AI may be killing jobs; will it pay off?

OPINION:

The U.S. employment market is in the doldrums.

Since May, the economy has added about 29,000 jobs monthly, compared with about 140,000 during the first Trump and Biden administrations.

If you’ve lost a job or recently graduated without skills in a hot area such as nursing, job searches are long and frustrating.

Several big firms with growing sales, including Amazon and Microsoft, have announced layoffs. Yet across the entire economy, layoff and quit rates have fallen below pre-COVID levels.

So have hiring rates because businesses are using artificial intelligence to boost productivity when workers voluntarily separate.

OpenAI, Microsoft, Meta and other tech companies are spending hundreds of billions of dollars to advance AI models. The broader economy is benefiting, but the unemployed are not.

Investments in equipment and software are holding up private investment spending. Along with moderate growth in consumer spending, those are sustaining growth in gross domestic product through the wrenching changes imposed by President Trump’s tariffs and immigration policies.

In the 1990s, the buildout of the internet featured businesses blanketing the country with millions of miles of fiber optic cable that far exceeded the demand that emerged. Stock prices soared but ended in a bubble that burst in March 2000.

Could this happen with AI?

AI investments consist of firms such as OpenAI, Microsoft and Meta spending heavily to develop programs such as ChatGPT, Copilot and Llama, building data centers with Nvidia chips and other hardware to train and run those, and creating agents to perform business and consumer tasks.

Similarly investing are spinoffs such as Anthropic, founded by several former OpenAI employees, and neoclouds such as CoreWeave, Nebius Group and Nscale Global Holdings Ltd. that build data centers to lease access to Nvidia’s chips.

McKinsey estimates that worldwide spending on data centers and electrical capacity could reach $6.7 trillion by 2030.

In 2023 and 2024 alone, leading tech firms committed sums to build data centers that exceed the amount spent more than 40 years to build interstate highway systems, adjusted for inflation.

More than 100 million Americans use AI programs, but 97% don’t pay a nickel for it.

According to a Massachusetts Institute of Technology study, 95% of businesses say they are getting little return on their AI investments. That may explain why they are reluctant to shell out more than the $30 a month per user that Microsoft charges for Copilot or something similar.

The free AI functions embedded in Google’s search engine serve my research needs well.

In 1794, Eli Whitney patented the cotton gin. It enabled the South’s cotton economy to flourish in the 19th century, but he didn’t profit much from it. For one thing, patent laws had lots of loopholes, but another problem was that farmers could easily copy the device — that is, build their own. It’s hard to profit from an invention if folks can build a similar machine for personal use.

That also may be happening with AI.

AI programs are displacing software engineers, and AI is learning how to build agents for ordinary folks.

In any case, economists are notoriously bad at measuring the consequences of transformative technologies in real time.

During the 1980s, desktop computers emerged and reduced head counts in clerical and administrative tasks. In the 1990s, the internet accelerated this process. Yet economists couldn’t find it in the data. The Nobel Prize-winning economist Robert Solow lamented, “You can see the computer age everywhere but in the productivity statistics.”

When I repeated this to my attorney in 1995, he responded that before desktop computers, his firm had seven lawyers and 10 support personnel, but with computers, those numbers had reversed.

Once again, economists might do less data mining and simply look around and talk with CEOs.

With robots aided by AI, Walmart, like Amazon, has automated its warehouses, and CEO Doug McMillon anticipates a major impact on virtually all white-collar jobs.

The company is building an AI agent for its customers, suppliers and workers, just like the Southern planter building his own cotton gin.

He wants to keep the head count constant over the next three years as the retailer increases sales.

The ability to apply AI and collaborate effectively with agents, colleagues and customers is what will keep someone employed.

How will AI pay for itself?

Likely the same way web browsers and social media do now: by selling advertising and products. Last year, OpenAI added a search engine to ChatGPT to challenge Google. This year, it will let users buy products by linking up with a shopping platform.

At Walmart, Mr. McMillon sees some existing positions, such as delivery drivers, increasing, but most others transformed or disappearing and new positions created. One of those is “agent builder.”

For most unemployed people and recent graduates, becoming AI-fluent and effective collaborators with both machine and man will be key to breaking through to start a new or first career ladder.

• Peter Morici is an economist and emeritus business professor at the University of Maryland, and a national columnist.