How ChatGPT Will Destabilize White-Collar Work

No technology in modern memory has caused mass job loss among highly educated workers. Will generative AI be an exception?

Illustration of vintage computer monitors with binary code superimposed
Michael Brennan; Getty; The Atlantic
Illustration of vintage computer monitors with binary code superimposed

Listen to this article

In the next five years, it is likely that AI will begin to reduce employment for college-educated workers. As the technology continues to advance, it will be able to perform tasks that were previously thought to require a high level of education and skill. This could lead to a displacement of workers in certain industries, as companies look to cut costs by automating processes. While it is difficult to predict the exact extent of this trend, it is clear that AI will have a significant impact on the job market for college-educated workers. It will be important for individuals to stay up to date on the latest developments in AI and to consider how their skills and expertise can be leveraged in a world where machines are increasingly able to perform many tasks.

There you have it, I guess: ChatGPT is coming for my job and yours, according to ChatGPT itself. The artificially intelligent content creator, whose name is short for “Chat Generative Pre-trained Transformer,” was released two months ago by OpenAI, one of the country’s most influential artificial-intelligence research laboratories. The technology is, put simply, amazing. It generated that first paragraph instantly, working with this prompt: “Write a five-sentence paragraph in the style of The Atlantic about whether AI will begin to reduce employment for college-educated workers in the next five years.”

ChatGPT is just one of many mind-blowing generative AI tools released recently, including the image generators Midjourney and DALL-E and the video generator Synthesia. The upside of these AI tools is easy to see: They’re going to produce a tremendous amount of digital content, quickly and cheaply. Students are already using ChatGPT to help them write essays. Businesses are using ChatGPT to create copy for their websites and promotional materials, and to respond to customer-service inquiries. Lawyers are using it to produce legal briefs (ChatGPT passes the torts and evidence sections of the Multistate Bar Examination, by the way) and academics to produce footnotes.

Yet an extraordinary downside is also easy to see: What happens when services like ChatGPT start putting copywriters, journalists, customer-service agents, paralegals, coders, and digital marketers out of a job? For years, tech thinkers have been warning that flexible, creative AI will be a threat to white-collar employment, as robots replace skilled office workers whose jobs were once considered immune to automation. In the most extreme iteration, analysts imagine AI altering the employment landscape permanently. One Oxford study estimates that 47 percent of U.S. jobs might be at risk.

No single technology in modern memory has caused mass job loss among highly educated workers. Will generative AI really be an exception? No one can answer this question, given how new the technology is and given how slowly employment can adjust in response to technological change. But AI really is different, technology experts told me—a range of tasks that up until now were impossible to automate are becoming automatable. “Before, progress was linear and predictable. You figured out the steps and the computer followed them. It followed the procedure; it didn’t learn and it didn’t improvise,” the MIT professor David Autor, one of the world’s foremost experts on employment and technological change, told me. ChatGPT and the like do improvise, promising to destabilize a lot of white-collar work, regardless of whether they eliminate jobs or not.

People and businesses are just figuring out how to use emerging AI technologies, let alone how to use them to create new products, streamline their business operations, and make employees more efficient. If history is any guide, this process could take longer than you might think. Consider electricity. The circuit, electric lights, and rudimentary electric motors were developed in the early 1800s. But another century passed before the widespread adoption of electricity in the United States began to lift GDP. Or take computers. They became commercially available in the early 1950s but did not show up in the productivity stats until the late 1990s.

Some technologies clearly improve productivity and reduce the need for labor. Automated machine tools, for instance, depress manufacturing employment while lifting output and productivity, as do many of the forms of machinery invented and employed since the Industrial Revolution. But other technologies—even amazing ones—show surprisingly muted effects. How about the internet, which has revolutionized almost every facet of communications in the past four decades? Despite altering how we date and talk and read and watch and vote and emote and record our own life stories, launching a zillion businesses, and creating however many fortunes, the internet “fails the hurdle test as a Great Invention,” the economist Robert Gordon argued in 2000, because it “provides information and entertainment more cheaply and conveniently than before, but much of its use involves substitution of existing activities from one medium to another.” Nearly a quarter century later, the internet still hasn’t spurred a productivity revolution. Smartphones haven’t either.

So is AI like the smartphone or is it like an automated machine tool? Is it about to change the way that work gets done without eliminating many jobs in aggregate, or is it about to turn San Francisco into the Rust Belt?

Predicting where technology will cause job losses is hard, Autor noted. Remember the freak-out several years ago over the possibility of self-driving automobiles eliminating work for truck drivers? But AI is much more flexible than a system like Excel, much more creative than a Google Doc. What’s more, AI systems get better and better and better as they get more use and absorb more data, whereas engineers often need to laboriously and painstakingly update other types of software.

As a rule, when companies can substitute machines for people, they will. AI can do work currently done by paralegals, copywriters, digital-content producers, executive assistants, entry-level computer programmers, and, yes, some journalists. That means such jobs might change, and soon. But even if ChatGPT can spit out a pretty good paragraph on AI, it can’t interview AI and labor experts, nor can it find historical documents, nor can it assess the quality of studies of technological change and employment. It creates content out of what is already out there, with no authority, no understanding, no ability to correct itself, no way to identify genuinely new or interesting ideas. That implies that AI might make original journalism more valuable and investigative journalists more productive, while creating an enormous profusion of simpler content. AI might spit out listicles and summaries of public meetings, while humans will write in-depth stories. “In many ways, AI will help people use expertise better,” Autor said. “It means that we’ll specialize more.”

AI could also make a wide variety of industries more efficient, with muted effects on overall employment. Matt Wampler is a co-founder of an AI-powered small business called ClearCOGS. He’s been a “restaurant guy” his whole career, he told me. Restaurants and grocery stores, he says, tend to run on thin margins, yet still tend to waste a considerable amount of food. People order more spaghetti than burgers; buns get thrown out. “Restaurants just lag behind on technology,” he told me. “They’re all about people. It’s people serving people; it’s people managing people. And in that very human-centric world, the default way of handling problems is to hand it to a person. Phil’s going to do it.

ClearCOGS takes restaurants’ customer-order history, supply data, and labor data and uses AI-powered modeling to make their books leaner and more profitable. If people are starting to order more spaghetti than burgers, the system will prompt the chef or manager to buy more pasta and fewer rolls. “We put this in place in some of my cousin’s sandwich shops,” Wampler told me. “Simple answers to simple questions. The question they needed answered was, there’s an assistant manager on the night shift and a couple hours before close, he has to decide whether to bake another tray of bread or not. We provide that answer.” This use of ChatGPT isn’t eliminating human jobs, really; neighborhood sandwich joints aren’t hiring McKinsey consultants. But it might make food service more efficient as a whole.

Even if it doesn’t boost the economy, AI could still change the texture of our lives and alter how we spend our time, like social media did before it. Video games might become more immersive. Shops might have far better copywriting and sales visuals. Movies might look cooler. Videos in the depths of YouTube might become far weirder and more beautiful. We might also see far more formulaic content than we already do. (Much more ominously, there might be a huge amount of plausible-seeming disinformation online.)

For workers, Autor noted, the great risk is that AI technologies cause too sudden a change in what kind of labor employers want. Certain specializations might get wiped out, leaving thousands of call-center operators or marketing workers unemployed. But he stressed the benefits of having such technology in our hands. Productivity has languished for decades. Machines doing a little more work would have a big upside, after all.