everywhere but in the productivity statistics. He said this in 1987, and, while there has been tremendous technological advancement since then, productivity, as measured by output per worker hour, has averaged a growth rate of slightly less than 2 percent year over year during this 34-year period.1
Productivity is central to economic growth, allowing higher standards of living and faster growth with less inflation. Economists often look to the 25 years following World War II as the golden age of productivity, as growth averaged 2.8 percent per year during the period. Similarly, today’s economists are eagerly studying developments in AI for its potential to bring about a period of rapid productivity growth in the years to come.
Economists posit that if AI can reduce the costs of the factors of production, as appears to have been the case in the COVID-19 vaccine-development process, AI has the potential to greatly improve productivity as it becomes adopted widely. Due to the iterative nature of general purpose technologies becoming adopted, one possibility of AI adoption is that it could have a logarithmic impact on future productivity. While automation has been improving on a 200-year continuum it has been limited to routine tasks since the spinning jenny advanced the industrialization of textile manufacturing. AI opens the door to automating non-routine tasks, such as self-driving cars, radiology, vaccine development, certain types of laboratory research, and even some legal services.
Even if these examples are fertile ground for more advances and widespread adoption, it is not guaranteed that AI will produce significantly higher growth. Economists consider the paradox outlined in Baumol’s cost disease.2 Sectors that experience high levels of productivity growth often become a smaller share of gross domestic product. This is because increased productivity reduces input costs and often results in lower absolute and relative prices.
The two prime examples of this paradox can be seen in agriculture and manufacturing. In the United States, the productivity of these sectors between 1950 and 2000 exceeded that of the economy as a whole. At the same time, these sectors’ GDP share fell: from 26.8 percent to 15.1 percent in manufacturing, and from 6.6 percent to 0.9 percent in agriculture.
Getting back to AI, a 2019 study produced by the National Bureau of Economic Research modeled the growth of AI specifically when it comes to automating the production of ideas.3 In this model, Baumol’s cost disease is mitigated and the model suggests explosive growth is possible if there are sufficient safeguards for protecting intellectual property. The authors suggest that AI could change the process by which new ideas and technologies are created, providing a significant boost to economies of scale. However, in order to realize this bright new future, consideration must be given to the problems of business stealing and creative destruction that is so rapid it discourages R&D, as well as business concentration and firm structure.
Ultimately, as Baumol’s insight shows, growth is determined not by what economies are good at, but by what is essential and yet hard to improve. If AI can tackle essential but hard-to-improve-upon problems, it could be the holy grail that delivers higher productivity for decades to come.
To read the entire KPMG 2021 Thriving in an AI World report, please visit: Thriving in an AI world.
- U.S. Bureau of Economic Analysis
- National Bureau of Economic Research, May 2006
- The Economics of Artificial Intelligence, Philippe Aghion, Benjamin F. Jones