AI jobs apocalypse unlikely soon, Anthropic study shows

In March, Anthropic released an analysis intended to assess whether AI would reshape employment as dramatically as earlier predictions suggested. The analysis directly challenged the company's own co-founder Dario Amodei, who claimed in May 2023 that AI could wipe out half of all entry-level jobs within one to five years. Amodei had also warned in January that AI would become a "general labor substitute for humans," and in June raised the specter of a world "stuck on the hypergrowth, hyper-inequality setting."

The March report's findings contradicted this doomsday narrative. It found "no systematic increase in unemployment for highly exposed workers since late 2022" and observed that AI deployment "remains a fraction of what's feasible." The analysis quantified this gap: Claude covers just 33% of all tasks in the computer and math category, whereas it could theoretically take over nearly 100% of them. This gap between capability and actual use suggests that real-world AI adoption is far slower than technological possibility allows.

Productivity data reinforced the cautious view. Despite surging datacenter spending, labor productivity growth in the first three years of the AI era was slower than during the information technology boom that began in the mid-1990s—a pattern economist David Autor summarized by noting that "the world is not changing as fast as they predicted." Even OpenAI's Sam Altman, AI's most prominent public figure, recalibrated expectations in May, saying "I don't think we're going to have the kind of jobs apocalypse that some of the companies in our space advocate or talk about."

This shift in narrative has opened space for more nuanced explanations of AI's labor market impact. One influential framework draws from the Challenger space shuttle disaster of 28 January 1986, when a rubber O-ring that failed in low temperatures destroyed the spacecraft. The "O-ring argument" suggests that as long as AI cannot perform every task perfectly, the remaining human tasks become more valuable. A recent study supported this logic, concluding that "despite strong substitution at the task level, overall employment effects are modest, as reduced demand in exposed occupations is offset by productivity-driven increases in labor demand at AI-adopting firms."

Yet formidable obstacles remain. Energy is a critical constraint: according to the International Energy Agency, power demand from datacenters will more than double by 2030 to about 945 terawatt-hours, exceeding Japan's entire energy consumption. Economist Daron Acemoglu warned that AI companies "are losing hundreds of billions of dollars every year," raising questions about the economics of the transition. Public opposition is also mounting—seven in 10 Americans oppose building AI datacenters in their area, driven partly by concerns about electricity costs. Fundamental technical limits persist as well: while AI excels at language, it cannot reliably connect language to external reality and still makes critical mistakes. The technology's scope, Autor noted, is inherently limited because "not everything is a computational problem."

Meanwhile, Amodei's one-to-five year window for mass job displacement still has almost four years remaining, and the Federal Reserve reports that AI adoption continues to expand rapidly across businesses. Autor observed that AI's progress shows no sign of hitting a ceiling soon, and Nobel prize-winning economist Daron Acemoglu noted that AI insiders remain "as gung ho as ever," still believing artificial general intelligence is around the corner. The history of the computer industry offers a cautionary parallel: Nobel laureate Robert Solow remarked in the early years of computing that "you can see the computer age everywhere but in the productivity statistics," yet computers eventually transformed economic output after a lag of roughly 10 years. Whether AI will follow a similar delayed-impact path, or whether structural energy and economic constraints will permanently limit its transformation, remains an open question.

The gap between AI's theoretical capability and actual deployment marks a turning point in the technology's narrative. Anthropic co-founder Dario Amodei warned last May that AI could wipe out half of entry-level jobs within one to five years, and the company's own leadership painted dire pictures of economic disruption and inequality. Yet the company's March analysis contradicts those alarms: no systematic unemployment rise has emerged since late 2022, and Claude—despite being a state-of-the-art system—covers only a fraction of feasible work. This disconnect is not merely a story of delayed impact. Productivity growth in AI's first three years has underperformed even the IT revolution of the 1990s, despite unprecedented datacenter investment. Even OpenAI's Sam Altman, the technology's most visible advocate, declared in May that he no longer expects a "jobs apocalypse."

The emerging alternative narrative rests partly on the "O-ring argument"—the idea that as long as AI cannot perform every task perfectly, it raises the value of the remaining tasks and the workers who perform them. A recent study supports this view, finding that while AI substitutes for specific tasks, overall employment effects remain modest because "reduced demand in exposed occupations is offset by productivity-driven increases in labor demand at AI-adopting firms." This framing acknowledges automation's real effects without endorsing wholesale labor displacement.

However, significant headwinds cloud AI's future. Energy demand from datacenters is projected to more than double by 2030 to about 945 terawatt-hours—more than Japan's entire energy consumption—and Acemoglu notes that AI companies "are losing hundreds of billions of dollars every year." The technology also faces fundamental limits: it excels at replicating language but cannot reliably connect language to external reality, and seven in 10 Americans oppose building datacenters locally due to electricity costs. Whether AI can eventually solve these problems or deliver value at a price society will bear remains an open question, leaving room for cautious skepticism alongside the continued optimism of industry insiders.

No comments yet. Be the first to share your thoughts!