Anthropic Says AI Could Make America Much Richer, Forecasting 33% GDP Growth by 2030. The Catch Is That Millions of Knowledge Workers Lose Their Jobs

AI could supercharge growth while pushing income from white-collar workers toward capital.

For millions of employees, the unsettling future sketched by Anthropic is not one in which artificial intelligence suddenly wipes out every job.

It is stranger than that.

In one of the economic scenarios modeled by the AI company famous for its Claude chatbot, the economy keeps growing. Companies become dramatically more productive. Construction workers, electricians and other people doing work that AI cannot easily replace may earn far more. The country, on paper, becomes astonishingly wealthy.

But many programmers, accountants, lawyers, managers, scientists and office workers watch the value of their own labor erode. Some are forced to take pay cuts. Others lose their jobs and struggle to enter entirely different professions with all the challenges that come with starting a new career from the bottom. Many may have to do this past age 40 and with kids.

And an ever-larger share of the prosperity flows not through paychecks, but to the owners of companies, compute and other capital.

The projection was made public in a new working paper from the Anthropic Institute, led by economists Anton Korinek, Charles I. Jones, Szymon Sacher, Tess Cotter and Peter McCrory. The researchers model three possible versions of the U.S. economy in 2030, ranging from an AI transition that barely registers in national statistics to one that produces economic change faster than anything the country has experienced before.

Anthropic stresses that these are “scenarios, not forecasts.” The model shows how sharply the outcome changes depending on how capable AI becomes, how quickly companies adopt it, and whether it mostly assists workers or replaces them.

In the more aggressive scenarios, AI creates enormous amounts of wealth while weakening the position of many knowledge workers.

×

Get smarter every day...

Stay ahead with ZME Science and subscribe.

Daily Newsletter
The science you need to know, every weekday.

Weekly Newsletter
A week in science, all in one place. Sends every Sunday.
No spam, ever. Unsubscribe anytime. Review our Privacy Policy.

Thank you! One more thing...

Please check your inbox and confirm your subscription.

AI assistant or AI replacement?

Anthropic starts from the idea that a job is really a collection of smaller jobs.

Consider a nurse who may draw blood, talk to patients, write discharge instructions, update records and organize care. AI might help with the paperwork while leaving the bedside work untouched. It might also create new responsibilities, such as checking whether an AI-generated treatment plan makes sense.

The same applies to lawyers, programmers and accountants. Some tasks can be sped up. Some can be handed to AI entirely. Others remain stubbornly human.

In Anthropic’s modest scenario, AI remains mostly another useful technology. By 2030, GDP is only 1.6 percent higher than it would have been without AI. Consequently, unemployment barely changes.

The substantial scenario is where human workers should start worrying. In this scenario, AI is used on roughly 12 percent of all task instances across the economy, and three-quarters of that AI use replaces human effort rather than merely assisting it. GDP ends up 8.3 percent above the no-AI path, driven by the major productivity gains delivered by the technology, while economic growth reaches 5.4 percent a year. Knowledge-work employment takes the biggest hit and falls about 4 percent.

Then comes the AI ‘optimistic’ scenario.

The ‘Optimistic’ Scenario

In Anthropic’s extreme case, AI is used on 30 percent of all task instances — roughly equivalent to half the work performed by knowledge workers today. On those tasks, it more than doubles productivity. Ninety percent of AI-performed work is automated outright.

This is also where something unprecedented happens. Whenever a new disruptive technology appears in human history, some professions suffer, but there are always new ones created to absorb the job market. Think of horse breeders who lost to automobile plant workers or handloom spinners replaced by machine loop operators. Now, when old human tasks disappear, there’s essentially no new knowledge-work tasks emerging to replace them.

Under those conditions, the U.S. economy takes off like a rocket.

GDP rises 32.4 percent above where it would otherwise have been. Annual growth reaches a staggering 15.4 percent, a pace the United States has not seen since the wartime mobilization of the early 1940s.

But cognitive employment falls 21.5 percent. Unemployment among workers who started in knowledge occupations reaches 17.9 percent, while unemployment across the entire economy hits 11.9 percent.

These are unemployment numbers typical of a major recession, but in this scenario the overall economy grows tremendously.

