AI’s Relationship With Costs and Prices
A recent working paper by researchers Tina Highfill and Jon D. Samuels of the U.S. Bureau of Economic Analysis (BEA), building upon their previous related paper, seeks early empirical evidence on the relationship between artificial intelligence (AI), costs, and prices at the macroeconomic level and how those relationships would potentially be observable in the U.S. national accounts.
If a firm can economize on the use of other inputs by using AI, the authors note, the decrease in demand for those other inputs may affect their prices. And if the firm operates in a competitive environment, input cost reductions would likely be passed on through prices paid by consumers or other businesses. Alternatively, deployment of AI could theoretically lead to increased costs and prices, especially in the shorter run. For example, if AI is expensive to employ and requires advanced engineers and equipment, along with drawing a significant amount of energy, AI could be associated with increased costs and prices.
The real-world adjustments to new technologies are, of course, not as seamless as a theoretical, perfectly competitive economic model would predict. Thus, the authors do not attempt to formulate a structural model to capture the many potential tradeoffs introduced with the application of AI; rather, they investigate macroeconomic outcomes empirically using currently available economic data. In particular, they ask the question: is the deployment of AI associated with observed measures of changes in prices of inputs and outputs within the current version of BEA's Industry Economic Accounts?
The authors frame that question within the Integrated Industry-Level Production Account, a collaboration between BEA and the U.S. Bureau of Labor Statistics. With some simple manipulation, this account shows the industry-level sources of price growth in the U.S. economy and how changes in the prices of factors of production relate to changes in purchaser prices. An objective of the paper is to discuss how these prices are measured, their potential relationship to the use of AI, and measurement challenges that may be pertinent, and to conduct a simple empirical exercise that examines the relationship between the use of AI and price change.
The paper finds early macro-level evidence that industries with relatively high AI intensity had slower output price growth in 2022–2023 than less AI-intensive industries, by about 2 percent per year in their baseline estimates. It also finds that these industries shifted their cost structure away from labor and toward capital and materials, with labor shares falling for both college and noncollege workers. The lower price growth in AI-intensive industries appears to be linked especially to smaller labor and materials price contributions, which is consistent with AI reducing some production costs, though the authors are careful not to assert causality.
For future study, the authors state that it is crucial to develop measures of industry spending on AI and constant-quality price deflators for this spending. This would enable macroeconomic analysts and researchers to directly assess the relationship between AI and the sources of economic growth instead of the indirect relationship this paper attempts to parse out.
