Last weekend, Planet Money traveled to San Francisco to attend the annual meeting of the American Economic Association (and other business organizations). This is the largest professional event for American economists. Eco-nerds met, networked, tried to get jobs, shared ideas, and published hundreds of academic papers. The hottest topic so far at this year’s conference was artificial intelligence. Research presented at the conference shows that large-scale job slaughter due to AI is not happening (at least not yet), and that AI will not completely replace employees (at least in most industries), but rather reduce employee productivity. The ultimate impact of AI on inequality is still a mystery.
AI wasn’t just a big theme within the conference. As you cross the Bay Bridge and enter the city, you’ll see billboard after billboard advertising AI products and services. Right now, Waymo’s driverless taxis are roaming the streets endlessly. In fact, our group each attended for the first time this weekend. It was a daunting experience. Sitting in the back seat and watching the car turn and move with no one in the driver’s seat, it felt like the future was already here.
All the buzz around AI has made Erik Brynjolfsson, a Stanford University economist and friend of Planet Money, one of the busiest people at this year’s conferences. Brynjolfsson is director of Stanford University’s Digital Economy Institute and one of the most prolific scholars in technology economics. Brynjolfsson spent the weekend hosting numerous panels, attending talks and lunches, answering questions, and posing for photos with fans as if he were a movie star. In fact, after he agreed to interview me after the panel discussion, I had to wait in front of him through a line of conference attendees asking for photos and a few minutes of their time.
“In some ways, the American Economic Association is being taken over by AI like other industries and professions,” Brynjolfsson told me after I was finally able to sit down for an interview. “People are using AI in research and doing research, and that’s a big change compared to a few years ago.”
It is still fresh in my memory that Planet Money rode in an unmanned taxi. Our drive went smoothly (unlike a recent passenger who got stuck in a spinning Waymo on the way to the airport and missed his flight). Despite some problems with self-driving cars, it seems clear that taxi drivers and other people who drive for a living have real reasons to fear for their jobs, and perhaps soon. To. Meanwhile, many other workers are worried that ChatGPT and other “generative AI” tools will be used in their jobs. However, perhaps unlike in the case of self-driving cars, there is reason to believe that fears that generative AI will displace office and other jobs are overblown.
Many of the predictions that AI will take away jobs have not come true, at least not yet. For example, AI godfather Jeffrey Hinton, who co-received this year’s Nobel Prize in Physics for his research in this area, said in 2016 that “people are now starting to think He should stop training as a radiologist immediately.” Read images within 5 years. That was more than eight years ago, and Brynjolfsson says not only do radiologists still exist, but their work is actually increasing.
Brynjolfsson said Hinton was “right about the system being very good, but what we’ve found is that there’s actually a higher demand for radiologists now than in 2016.” said.
A big reason for this, Brynjolfsson said, is that radiologists do more than just analyze medical images. “According to my research, they perform about 27 different tasks. One of them is reading medical images, but they will also do many other things, such as consulting with doctors and patients. These That’s what we don’t want machines to do.” It’s what we should do to people, and that’s what we do in almost every job in the economy. ”
Forecasters predicted that once AI proves it can do the job better and faster, the jobs of language translators and interpreters will soon disappear as well. However, as we reported in our Planet Money newsletter, jobs for translators and interpreters still appear to be on the rise.
Almost every job is actually a bunch of different tasks. AI may help with some of these or fully automate them, but Brynjolfsson says it’s “nearly impossible” for AI to do all of them. “If that were the case, it would actually eliminate entire jobs. But in most cases, there are some parts of the job that humans can do better, and we need to keep that in mind when designing systems.” Rather than thinking that machines can do everything, we want to create systems where humans and machines work together to produce better results.”