The closest American parallel came during the recovery from the Great Depression, when GDP grew at double-digit rates while millions remained unemployed. But even that comparison breaks down: the 1930s economy was recovering from a collapse and unemployment was falling. Anthropic imagines something stranger — an economy creating wealth at extraordinary speed while simultaneously deciding it needs far fewer human workers.

These are truly unprecedented times.

A booming economy can still leave workers behind

The most startling part of the model is what happens to the money, although, on second thought, this shouldn’t surprise anyone.

Anthropic’s model assumes that, before the AI shock, workers receive 60 cents of every dollar of economic output as compensation, with the remaining 40 cents attributed to capital — the owners of companies, machinery, software, computers and other productive assets. This is a simplified benchmark rather than a measurement of the U.S. economy today.

RelatedPosts

AI Ping Pong Robot Beats Elite Human Table Tennis Players
Endowing AI with confidence and doubt will make it more useful, paper argues
Chat GPT’s new O1 model escaped its environment to complete “impossible” hacking task — should we be concerned?
AI enables mind-controlled handwriting in paralyzed person

In the extreme scenario, those shares flip.

Labor receives just 45.2 percent. Capital takes 54.8 percent.

Average wages still rise, but that number hides a lot of disruption. Workers outside AI-exposed occupations see pay climb 33.6 percent relative to the no-AI economy. That’s good news for plumbers, electricians, construction workers and others whose jobs still depend heavily on physical skill and being there in person.

Meanwhile, knowledge workers see their wages fall 11.5 percent. That is bad news for programmers, accountants, lawyers, analysts and many office workers whose most valuable tasks are exactly the kind AI can increasingly perform.

In other words, America becomes much richer while a large group of people who spent years building valuable professional skills discover that the economy suddenly values those skills less.

This month, the Financial Times reported that UBS will require future junior bankers to demonstrate AI proficiency. This is a good hint of what’s to come as work once handed to young analysts — research, financial analysis, presentations — is increasingly being pushed toward AI. At some point, the entry-level knowledge work will be the first to go, except for a few positions required to train the next generation of seniors.

Another Financial Times report described AI as causing “mass toe-treading” inside companies, as employees use AI to perform tasks that once belonged to other specialists. A marketer can produce a rough design, an analyst can write code, and a manager can draft a presentation without turning to a designer, programmer or junior colleague. That can make teams faster and leaner — but it can also reduce demand for some specialist roles altogether.

And Reuters recently detailed how Meta experimented with an aggressive plan to replace large portions of its workforce with AI agents. The plan ran into reliability problems and employee resistance and was ultimately scaled back.

The nightmare outcome is not inevitable

The actual U.S. labor market still looks nothing like Anthropic’s extreme world.

Employers added 162,000 jobs in August and unemployment remained at 4.1 percent, according to the Associated Press’s latest report on U.S. hiring.

Anthropic’s bleakest results require several bleak assumptions to arrive together: AI becomes extremely capable, companies adopt it quickly, 90 percent of AI use replaces knowledge workers, new human tasks fail to appear and displaced professionals have unusual difficulty finding work elsewhere.

Suppose employers are free to cut salaries quickly as AI makes some knowledge work less valuable. Companies then have less reason to fire people because they can keep them at much lower wages. In that version of Anthropic’s extreme scenario, unemployment among knowledge workers stays relatively low, at 2.6 percent.

But there is a brutal trade-off. Their wages fall 42.2 percent below what they would have earned in a world without AI.

If wages do not fall so easily, the pain shifts elsewhere. Employers cannot make workers dramatically cheaper, so they employ fewer of them. Under the model’s stickiest-wage assumption, knowledge-worker unemployment rises to 24 percent.

That may be the most important message. The future of work will not be decided simply by how intelligent AI becomes. It will depend on whether thinking machines become colleagues or replacements, whether society invents new work as quickly as old work disappears, whether displaced professionals can realistically start new careers, and whether the wealth created by AI reaches ordinary households or accumulates mainly with those who own the technology.

If AI pushes the economy in that direction, the biggest argument of the next decade may not be about what the technology can do. Its usefulness will be proven without a doubt. It may be about who gets paid the productivity gains when it does.