Unlocking the Future: Insights from America’s Leading Economists on AI and the Inequality Challenge

This past weekend, Planet Money made its way to San Francisco to participate in the yearly gathering of the American Economic Association (along with other economic associations). It is the largest professional event for economists in the United States. Economic enthusiasts gathered, networked, sought employment, exchanged ideas, and showcased numerous academic papers. By far, the most discussed issue at this year’s conference was artificial intelligence. The research shared at the event indicated that the anticipated mass job losses due to AI are not occurring (at least not just yet), that AI is more likely to enhance worker productivity rather than simply replace them (at least across most sectors), and that AI’s ultimate impact on inequality remains uncertain.

AI was not only a significant subject within the confines of the conference. While crossing over the Bay Bridge into the city, billboard after billboard promoted AI products and services. A continuous flow of Waymo self-driving taxis now zoom around the city. In fact, a group of us experienced one this weekend, each for the first time. It was astonishing. Sitting in the backseat, witnessing the car maneuver and operate without anyone in the driver’s seat — it felt as if the future had arrived.

All the excitement surrounding AI made Stanford economist and acquaintance of Planet Money Erik Brynjolfsson one of the most sought-after individuals at this year’s conference. Brynjolfsson leads the Stanford Digital Economy Lab and is a highly regarded scholar of technology’s economic aspects. Brynjolfsson dedicated the weekend to moderating various panels, engaging in discussions and meal gatherings, addressing inquiries, and taking photographs with admirers as if he were a celebrity. Indeed, after he consented to let me interview him following a panel session, I had to wait momentarily as a queue of conference attendees formed in front of him, each eager for pictures and/or a moment of his attention.

“The American Economic Association is, in a sense, being influenced by AI just as every other industry and profession,” Brynjolfsson informed me, after he finally had the chance to sit for an interview. “Researchers are employing AI in their work and studying it — and it’s a transformative shift compared to what it was just a few years ago.”

Fresh on my mind was Planet Money’s journey in a self-driving taxi. The ride was quite pleasant for us (unlike a recent passenger who found himself trapped in a circling Waymo on his way to the airport and missed his flight). Despite certain issues with driverless vehicles, it appears evident that taxi drivers and others who rely on driving for a living have valid reasons to be concerned about their job security — and quite soon. Simultaneously, many other employees fear that ChatGPT and other “Generative AI” technologies will threaten their employment — although, perhaps unlike autonomous vehicles, there are indications that apprehensions about Generative AI jeopardizing office jobs and other roles may be exaggerated.

Many of the forecasts regarding AI wiping out jobs have — thus far — not come to fruition. For instance, AI pioneer Geoffrey Hinton, who received the Nobel Prize in physics this year for his contributions to this field, proclaimed back in 2016 that “people should cease training radiologists immediately” because it was “completely evident” that AI would surpass them in interpreting images within five years. That was over eight years ago — and not only are radiologists still present — but Brynjolfsson asserts that opportunities for radiologists are actually increasing.

Hinton, as Brynjolfsson explains, was “accurate about the systems becoming extremely proficient. However, as it turns out, we actually demand more radiologists now than we did in 2016.”

The primary reason, according to Brynjolfsson, is that radiologists handle many more responsibilities than merely evaluating medical images. “According to my research, they conduct approximately 27 distinct tasks. One of those is analyzing medical images, but they also engage in numerous other duties, such as consulting with doctors and patients. These are tasks that you would not want a machine performing on people, and this holds true in nearly every profession in the economy.”

Forecasters anticipated that jobs for language translators and interpreters would swiftly vanish as AI demonstrated its ability to perform the tasks more efficiently. However, as we’ve covered in the Planet Money newsletter, positions for translators and interpreters seem to still be on the rise.

Virtually every job encompasses a collection of various tasks. AI may assist with or potentially automate some of them entirely, but, as Brynjolfsson points out, it can “almost never” handle all of them. “If it did, then you would indeed see the removal of the entire job. However, in most instances, there are components of the job that humans excel at. I believe, as we develop these systems, it’s essential to remember that we want to create environments where humans and machines collaborate to achieve superior outcomes rather than presuming that machines can accomplish everything.”

Brynjolfsson remarks that one of his major insights from the research presented at the economics conference this year was that AI is not merely eradicated human jobs. “Many individuals are concerned about AI’s impact on employment,” Brynjolfsson states. “Yet, one of the dominant themes has been that AI has been augmenting workers just as much as, or even more than, it has been replacing them. For those not familiar with economics, to substitute something means to replace it and diminish its value. Conversely, to complement it means to enhance its worth. And whether it involves radiologists, other types of doctors, call center representatives, or attorneys, economists are discovering that numerous workers are performing their roles more effectively with AI assistance, they’re able to cater to a larger customer base, and the demand [for their human skills] is actually increasing.”

