Americans are using AI at a significantly higher rate. What does this mean for the economy? : Planet Money : NPR

Before Harvard University economist David Deming conducted a recent survey of Americans to understand how much they were using generative AI, he considered himself a staunch “AI skeptic.” He said he was in a position. In other words, he was skeptical that the explosion of generative AI would bring significant benefits to the U.S. economy anytime soon. But he says he’s more optimistic now.

“We were very surprised by the numbers in our study,” Deming said. “And I started to think that AI might be a bigger problem than I thought.”

Deming said the study was motivated by the question of whether and to what extent Americans are using generative artificial intelligence. Doing what economists tend to do, he and colleagues Alexander Bick and Adam Brandin wanted to get good data.

They modeled their survey after the Current Population Survey (CPS), sponsored by the U.S. Census Bureau and the U.S. Bureau of Labor Statistics. The CPS is conducted monthly and is something of a gold standard for surveys. By doing so, you can find out things like the unemployment rate and the state of the labor market.

This means that the research conducted by Deming, Bick, and Blandin is of high quality. The sample size is large. It’s a national representative. Additionally, it included some of the same questions that CPS asks, so we could cross-reference the survey with CPS to make sure the numbers were good. They conducted the survey twice, in June and August 2024.

Deming said he was shocked by the results. He and his colleagues found that nearly 40% of Americans between the ages of 18 and 64 use generative AI. And a significant percentage seem to use it on a regular basis. Economists found in an August survey that more than 24% of U.S. workers had used it “at least once in the week preceding the survey, and nearly 1 in 9 reported it to work every day.” did.

Even more surprising, Deming says, the use of AI appears to be almost universal. He expected young, educated Americans to be the biggest users. They confirmed that to be the case. “However, we also found that 22% of blue-collar workers say they use AI, with usage rates exceeding 20% ​​in every major occupation except personal services. , it was about 15%,” he says.

The authors note that this rapid rate of adoption is much faster than the rate at which Americans adopted personal computers and the Internet. But this is not surprising, considering that in the 1980s and 1990s, personal computers cost thousands of dollars and required physical hardware to be acquired, connected, and understood. Maybe not. Even with the internet, people had to do that. Buy a modem, get a subscription, or find a way to “surf the web.” Generative AI is more plug-and-play, often free or with a low monthly subscription cost, and has a user interface that will be familiar to anyone who has used Google.

What does this mean for the economy?

In a recent two-part series on The Indicator, Darian Woods and I discussed whether AI is overrated. Given that there is so much uncertainty about the impact of AI on the economy, and given our reluctance to make predictions that may be laughed at in a few years, we have taken a personal approach to this issue. I decided to make my feelings vague. I flipped a coin generated by an AI. I learned that AI is overrated, and wrote an episode and newsletter to make the case.

During our research, we found that only about 5% of U.S. companies surveyed by the Census Bureau this year said they had used AI in the past few weeks. I used this as evidence that the use of generative AI is pretty pathetic, especially considering the hype around it.

How does Deming apply his results to that discovery?

As an example, Deming and his colleagues say they found that Americans actually report using AI more in their personal time at home than at work.

Hmm. got it. Going back to my idea that AI is overrated, I believe that much of the use of AI is for fun rather than for productive work purposes. This suggests that AI’s impact on the economy will be limited.

An economic dream of sorts for AI would be to rescue the U.S. economy from a long period of stagnant productivity growth. Increased productivity, the ability of workers to produce more in less time, is the magic sauce for improving living standards. And recent technology has been pretty disappointing in that regard.

I mean, look at your smartphone. If I had told you in 2006, the year before the iPhone was released, that we would soon have supercomputers in our pockets that could search the Internet, give us precise directions anywhere, send emails, and more. , you will be able to make video calls. – Both employees and customers can order basically any product or service, translate languages, and much more. You might think your productivity will skyrocket. But smartphones seem to be more of a tool for entertainment and distraction than an incredibly impactful work tool. Since its mass introduction over a decade ago, there has been no significant increase in productivity.

But Deming emphasizes that people are using generative AI to do their jobs. He said their research found that “about one in four people used AI at least once in the past week for work,” compared to “about one in three” who used it on their days off. The results are in. When it came to their jobs, respondents said that writing, interpreting, and administrative support were most helpful (although a significant proportion also used it for other tasks, such as coding and supporting when dealing with customers and colleagues. ).

Comparing his findings to those of the Census Bureau, Deming said he believes there is a “huge disconnect between formal corporate policies and workers regarding the use of AI.” If I want to email my boss, I just use ChatGPT to write emails faster, but I don’t tell anyone. ” Deming says many students are using generative AI in this way.

We spoke to Daron Acemoglu, an economist at the Massachusetts Institute of Technology (and one of the leading AI skeptics these days), about Deming et al.’s work. Does the fact that it turns out that a large percentage of Americans are using generative AI change his view at all?

“My concern about their numbers is that they do not distinguish between fundamentally productive uses of generative AI and occasional/frivolous uses,” Acemoglu said in an email. said. “If you want to know what ChatGPT says to introduce one of our guests and what it uses to understand it, you’re a user, but this is fundamentally not integrated into NPR’s production process. Hmm. Or worse, if you ask me, I use generative AI. Honestly, I have to say that because when I search, I see the output that generative AI produces. But this does not affect at all the productivity improvements that are needed for me: the fundamental integration and reorganization of production processes and complementary investments in organizational capabilities and employee skills. We know that ChatGPT has around 200 million unique users per month, but the question is how many of them are using it in any way to significantly increase productivity or reduce costs. I don’t really know the answer to that question, and I don’t think I’ll find the answer in this paper.”

How much does generative AI improve productivity?

To predict how much productivity gains AI would increase, Deming and his colleagues did some rough calculations. They reviewed five randomized studies that analyzed how much the use of generative AI improved productivity in different work environments. They chose a productivity improvement number that was in the middle of these studies: about 25%. Deming says they multiplied this by “an estimate of the number of work hours currently supported by generative AI,” which they calculated was 0.5% of all U.S. work hours. 3.5%).

They ultimately estimated that generative AI would deliver “a 0.125 to 0.875 percent increase in labor productivity at current levels of use.” It may not seem like it. But Deming says, “consider that productivity growth over the past few decades has been about 1.5% a year. So when you add that to that 1.5%, it’s actually quite a large increase.” ”.

Interestingly, Acemoglu also makes some rough calculations about the potential impact of generative AI on the economy in the near future. And despite using a different method to calculate the potential productivity gains from AI, his estimates are actually quite similar to a recent study by Deming and colleagues.

However, Acemoglu and Deming had different interpretations of the results. Acemoglu seems to be reacting further to all the media and industry hype around generative AI. And he seems to be saying that we are by no means witnessing an economic revolution. Noting the potential productivity gains, Deming further emphasized that generative AI could have a meaningful impact on the U.S. economy.

“Will that translate into a 7% productivity increase? No, probably not,” Deming says. “Although not in the way it is currently used, incremental additions can be made and every little bit counts. Doing so could generate millions of dollars in additional GDP growth and improved standards of living. That’s really important.

Deming and his colleagues plan to continue their investigation. It’s worth noting that before Apple launched the iPhone with ChatGPT integration, they conducted one last study.

Soon, he says, “AI will be in so many things, including the iPhone, that it will be difficult to even ask people, ‘Are you using AI?'”