In recent months, some of the leading voices in artificial intelligence (AI) have tempered earlier predictions of widespread job displacement driven by automation. Anthropic, a prominent AI company known for its chatbot Claude, published an analysis in March assessing AI’s impact on employment, challenging the notion that intelligent machines will imminently replace human labor on a massive scale.

Dario Amodei, Anthropic’s co-founder, had previously forecast significant disruption: in May 2023, he suggested that AI could eliminate half of all entry-level positions within one to five years, and by January 2024 described AI as likely becoming “a general labour substitute for humans.” He warned in June of a potential future marked by rapid economic growth coupled with deepening inequality. Yet, Anthropic’s own March report noted that so far, AI’s deployment remains limited and has not triggered a surge in unemployment among workers exposed to automation risks. For example, Claude currently addresses only about one-third of tasks categorized under computer and mathematics professions, although theoretically it could handle nearly all of them.

Despite growing investment in data center infrastructure, productivity gains attributable to AI have not matched early expectations. Labor productivity growth over the initial three years of widespread AI availability has lagged behind earlier technological waves such as the information technology boom of the 1990s. OpenAI’s CEO, Sam Altman, who is often seen as a primary spokesperson for AI’s potential, acknowledged in May that a “jobs apocalypse” is unlikely, a position echoed by economist David Autor of MIT, who observed that world economic dynamics have not shifted as rapidly as anticipated.

This evolving understanding has sparked a reassessment of AI’s broader economic and social effects. Some experts draw on what has been dubbed the “O-ring effect”—an analogy from the 1986 Challenger space shuttle disaster caused by a small but critical component failure—to point out that AI’s inability to flawlessly perform every task may in fact increase the value of complementary human work. Depending on which tasks AI assumes, it could augment the productivity and opportunities available to both high- and low-skill workers.

Recent studies support this nuance: while automation substitutes for certain job components, overall employment impacts tend to be modest, as productivity improvements at AI-adopting firms foster greater labor demand. Nonetheless, views within the technology sector remain divided. Nobel laureate economist Daron Acemoglu notes that many insiders continue to anticipate the advent of artificial general intelligence, while figures like Elon Musk advocate for a future where AI combined with robotics can accomplish almost all work, potentially making employment optional.

History offers a cautionary parallel. Robert Solow, a Nobel-winning economist, famously remarked during the early computer era that the transformative effects of computing were “everywhere but in the productivity statistics.” Eventually, businesses adapted and benefits became measurable, suggesting that AI’s full impact may be similarly delayed.

However, public sentiment toward AI infrastructure is increasingly negative. In the United States, about 70% of Americans oppose construction of AI data centers in their communities, a resistance partly driven by concerns over local energy demands and costs, and partly by apprehensions regarding societal disruption.

Additional challenges persist, including doubts over AI’s ability to meet all economic needs and questions about the financial viability of sustaining expansive AI operations. Estimates predict AI-related data centers could consume a significant portion of gross domestic product in coming years, raising concerns about affordability. Moreover, AI companies face rapid depreciation of technology value as newer models continually emerge, resulting in substantial financial losses.

While some of the recent skepticism expressed by AI leaders may be influenced by public relations considerations, the growing recognition of economic and technical hurdles points to a more complex future for AI than previously imagined. The much-heralded transformation may still be unfolding, but whether artificial intelligence can fulfill its ambitious promises without excessive costs remains an open question.