In other words, AI — or, at least, this iteration of Generative AI — is demonstrating not to be a threat to the employment of most or all individuals. In fact, if anything, Generative AI appears to be assisting humans in improving their performance in their roles. However, just because AI cannot fulfill entire job functions does not imply that the partial automation of these functions will not have adverse implications.

effects for employees. The incorporation of AI might, for instance, render jobs less fulfilling, diminish required skills, or lead to reduced pay, and so on. Furthermore, after experiencing a ride in a Waymo, I must express that autonomous vehicles (which are not influenced by Generative AI) appear to pose an impending danger to the livelihoods of numerous individuals currently engaged in driving as a profession.

Will Generative AI Diminish Or Expand Income Inequality?

One of the significant inquiries in economics today is determining which categories of workers gain from the application of AI and which ones do not. As previously discussed in the Planet Money newsletter, some initial research concerning Generative AI has indicated that less skilled, lower-performing employees have profited more than their higher skilled, higher-performing counterparts.

For economists such as David Autor from MIT, these preliminary findings have been thrilling. Leading economic studies have revealed that, over the years, technology has predominantly favored workers at the upper end rather than those at the lower end. This is a significant reason why economists think income inequality has escalated in recent years. If AI is behaving differently — supporting those at the bottom and not significantly harming or even affecting those at the top negatively — it might help in lessening inequality and, with appropriate policies in place, perhaps even assist in rejuvenating the middle class.

Brynjolfsson himself has released studies indicating that lower-skilled workers derive more benefits from the implementation of Generative AI compared to higher-skilled workers.

“I would argue that the majority of the evidence currently points to instances where AI indeed supports workers, particularly those with fewer skills, which could, you know, potentially narrow some of the income disparity,” Brynjolfsson states. “In fact, overall income disparity has decreased in recent years, although not necessarily solely due to AI. But the technology we have today is distinct from what was available 10 or 20 years ago.”

Nevertheless, the most recent empirical data is inconclusive. Similar to personal computers and the internet in the past, AI might ultimately exacerbate inequality.

At the conference, I observed a working paper presented that proposed that AI could heighten inequality by elevating the success of those already prosperous. The paper is named, “The Uneven Impact of Generative AI on Entrepreneurial Performance,” and it is authored by a group of five economists from UC Berkeley and Harvard Business School. The economists executed a randomized controlled trial involving entrepreneurs in Kenya. They provided these entrepreneurs access to an AI mentor (powered by GPT-4), which offered business guidance. The economists discovered that “high performers benefited by just over 20% from AI advice, whereas low performers did roughly 10% worse with AI assistance.” The study implies that high-performing entrepreneurs had the acumen to pose insightful questions to the AI and identify practical and profitable strategies to enhance their businesses. Conversely, low-performing entrepreneurs solicited guidance “on particularly challenging issues” and lacked the judgment to discern which AI recommendations would genuinely benefit them. The outcome was that the use of AI negatively impacted their performance. In essence, AI appeared to amplify inequality within this group of entrepreneurs.

Another recent study conducted by MIT economist Aidan Toner-Rodgers discovered similar findings. It examined the productivity changes of over a thousand scientists at a research and development lab of a large corporation after they gained access to AI. Toner-Rodgers found that “while the bottom third of scientists see minimal benefit, the output of top researchers nearly doubles.” Again, AI favors those capable of effectively utilizing it, and it indicates that in numerous fields, top performers may enhance their performance even further, thereby intensifying inequality.

“There is evidence in both differing directions,” Brynjolfsson remarks. However, Brynjolfsson emphasizes that the impact of AI on inequality or any other vital economic outcome is not predetermined. “We have options regarding how we structure our systems.” With the appropriate design of AI frameworks (and also public policies), Brynjolfsson asserts, AI could facilitate more broadly shared abundance. “And one of my appeals to business leaders and technologists is to contemplate more thoroughly about leveraging AI to enhance human capabilities and to benefit humanity at large rather than allowing all advantages to concentrate within a very limited group.”

As I wandered through this year’s economics conference, I found myself pondering what transpired during this annual gathering back in the mid-to-late 1990s when the internet was burgeoning. What were economists pondering at that time? Were their insights and forecasts accurate? Did they recognize that while the internet would revolutionize the economy, they were simultaneously amidst a dotcom bubble that would soon burst? Are we currently in the midst of a similar bubble, and are economists insufficiently attentive to it?

Indeed, Brynjolfsson suggests, there may be some overvalued AI companies or applications that may not yield substantial results. However, he remarks, “I am quite confident AI will have a more profound impact on the economy than the internet did. And the internet made a significant impact.”