AI Could Make The Economy Far Richer. Anthropic Says. But Knowledge Workers Could Take a Hit.

Artificial intelligence could sharply accelerate U.S. economic growth while at the same time leaving many knowledge workers with lower wages or without jobs if the technology becomes capable enough to automate most cognitive work, a new economic model shows.

Researchers at the Anthropic Institute modeled three scenarios for the U.S. economy through 2030, ranging from relatively modest AI adoption to an extreme case in which AI becomes more productive than humans at most knowledge-work tasks. The researchers stressed that the scenarios are not predictions and assigned no probabilities to them.

The three scenarios produce sharply different outcomes for the U.S. economy. Under the modest scenario, gross domestic product in 2030 would be 1.6% higher than in an economy without AI, reaching $34.1 trillion in 2025 dollars. The substantial scenario puts GDP 8.3% higher at $36.3 trillion, while the extreme scenario lifts it 32.4% to $44.4 trillion.

In the most aggressive scenario, AI can perform nearly all knowledge-work tasks autonomously, adoption is rapid and essentially no new knowledge-work tasks are created for humans. Under those assumptions, annual GDP growth reaches 15%, a pace that would double the size of the economy roughly every 4.5 years.

In the extreme scenario, wages for knowledge workers fall more than 10% below where they would have been without AI by 2030, while pay rises substantially for workers in occupations less directly exposed to the technology.

The extreme scenario also shows significant job losses in cognitive occupations as companies automate more tasks. The underlying working paper, written by Anton Korinek, Charles I. Jones, Szymon Sacher, Tess Cotter and Peter McCrory, estimates that nearly one in five cognitive workers would be unemployed by 2030 under those assumptions.

Cognitive occupations in the model include management, professional, sales and office jobs. Workers displaced from those fields would need to move into occupations less directly exposed to AI, such as construction or electrical work, with the model accounting for the time and difficulty involved in making such a switch.

Under the substantial scenario, the labor-market effects are less severe. AI becomes capable of doing half of knowledge work by 2030, but most of those tasks continue to be performed without the technology because adoption remains incomplete. Economic growth runs at about twice its normal rate, while wages for knowledge workers are essentially flat and workers in other occupations see gains.

The model also shows labor receiving a smaller share of economic income as AI becomes more capable. Workers receive 59.4% of income in the modest scenario, compared with 56.1% in the substantial scenario and 45.2% in the extreme case. Capital's share reaches 54.8% under the most aggressive assumptions.

Average wages still rise across all three scenarios because the gains are concentrated among workers outside knowledge-intensive occupations. In the extreme case, total labor income is little changed by 2030 even as the overall economy becomes substantially larger.

Anthropic also surveyed 10,980 U.S. adults in August about their expectations for AI capabilities, adoption, productivity, automation and how long displaced workers would need to find new jobs. The median responses produced results close to the substantial scenario, with GDP about 10% higher by 2030 than it would be without AI and overall unemployment at around 5%. About 10% of respondents gave answers consistent with the assumptions behind the extreme scenario.

The model uses the U.S. Department of Labor's O*NET occupational database to build occupations from individual tasks, allowing researchers to distinguish between work augmented by AI, tasks that are automated and new work created as technology changes jobs.

Anthropic's earlier labor-market research has found little evidence so far of widespread AI-driven unemployment. A March study found no systematic increase in unemployment among workers in occupations with high observed AI exposure, although researchers found signs that hiring of younger workers had slowed in some exposed fields.

A separate survey of 81,000 Claude users published in April found that concerns about losing jobs to AI were higher among workers in occupations with greater exposure, early-career workers and people reporting the largest productivity improvements from using AI.

Several potential economic effects are outside the scope of the new model, including financial-market disruptions, political responses, business-cycle effects and catastrophic risks. The researchers also treat workers in broad occupational groups rather than modeling differences based on skills, tenure or geography.

Highly capable robots that could automate large amounts of physical work are also excluded from the scenarios, which focus primarily on the effects of increasingly capable AI systems on cognitive work through 2030.