Brynjolfsson says the big takeaway from the paper presented at this year’s Economics Conference is that AI won’t just take jobs away from humans. “A lot of people are concerned about AI and jobs,” Brynjolfsson says. “But I think one of the big themes is that AI has complemented workers as much as it has replaced them, if not more so. When you replace something, you are replacing it and alleviating it.”But when you complement it, it becomes more valuable. And economists are finding that many workers, including radiologists and other types of doctors, call center operators, and lawyers, are improving their jobs with the help of AI. They are able to serve more customers and the demand (for human labor) is actually increasing. ”
In other words, AI, at least this generation of generative AI, is proving to be no threat to the jobs of most or all workers. In fact, if anything, generative AI seems to be helping humans do their jobs better. However, just because AI can’t do the entire job doesn’t mean partial automation won’t have a negative impact on workers. For example, the introduction of AI could lead to lower job satisfaction, lower skills, and lower salaries. Also, a ride on Waymo told me that self-driving cars (without generative AI) appear to be an immediate threat to the livelihoods of many people who currently drive cars for a living. I have no choice but to.
Will generative AI reduce or widen income inequality?
One of the big questions in current economics is which types of workers will benefit from the use of AI and which will not. As previously covered in the Planet Money newsletter, early research on generative AI found that lower-skilled, lower-performing workers benefited more than higher-skilled, higher-performing workers. .
For economists like MIT’s David Autor, these early studies were exciting. Leading economics research has found that for decades, technology primarily benefits workers at the top and not those at the bottom. This is a big reason why economists think income inequality has widened in recent decades. If AI is doing something different, if it only complements workers at the bottom and doesn’t really impact or even harm those at the top, then It could help reduce inequality and, with the right policies in place, it could also help rebuild the middle class.
Brynjolfsson himself has published research that found that lower-skilled workers benefit more from the use of generative AI than higher-skilled workers.
“The preponderance of evidence at this point is that there are examples of AI complementing workers, particularly low-skilled workers, which could potentially eliminate some of the income inequality,” Brynjolfsson said. says. “In fact, overall income inequality has decreased in recent years, and it’s not necessarily just because of AI. But the technology is not the same as it was 10 or 20 years ago.”
However, current empirical evidence is mixed. Like computers and the internet before them, AI could ultimately increase inequality.
At the conference, we saw research papers presented suggesting that AI could increase inequality by making already successful entrepreneurs even more successful. The paper is called “The Heterogeneous Impact of Generative AI on Entrepreneurial Performance” and is by a team of five economists from the University of California, Berkeley and Harvard Business School. Economists conducted a randomized controlled trial among Kenyan entrepreneurs. They provided entrepreneurs with access to an AI mentor (powered by GPT-4) who provided business advice. The economists found that “high performers benefited just over 20% from AI advice, while low performers fared about 10% worse off from AI assistance.” This study suggests that high-performing entrepreneurs had the judgment to ask AI the right questions and find viable and profitable strategies to improve their businesses. Masu. Low-performing entrepreneurs, on the other hand, sought advice “on particularly difficult problems,” but lacked the judgment to determine which AI advice would actually help them. As a result, the use of AI had a negative impact on performance. In other words, AI appears to be increasing inequality among this group of entrepreneurs.
Another recent study by Massachusetts Institute of Technology economist Aidan Toner Rogers found similar findings. The study looked at what happened to the productivity of more than 1,000 scientists working in research and development labs at large companies after they had access to AI. Toner-Rogers found that “the bottom third of scientists saw little benefit, while the output of top researchers almost doubled.” Again, AI will benefit those who can figure out how to use it well, and in many fields top performers may become even more top performers, thereby increasing inequality. suggests.
“So there’s evidence in both different directions,” Brynjolfsson says. But Brynjolfsson stresses that how AI will impact inequality and other important economic outcomes is not a given. “We have a choice in how we design our systems.” If we design AI systems (and public policies) well, AI could lead to more broadly shared prosperity. , says Brynjolfsson. “And one of the things I would ask business leaders and technologists to do is to use AI to enhance humans and benefit humanity more broadly, rather than all the benefits going to a very small group. I want them to think more seriously about what they can bring to the table.”
As I walked around the grounds of this year’s Economics Conference, I found myself wondering what might have been going on at this annual conference back in the mid-to-late 1990s, when the Internet was just starting to take off. What were economists saying at the time? Were their insights and predictions correct? Did they also know that while the Internet was transforming the economy, they were also in the midst of a dot-com bubble that was about to burst? Are we in the midst of a similar bubble and economists aren’t focusing on it enough?
Indeed, some overvalued AI companies and applications may never come to fruition, Brynjolfsson says. But, he says, “I do believe that AI will have a bigger impact on the economy than the internet. And the internet was pretty big.”
