Loading the Elevenlabs Text to Speech AudioNative Player…


Fears over job losses in an AI-led economy have become more real and clearer than ever in recent months. Major companies including Walmart, Meta, Amazon, Microsoft, Ford and Lufthansa have announced layoffs or reconfigured roles, and plans for new hires in areas like robotics won’t offset those.

A new brief by the Penn Wharton Budget Model brings into sharp relief both the macro outlook on productivity, and the potential micro-level impact of AI automation. Alex Arnon, PWBM’s director of policy analysis, produced the brief with research assistance from data analyst Vidisha Chowdhury. PWBM faculty director Kent Smetters directed the study; he is also Wharton professor of business economics and public policy.

According to the brief, titled “The Projected Impact of Generative AI on Future Productivity Growth,” AI will increase productivity and GDP by 1.5% by 2035, nearly 3% by 2055, and 3.7% by 2075. AI’s strongest boost to productivity growth will occur in the early 2030s (0.2 percentage points in 2032), but it will eventually fade to leave a permanent growth effect of 0.04 percentage points annually as the economy adjusts to AI.

“In about 40% of the employment in [the occupations we looked at], at least 50% of the tasks will be replaceable in the future,” Smetters said in a recent episode of Wharton’s This Week in Business podcast. (Listen to the episode.) “It’s not a small impact by any measure. This does not mean that these jobs are replaceable; it could mean that these jobs become more productive. More time and data are required to understand the full impact.”

While noting that it is premature to project the impact of AI on the federal budget, the brief estimated that it could reduce federal deficits by $400 billion over the 10-year budget window between 2026 and 2035.

The findings in the brief are based on a study of the automation potential in 784 occupations to gauge the implications for productivity growth. It builds on a task-based framework developed by MIT professor and Nobel economics laureate Daron Acemoglu, with a projected timeline for gen AI adoption based on the adoption path of comparable technologies such as the commercial web and cloud computing services.

“This does not mean that these jobs are replaceable; it could mean that these jobs become more productive.”—Kent Smetters

Key Findings

The brief analyzed the impact of the potential exposure to AI automation across both income groups and specific occupations. Among its key findings:

  • Forty percent of current labor income, or GDP, is potentially exposed to automation by generative AI, based on an analysis of occupation-level employment, wages, and exposure. The brief defined a job as exposed if at least 50% of the activities performed could be automated by gen AI.
  • Occupations with the highest exposure to AI automation are office and administrative support (75%), business and financial operations (68%), and those involving computers and mathematics (63%).
  • The highest-earning occupations are less exposed, and the lowest-earning occupations are the least exposed. Occupations at the bottom of the wage distribution are the least exposed to AI, since many of these jobs are predominantly manual labor or personal services.
  • Exposure rises with earnings until the 80th-90th percentiles, which include programmers, engineers, and other professionals. In these high-wage occupations, around half of the work could be performed by generative AI on average.
  • The exposure to gen AI automation is sharply lower for those in the highest-earning occupations, such as business executives, athletes, and medical specialists. Such exposure is also among the lowest for building and grounds cleaning and maintenance (2.6%), construction and extraction (9%), and farming, fishing, and forestry (10%).
  • For 29% of jobs, there is no potential to substitute AI for workers. For another 29%, AI could automate less than half of the activities required. Around 1% of jobs are completely exposed to automation, so AI could perform them entirely without significant human oversight. For more than a quarter of U.S. employment, AI could perform between 90% and 99% of the work required with minimal oversight, the brief noted.

Gen AI Adoption Curve and Jobs Impact

The PWBM brief predicted that the adoption timeline for AI’s productivity-enhancing tools will be similar to those of other mass-market technologies such as the PC, the internet, smartphones, and cloud computing. With most of those earlier technologies, adoption rose sharply in the first decade, with 40-50% of workers using them, but it slowed sharply in the next decade. The use of gen AI in 2024 suggested a faster adoption rate than previous technologies.

Smetters cited PWBM’s Arnon, who supervised the PWBM brief, as noting that AI’s impact across occupations will be similar to how email changed how people communicate. “But it’s not a magic bullet either. It’s not electricity, it’s not refrigeration — it’s not that transformative.”

“AI is not a magic bullet. It’s not electricity, it’s not refrigeration — it’s not that transformative.”— Kent Smetters

The study’s projections are already beginning to show up in employment patterns. Job growth has stagnated in occupations with the most AI automation potential. Jobs that AI can completely replace saw a sharp fall in employment between 2021 and 2024 (0.75%), although these account for just about 1% of the total employment. Employment growth has slowed significantly for other occupations with high exposure to AI automation, where the technology can automate 90-99% of the tasks.

In the future, gen AI tools will be used in more and more tasks exposed to AI productivity gains, alongside advancements in the technology and cost savings for employers, the brief stated. The share of economic activity exposed to AI will also grow faster than the rest of the economy, it added.

Smetters cautioned against expecting AI to fix everything for the economy. “There’s this belief among policymakers that in this new era of AI, we don’t have to be fiscally responsible because AI is going to solve everything,” he said. “That’s simply not true. We’re not even close.”

Are the Stock Markets Oversold on AI?

“AI is having a very big and salient impact [on stock values],” Smetters said. Barring the Magnificent Seven stocks (Google parent Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla), which are the biggest AI adopters, earnings expectations haven’t changed much for 493 of the S&P 500 stocks, he added, citing a note by Torsten Slok, chief economist at Apollo Global Management.

Companies may prepare for AI with varying degrees of optimism or skepticism, but they don’t have the option of ignoring it. “When you have really unusual transformative technology like [AI], you have two macro choices: You’re going to either lean into it, embrace it, and figure out how to make your customer experience better and shape it, or you can wish it away and have it happen to you,” Amazon CEO Andy Jassy said in a recent CNBC interview.