Labor market impacts of AI: A new measure and early evidence

The rapid diffusion of artificial intelligence (AI) is reshaping the labor market in profound ways. As organizations increasingly adopt AI technologies, understanding the potential impacts on employment becomes crucial. This article introduces a new measure of AI displacement risk, termed “observed exposure,” and presents early evidence of AI’s effects on the labor market.

Key Findings

• We introduce a new measure of AI displacement risk, observed exposure, which combines theoretical large language model (LLM) capability with real-world usage data.
• AI is currently far from reaching its theoretical capability, with actual coverage remaining a fraction of what is feasible.
• Occupations with higher observed exposure are projected by the Bureau of Labor Statistics (BLS) to experience slower growth through 2034.
• Workers in the most exposed professions tend to be older, female, more educated, and higher-paid.
• No systematic increase in unemployment has been observed for highly exposed workers since late 2022, although hiring of younger workers in exposed occupations has shown signs of slowing.

Introduction

The advent of AI technologies has spurred a wave of research aimed at measuring and forecasting their impacts on labor markets. However, the track record of previous approaches suggests a need for caution. For instance, a notable attempt to measure job offshorability identified approximately a quarter of U.S. jobs as vulnerable, yet a decade later, most of those jobs continued to see healthy employment growth.

Similarly, while the government’s occupational growth forecasts have been directionally correct, they have provided little predictive value beyond simple extrapolation of past trends. The effects of significant economic disruptions on the labor market often remain ambiguous, as seen in studies examining the employment impacts of industrial robots and the ongoing debate regarding job losses attributed to the China trade shock.

This paper presents a new framework for understanding AI’s labor market impacts and tests it against early data, revealing limited evidence that AI has significantly affected employment thus far. Our aim is to establish a robust approach for measuring how AI influences employment and to revisit these analyses periodically. While this framework may not capture every potential channel through which AI could reshape the labor market, it lays the groundwork for more reliable future findings.

Counterfactuals

Causal inference is more straightforward when the effects are large and sudden. The COVID-19 pandemic, for instance, caused such stark economic disruption that sophisticated statistical approaches were often unnecessary. Unemployment surged sharply in the early weeks of the pandemic, leaving little room for alternative explanations.

In contrast, the impacts of AI may resemble the gradual changes brought about by the internet or trade with China. The effects might not be immediately apparent from aggregate unemployment data, as factors like trade policy and the business cycle could obscure trend interpretations.

A common analytical approach is to compare outcomes between workers, firms, or industries with varying levels of AI exposure to isolate the effect of AI from other confounding forces. Exposure is typically defined at the task level; for example, AI can grade homework but cannot manage a classroom, making teachers less exposed than workers whose entire job can be performed remotely.

Measuring Exposure

Our approach integrates data from three primary sources:

1. The O*NET database, which outlines tasks associated with around 800 unique occupations in the U.S.
2. Our own usage data, measured through the Anthropic Economic Index.
3. Task-level exposure estimates from Eloundou et al. (2023), which assess whether it is theoretically possible for an LLM to perform a task at least twice as quickly.

Eloundou et al. score tasks on a scale: 1 if a task can be doubled in speed by an LLM alone, 0.5 if it requires additional tools or software, and 0 otherwise. Actual usage may fall short of theoretical capability due to model limitations, legal constraints, specific software requirements, or human verification steps.

A New Measure of Occupational Exposure

Our new measure, observed exposure, quantifies the tasks that LLMs could theoretically speed up and are actually seeing automated usage in professional settings. Theoretical capability encompasses a broader range of tasks, and by tracking the narrowing gap between theoretical capability and observed exposure, we gain insights into emerging economic changes.

We assess a job’s exposure based on several factors:

• The tasks are theoretically possible with AI.
• The tasks see significant usage in the Anthropic Economic Index.
• The tasks are performed in work-related contexts.
• There is a higher share of automated use patterns or API implementation.
• The AI-impacted tasks constitute a larger share of the overall role.

Conclusion

The labor market impacts of AI are still unfolding, and our new measure of observed exposure provides a framework for understanding these changes. As AI capabilities advance and adoption spreads, the implications for various occupations will become clearer. Continuous monitoring and analysis will be essential to navigate the evolving landscape of work in an AI-driven world.

Frequently Asked Questions