Artificial intelligence (AI) is rapidly advancing, with generative AI (genAI) models demonstrating continuous improvements in capability. These developments have sparked discussions about AI’s potential to enhance productivity and economic growth, as well as concerns about their broader impact on the labor market and business processes. As a part of a project with David M. Byrne and Paul E. Soto, we explore the economic and productivity implications of AI—particularly genAI. We investigate whether genAI qualifies as a general-purpose technology (GPT), akin to past innovations like the electric dynamo and computer that transformed the economy and fostered strong productivity growth. In addition, we examine the technology as an invention in the method of invention (IMI),1 examining the role AI is playing in scientific research and R&D. Research shows the difficulty and expense of scientific and applied development has greatly increased, making it harder to push out the frontier of knowledge. If AI can mitigate this trend and raise the productivity of research, it can, over time, raise overall productivity in the economy.
As part of this research effort and to ground our research in real-world applications, we examine four key industries where AI is poised to make a significant impact:
- Electricity generation and transmission: AI optimizes grid management and prevents outages using predictive analytics and satellite imaging.
- Health care: AI can assist with diagnosis and reduce administrative burden on tasks such as scheduling appointments and transcribing case notes.
- Finance: AI-driven risk management, fraud detection, and algorithmic trading can transform banking and investments.
- Information: AI enhances software development, customer service, and graphic design.
What are the most important lessons learned from these case studies?
First, the technology is changing at a breakneck pace. One of the great limitations of genAI has been that it predicts the right answer to questions using a model of the language found on the web, but because it might not understand the ideas in that text, it can be prone to producing incorrect and misleading outputs, known as “hallucinations.” Popularized in the past year, “reasoning” AI—where the AI is trained to provide sensible explanations for its answers—is being developed to overcome these limitations. The rapid pace of this technology’s evolution will make it exceedingly difficult for researchers (like ourselves) to predict impacts far into the future.
Second, we find enormous capability of this technology, which suggests great potential for improvements in productivity going forward. However, partly due to difficulties in measurement of productivity, particularly for industries like health care and information, these case studies are unable to pull out reliable productivity forecasts at this early stage. Therefore, we have not attempted to make quantitative, industry-wide estimates for the productivity impact of AI.
Third, barriers to adoption will be challenging to overcome. Changing business processes takes time. It is a risky effort for any company, especially small companies. It requires learning new skills, both for workers using the technology and for managers who must understand what it can do and where it can fail. Another barrier to progress is institutional inertia. In the health care sector, doctors largely control how care is delivered and, along with hospital managers, they must be persuaded that investing in a new technology is worthwhile and will pay off for them financially.
The evidence of the cases, together with the evidence presented in our forthcoming paper, strongly support the idea that new developments in AI make it a GPT that will raise productivity over time. We also find that AI will help reduce the cost and increase the productivity of research, making it an invention in the method of invention, leading to faster productivity growth in the future. We are optimistic about the productivity benefits of AI. However, we are also cautious and do not yet know how fast change will take place. Adoption of genAI has been slow, and whether the engineering achievements of AI developers can be delivered cost-effectively is unknown. However, it remains to be seen to what extent it will actually be a GPT or an IMI.
Download the AI in the electricity sector case study
Download the AI in the finance sector case study
Authors
-
Acknowledgements and disclosures
Martin Baily led the work on the health care and finance sector case studies. Aidan Kane led the work on the electricity and information sector case studies. These case studies were written as part of a joint project with David M. Byrne and Paul E. Soto of the Federal Reserve Board. We are indebted to them for assistance and helpful comments. We would also like to thank Eli Schrag for his factchecking.
-
Footnotes
- This term was coined by Alfred North Whitehead in the 19th century. However, Zvi Griliches popularized the phrase among economists.
The Brookings Institution is committed to quality, independence, and impact.
We are supported by a diverse array of funders. In line with our values and policies, each Brookings publication represents the sole views of its author(s).
