How Much Are Businesses Using Artificial Intelligence?

Big improvements to artificial intelligence have brought widespread attention to the technology, along with questions about AI in the workplace. For example, how much could AI help increase productivity—which could mean higher compensation for workers? And will it displace workers?

The answers to those questions depend in part on how much AI is being used by businesses. Federal Reserve Bank of St. Louis researchers have explored aspects of companies’ AI use, including:

• What percentage of workers reported using generative AI
• How quickly businesses are adopting AI
• How business investment in AI shifted after the introduction of ChatGPT

Some takeaways from their findings are below.

But first, let’s look at what AI is and what it does.

What Is Artificial Intelligence?

Artificial intelligence describes technologies from autonomous cars to chatbots that “exhibit human-like intelligence,” as a 2018 working paper quoted in a December 2024 Page One Economics article explained.

But AI works on prediction rather than thinking or reasoning like a human. “Prediction” is the process of filling in missing information, and AI uses the information, or data, it has available to do that, according to the article, “AI and the Future of Work: Opportunity or Threat?”

Some examples? Services that predict movies or music a consumer might like based on previous selections.

“Generative AI” can “generate text, images, videos and other content in response to a user prompt,” according to a New York University guide.

One class of AI is large language models, or LLMs, which predict how humans might answer when asked a question. Popular chatbot ChatGPT, trained on “huge quantities of text,” is in the LLM class, Scott Wolla, St. Louis Fed economic education officer, wrote in the Page One Economics article.

What Percentage of Workers Are Using Generative AI?

AI has clocked a faster adoption rate than two other “transformative” technologies—personal computers and the internet—when work plus home use is taken into account, according to a September 2024 On the Economy blog post, “The Rapid Adoption of Generative AI.”

In August 2024, almost 40% of 18- to 64-year-olds in the U.S. who responded to the Real-Time Population Survey (RPS) said they used generative AI, as the bar graph below shows.

Use of Generative AI at Work and at Home, August 2024

SOURCES: Real-Time Population Survey and calculations by the authors of a Sept. 23, 2024, On the Economy blog post, “The Rapid Adoption of Generative AI.”

NOTES: The figure shows the share of RPS respondents who used generative AI for work, outside of work and overall (either for work or outside of work). Intensity of use is broken down into every day in the week before the survey, at least one day in the week before the survey but not every day, and not used in the week before the survey. Data are from the August 2024 wave of the RPS and for respondents ages 18 to 64. The “for work” sample includes only employed individuals (N=3216); the other samples include all respondents (N=4682).

That generative AI adoption rate almost two years after mass market introduction via ChatGPT compares with about 20% for internet adoption at the same point, authors Alexander Bick, Adam Blandin and David Deming noted. Bick is a St. Louis Fed economist; Blandin and Deming are economics professors at Vanderbilt University and the Harvard Kennedy School, respectively.

Personal computer adoption, meanwhile, reached 20% three years after mass market introduction.

How does AI adoption translate into workplace use and productivity? Bick, Blandin and Deming looked at how much respondents used generative AI on the days they reported using it to project a range for the share of total work hours using generative AI for the U.S. economy. The authors then used those figures, plus a median increase of 25% in task productivity from generative AI adoption, to estimate labor productivity growth. (The 25% median increase is “consistent with that observed in several studies,” the authors wrote.)

“We estimated that generative AI could plausibly grow labor productivity by between 0.1% and 0.9% at current levels of usage,” they wrote.

How Quickly Are Businesses Adopting Artificial Intelligence?

The use of “smart” devices in the U.S. increased sharply over about two years; 3D printing use has had ups and downs while trending up overall; and the growth of cloud computing has been gradual, an April 2024 On the Economy blog post showed. The researchers gauged the growth of the technologies’ use by how often they were mentioned in job postings across different urban areas.

The authors noted that job postings mentioned cloud computing in about 50% of urban areas in the U.S. in 2019. That figure was about 1% for 3D printing, while it was nearly 100% for smart devices.

“These technologies exhibit contrasting patterns of adoption, hinting at the many paths AI could take,” Aakash Kalyani, a St. Louis Fed economist, and Marie Hogan, a research associate, wrote in the blog post, “AI and Productivity Growth: Evidence from Historical Developments in Other Technologies.”

On which path did AI seem to be heading at the time of the post? Early evidence on the technology’s spread seemed to point to a pattern like that for personal computers and cloud computing, the post said.

The U.S. Census Bureau had asked businesses in several survey waves over the previous five years whether they use AI in production. The researchers examined reported use among businesses for three time periods and found:

• 2018: Around 3%
• 2023-24 (average): Around 4.4%
• Six months from survey (expected use): Less than 7%

“If history is any guide, AI productivity gains might take a long time to realize,” the authors concluded.

How Has Business Investment in Artificial Intelligence Changed?

After the release of ChatGPT in November 2022, sentences mentioning AI in quarterly earnings calls increased more than fivefold, a September 2024 On the Economy blog post reported.

Following up with an October 2024 blog post, authors Aakash Kalyani, Serdar Ozkan, Mickenzie Bass and Mick Dueholm examined whether the increase in talk translated into more action—that is, more investment. They didn’t find the increases matched, as they discussed in the blog post, “AI Hype or Reality? Shifts in Corporate Investment after ChatGPT.”

Economists Kalyani and Ozkan and research associates Bass and Dueholm analyzed the relationship between AI discussions in earnings calls and firms’ investment decisions as reported in their 10-K reports, which are annual regulatory filings with the U.S. Securities and Exchange Commission. The researchers looked in particular at discussion of AI that largely was positive in sentiment.

Before ChatGPT was released, businesses with positive views of AI spent more on capital expenditures and R&D than their peers, the authors found.

“For every net positive-sentiment sentence in a firm’s earnings call, we observed on average a 3.1% increase in capital expenditures and 9.7% increase in R&D expenditures by the company,” the authors wrote.

After ChatGPT was released, sentiment about AI was no longer linked in a significant way to capital expenditures, and the increase in R&D expenditures associated with positive AI sentiment was half what it was before ChatGPT, the blog post said.

The “before” and “after” growth percentages for R&D investment can be seen in the bar chart below. The “before” period covers the first quarter of 2012 to the third quarter of 2022, and the “after” period covers the fourth quarter of 2022 to the first quarter of 2024.

R&D Investment Growth by Positive AI Sentiment

SOURCES: S&P Global, Compustat and calculations by authors of the Oct. 3, 2024, On the Economy blog post, “AI Hype or Reality? Shifts in Corporate Investment after ChatGPT.”

What might explain the weaker connection between positive AI sentiment and investment after the announcement of ChatGPT?

“While some of this weakening might be attributable to cheap talk and corporate hype around AI, time will tell whether these changes are driven by fundamental differences in the implementation of traditional AI and new LLMs,” the authors wrote.