Who’s Afraid of an AI Jobs-Apocalypse? Not America, Apparently

The alarms began sounding soon after the arrival of ChatGPT: Artificial intelligence is coming for jobs—good jobs, white-collar jobs.

“Generative AI could substitute up to one-fourth of current work,” Goldman Sachs warned. Automation might take over tasks accounting for nearly 30% of hours worked across the US economy by 2030, McKinsey said. The International Monetary Fund said that in advanced economies “about 60% of jobs are exposed,” and that of these, “about half may be negatively affected by AI.”

Generative AI is threatening because it cheaply does things that lie at the heart of the white-collar workforce—tasks like writing, analyzing, programming, and creating.

There are 168 million workers in the US workforce, of whom about 60%, or 100 million, are classified as knowledge workers. If “only” 10% of them were to lose their jobs, that would be 10 million people newly on the dole, compared with the current 6.5 million unemployed. Another 10 million not paying taxes, or contributing to Social Security, and lacking the means to pay for their families’ livelihoods.

It isn’t Great Depression levels, but it would be a societal crisis nonetheless.

So where’s the urgency? There isn’t much in the way of big-picture discussion or planning. President Joe Biden’s AI executive order does call on the Labor Department to consider “AI-related workforce disruptions,” but no one seems terribly worried in Washington. And with a few exceptions—notably the Hollywood actors who walked out last year with AI-related concerns at the center of their demands—unions have mostly been silent on the issue.

Probably one reason is that, despite some dire forecasts, the history of technological change is pretty reassuring. Innovation has always been disruptive, but even as some jobs have been lost, new ones have been created; buggy whip makers, famously, made way for auto mechanics. Technological innovation has raised growth and living standards.

Indeed, none of the economists I spoke with see an AI-induced employment crisis as likely.

“I don’t believe in any doomsday scenarios,” said Stijn Broecke, who leads the Organization for Economic Cooperation and Development’s Future of Work Initiative. “I think we’ll have more jobs, better jobs.”

Different This Time?

Still, there’s never been anything quite like generative AI: It works 24 hours a day all year long, never complains or seeks to unionize, and appears in the financial accounts as a capitalized software cost—rather than as wages and benefits. The dollars-and-cents arguments for replacing humans with machines might never have been stronger.

Maybe we’re having trouble getting our heads around it because of generative AI’s unprecedented speed of advance. Previous technological revolutions played out over years (the internet), decades (internal combustion engines) or even centuries (ironworking). ChatGPT, in contrast, had hundreds of millions of users within months of its launch, and we’re very early in the game: Companies like Google, Nvidia, Apple, and Meta are directing billions of dollars and top talent into their own large language models, and they’re all getting better quickly.

To the extent that labor advocates are focused on AI, they are worried not mainly about job losses, but rather about potential AI-related harms, like algorithmic hiring bias and employer monitoring of workers, said Virginia Doellgast, professor of employment relations at Cornell University. That’s because we continue to enjoy tight labor markets, and there’s no evidence so far of an AI jobs shock, she said.

Broecke, the OECD economist, said the main concern was frictional unemployment, in which technology changes so fast that it brings short-term job losses that we’re ill-equipped to manage.

“Even though AI will be net positive over the longer term, I can imagine a scenario where the frictional unemployment is quite high,” he said. “That’s why it’s important that we invest in our systems—unemployment, training, social dialogue.”

Curiously, some of the most urgent warnings have come from the tech industry. OpenAI CEO Sam Altman has cautioned about “the speed and magnitude of the socioeconomic change” AI will bring.

“GPT- 4 didn’t have this huge detectable impact on the economy, and so people were kind of like, ‘Oh well, we were too worried about that, and that’s not a problem,’” Business Insider cited him as recently saying. “I have a fear that we just won’t take that one seriously enough going forward, and it’s a massive, massive issue.”

What if we’re actually heading for a situation in which we suddenly have millions of unemployed and potentially unemployable white-collar workers?

One proposal, endorsed by politicians like Andrew Yang and techbros including Altman and Elon Musk, is for the creation of a universal basic income. Yang’s version would be a “Freedom Dividend, a universal basic income of $1,000/month, $12,000 a year, for every American adult over the age of 18.”

Covid proved the government could step up quickly in an emergency to keep people solvent and the economy afloat. If the most dire forecasts about AI and jobs turn out to be true, the economic winners might see the logic of supporting a universal basic income for the sake of social stability.

Still, it seems unlikely that a Washington establishment that can barely agree to continue funding itself would be capable of moving rapidly to build a new safety net on the necessary scale.

Tax Policy to the Rescue?

In the meantime, we could try to slow the uptake of AI with “robot taxes”—endorsed at various times in another context by luminaries including Bill Gates and Sen. Bernie Sanders. The theory is that we could tax artificial intelligence to raise the cost of using it in jobs that could still be done by humans.

One runs into obvious problems. Taxing tangible robots on an assembly line would be one thing, but we don’t have experience in taxing machines that exist only on a server.

Daron Acemoglu, a professor at Massachusetts Institute of Technology who has studied the issue, wrote me that he doesn’t support robot taxes, “because robots are not the only problem. Automation technologies in general are being applied excessively, so there may be a case for a broader automation tax, but how do you deal with AI?”

Orly Mazur, a professor at the SMU Dedman School of Law in Dallas, said a robot tax would be “very hard to design, implement and administer. How do you know if it’s the AI producing the income or if it’s the labor?” And maybe taxing AI would just drive the capital offshore.

“At the end of the day,” she said, “the robot tax isn’t going to do what it purports to do, it’s not going to stop the job losses or reduce social inequality and inequity.”

Perhaps we could instead target a pro-robot bias in the tax code? Acemoglu and his colleagues have shown that the US tax system actually subsidizes automation.

“For every dollar a worker receives,” they note, “the business that employs them has to pay an additional 25 cents in taxes, effectively making the cost of labor to employers 25% higher. In contrast, the average tax rate on software and equipment stood at about 15% in the 1990s and fell to about 5% after a series of tax reforms in the 2000s and 2010s.”

Ryan Abbott, University of Surrey law professor who studies AI and the law, put it this way: “If my university replaces me with a chatbot, they don’t have to pay employment taxes to the federal government, and that’s a pretty substantial savings, just on that. If I’m even marginally close to a chatbot in performance, there’s an incentive to replace me.”

Real Talk on AI is an occasional column exploring artificial intelligence and the changing workplace. If you have a column idea email djolly@bloombergindustry.com, subject line: real talk.