Cutting through the AI noise

When the “godmother of AI” says the rhetoric has gotten out of hand, it’s time to listen.

“It’s the hyperbole,” said AI pioneer Fei-Fei Li when asked at a recent Policy Forum hosted by the Stanford Institute for Economic Policy Research (SIEPR) if anything disappoints her about the technology’s sudden shift from sleepy science to a world-changing phenomenon akin to the discovery of electricity.

Li, who entered the field a quarter century ago and is now the founding co-director of the Stanford Institute for Human-Centered Artificial Intelligence, says today’s AI conversation centers on two extremes: It’s either “total extinction, doomsday, machine overlord” or “total utopia, post-scarcity, infinite productivity.”

People, Li said, “need to hear the facts.”

With that, Li captured the whole point of the SIEPR Fall Policy Forum. With its straightforward title, “AI & the Economy,” the Nov. 20 event brought together top technologists, economists, business leaders and government officials who, like Li, are committed to understanding AI’s opportunities and risks and to finding ways to ensure that the benefits are widely shared.

“It’s not a stretch to say that there is no issue more timely, more engaging and more important than the economics of AI,” Neale Mahoney, the Trione Director of SIEPR and the TG Wijaya Professor of Economics in the Stanford School of Humanities and Sciences, told a packed audience in kicking off the daylong event.

Challenged by Mahoney to focus on the concrete and to strive for consensus as to the most critical questions, an extensive lineup of speakers — among them U.S. Senator Mark Kelly; James Manyika, a Google senior vice president; and Samuel Hammond, chief economist at the Foundation for American Innovation — delivered. They explored AI’s impacts on the workforce and productivity, on energy supplies, on the U.S.-China rivalry and developing countries, and on market competition if only a handful of tech companies dominate AI services.

A key theme throughout the discussions: the role of government and how public policy can help minimize AI’s risks while maximizing its promise.

Automation vs augmentation

When it comes to AI’s impacts on workers and productivity, speakers set the record straight.

“One thing that people find surprising is that [AI’s] adoption [in the workplace] is actually pretty low and it’s very concentrated in tech, in professional services and in finance,” said Erika McEntarfer, who ran the Bureau of Labor Statistics until recently and is now a distinguished policy fellow at SIEPR, while moderating a session on preparing workers for AI. “Lots of sectors are doing very little with AI.”

This suggests that AI’s spread through the U.S. economy will take time, and will be much slower than the pace of technological breakthroughs — following what SIEPR Senior Fellow Erik Brynjolfsson describes as a “productivity J-curve” trend line. 

“We’re still in the very early innings of a long game,” said Maria Flynn, the CEO of Jobs for the Future, before warning against overstating AI's jobs impacts today but also underestimating what they might be in the future. 

Because there’s no one-size-fits-all AI strategy, companies will need to experiment first to see what works for their particular business. What’s more, companies face many bottlenecks, big and small, to integrating AI into their operations, said Neil Ashe, the CEO of Acuity, an industrial lighting company based in Atlanta.

“The technology is not the hard part,” he said. “It’s the changing-the-company part that’s hard.”

If, for example, professional services companies like law firms stop hiring entry-level workers, they will close off a key pipeline for developing rainmakers and future leaders — a phenomenon that Policy Forum speakers called “eating your seed corn.”

Lorraine Bardeen, a Microsoft corporate vice president for AI strategy and transformation, said many global companies, Microsoft included, are more interested in figuring how AI can augment workers rather than fully automate jobs. “We [now] have an autonomous sales agent that works 24/7 and responds to digital leads for small businesses that our workforce would never get to,” Bardeen said by way of example.

Research, too, suggests AI may be enabling a “democratization of expertise” within companies, said Fabrizio Dell’Acqua, a postdoctoral researcher at Harvard Business School. His recent co-authored study of a Procter & Gamble experiment with AI in its research and development unit suggests that AI helps workers with lower technical skills contribute to the innovation process but has almost no impact on higher ups.

Lindsey Raymond, who also studies the effects of digital technologies on labor markets, described research with Brynjolfsson, director of the Stanford Digital Economy Lab, that reached a similar conclusion: AI helped less-experienced workers in a Fortune 500 call center learn and work faster, but not for more experienced employees.

Speakers described other ways that AI supports workers or could be doing so soon. Bardeen, for example, talked how Land O’ Lakes, a Microsoft partner, is deploying AI to help its farmers boost crop yields and control costs. McEntarfer described the nation's shortage of electricians and how an AI avatar working with trainees can accelerate learning while lowering high turnover rates by flagging safety issues right before they happen. "We lose a lot of apprentices the first time they shock themselves," she said.

Policy challenges

The experts weren’t being pollyannish. AI will disrupt the labor market, they agreed. Mohamed El-Erian, a professor at the University of Pennsylvania’s Wharton School, said he worries that over time the U.S. economy will fall into a pattern of declining employment as workers get replaced and GDP rises because of AI.

Avoiding that scenario of widening inequality, according to El-Erian, requires that workers be open to retooling, companies adopt AI as labor-enhancing and not just cost-minimizing, and governments proactively adopt policies to mitigate job losses.

“I think of this as a shared responsibility,” El-Erian said. “If you focus on just one element of these three, you’re going to get it wrong.”

Getting the public policy piece right is especially tricky, several speakers agreed, in part because AI is advancing so rapidly that governments may not be able to keep up with its far-reaching effects, including on energy infrastructure, public trust, and national security.

“When it comes to AI policy, the picture of what needs to get done and what will get done one year from now, two years from now, is still very unclear,” said Vivek Viswanathan, who served in the Biden administration before joining SIEPR as a policy fellow. Viswanathan moderated a session on the public policy response to AI.

Senator Mark Kelly (D-Ariz.), in a keynote speech, said that if the U.S. doesn’t lead on addressing AI’s threats and seizing its opportunities, the country risks ceding the AI race to geopolitical rivals like China. In his remarks and subsequent conversation with SIEPR policy fellow Ramin Toloui, a chief organizer of the event, Kelly talked about his new roadmap, “AI for America.” His proposal would, among other things, establish a fund, financed by the AI industry, to pay for investments in energy and water infrastructure and in programs to help workers.

“Major tech companies benefiting from the rise of AI should pitch in,” Kelly said.

The policy challenges aren’t just a U.S. problem. Much of the session on the public policy response focused on the dilemma faced by governments in developing countries: While AI promises to enable giant leaps forward in health care, education and entrepreneurship, taking advantage of those opportunities either means pouring scarce resources into developing homegrown AI or handing foreign AI companies — likely American or Chinese — an uncomfortable amount of control over government services.

“If somebody somewhere else has a kill switch for your economy, that’s bad,” said Susan Athey, a SIEPR senior fellow and The Economics of Technology Professor at the Stanford Graduate School of Business who is currently an adviser to the World Bank on what AI means for developing countries.

Era Dabla-Norris, deputy director at the International Monetary Fund, agreed governments around the world have to make some hard decisions around AI. But there’s one step they can be taking today, at least to address the possibility that AI will drive up unemployment: Get rid of corporate tax incentives that effectively encourage companies to automate jobs.

“This is a policy choice,” Dabla-Norris said. “And this is something countries can do now. They don’t have to wait five years [for AI’s effects to become clear].”