Advancing AI Capabilities and Evolving Labor Outcomes

summary

The gist

This study investigates the labor market consequences of advancing Artificial Intelligence (AI) capabilities by analyzing near real-time changes in employment status and work hours across occupations…

In short

Researchers tracked how increasing AI capabilities affect jobs using near real-time US labor data. They created a dynamic score for AI exposure across occupations and found that higher exposure leads to reduced employment, more unemployment, and shorter work hours. These negative effects are strongest for older workers, younger workers, men, and college graduates.

Key concepts

Dynamic Occupational AI Exposure Score (OAIES)
A score built over five stages of AI development (from basic machine learning to agentic AI) that measures how much an occupation's tasks can be performed by evolving AI technology. It dynamically updates as the technology advances.
First-Differenced Analysis
A statistical method used to compare labor market outcomes between two distinct time periods (e.g., 2022–2023 versus 2024–2025). This helps isolate the specific impact of AI exposure changes on employment, unemployment, and hours worked.
Extensive Margin vs. Intensive Margin
These terms describe two ways to measure labor market shifts. The extensive margin measures overall job status (like unemployment rates), while the intensive margin measures how much workers spend in their jobs (like work hours). The study found effects on both.

Terminology used across episodes

This episode discusses

The paper

Advancing AI Capabilities and Evolving Labor Outcomes · Read on arXiv

Institute for Ethics and the Common Good, University of Notre Dame · Keough School of Global Affairs, University of Notre Dame

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Advancing AI Capabilities and Evolving Labor Outcomes".

Warren: Here's the gist of the paper. This study investigates the labor market consequences of advancing Artificial Intelligence (AI) capabilities by analyzing near real-time changes in employment status and work hours across occupations in relation to these technological shifts.

Ruth: Researchers construct a dynamic Occupational AI Exposure Score, built upon a five-stage framework that evaluates how AI's capability to perform tasks evolves as technology progresses from traditional machine learning to agentic AI.

Warren: These scores are linked to the US Current Population Survey (CPS), enabling near real-time analysis of employment, unemployment, work hours, and full-time status.

Ruth: A first-differenced analysis was conducted comparing the period from October 2022 to March 2023 with the period from October 2024 to March 2025.

Title and authors: Ruth: Let's talk about the title and who wrote this paper, "Advancing AI Capabilities and Evolving Labor Outcomes." It sets a very clear agenda for what they’re trying to prove about the relationship between technology and work.

Warren: Yeah, the authors are Jacob Dominski and Yong Suk Lee. They seem to be coming from different angles, which is good because you need that mix of economics and quantitative modeling here.

Iris: I see them focusing on how the capabilities of AI aren't static; they’re evolving rapidly, so the study needs to track that evolution in their exposure score over time.

Ruth: Right. The core idea is tracing those changes in AI capability—from just predictive machine learning to these newer agentic systems—and linking that directly to observable labor market outcomes like job status and hours.

Warren: So it’s not just about the current state of AI; it’s about the trajectory of its development influencing how people actually find work and spend their time.

Iris: That evolution is key because if you only measure today's capability, you miss the potential impact from what's coming next in terms of reasoning or planning models.

Ruth: Exactly. They are building this framework to capture that progression, which is what makes this study different from just looking at a single snapshot of AI adoption.

Warren: So the title tells us they’re not just making a prediction; they’re trying to model the actual labor market response as it unfolds.

Iris: It positions the work as an attempt to quantify something that used to be more anecdotal, giving us measurable evidence about job displacement risks.

The paper's summary: Ruth: So, summarizing what they did, this paper investigates the labor market consequences of AI by constructing this dynamic Occupational AI Exposure Score. It’s a way to measure how much an occupation is exposed to different stages of AI advancement.

Warren: They use task-level data to build these scores using state-of-the-art models like ChatGPT and Claude, then they connect those scores to CPS data for near real-time tracking of employment and hours.

Iris: The paper summarizes the key finding that higher exposure to AI leads to reduced employment, increased unemployment, and shorter work hours across occupations.

Ruth: And they add the nuance that these negative effects are more pronounced for specific groups like older workers or younger workers under thirty or over fifty, men, and those with college education.

Warren: They also found that roles needing complex reasoning and problem solving see the biggest declines in full-time work when AI exposure rises.

Iris: Conversely, manual physical tasks seem less affected by AI exposure and some even show employment gains, which suggests a difference in how the technology is being applied to different kinds of work.

Ruth: Essentially, they’re showing that the impact isn't uniform; it depends heavily on the specific nature of the tasks within an occupation.

Warren: It’s a shift from a simple correlation to something more granular, where you can see *which* tasks are being hit hardest by AI progress.

Iris: And they're using these scores to analyze shifts along both the extensive margin—unemployment—and the intensive margin—work hours—which gives us a dual picture of the labor market change.

The paper's improvements: Ruth: Now, moving on to how this study could be improved, the authors suggest several ways to make this framework even more robust and detailed. They want to build on what they’ve already done.

Warren: I think their suggestion about using prompt engineering with models like ChatGPT 4o and Claude three point five Sonnet to assess exposure at those five stages is a really concrete way to operationalize the dynamic score construction <ref:2507.08244#pg1>.

Iris: That sounds like a strong methodological improvement because it moves the scoring from being purely theoretical to being based on actual performance assessments by cutting and testing prompts against different AI models.

Ruth: And they suggest linking these scores to task content heterogeneity modeling by incorporating indices like the non-routine cognitive analytical index and routine manual index into the econometric framework.

Warren: That’s smart because it directly addresses the question of *why* some tasks are hit harder than others, rather than just observing that they are. It helps pinpoint vulnerability.

Iris: I think their proposal to use indicators for occupations in the top quartile of different task indices to assess if labor market consequences vary with task intensity is a way to really test that hypothesis rigorously.

Conclusion: Ruth: So, to wrap up, this paper on "Advancing AI Capabilities and Evolving Labor Outcomes" confirms that increasing AI exposure is tied to lower employment and higher unemployment rates, especially for certain demographics and job types.

Warren: It’s a clear finding that we should be watching the shift toward part-time work because the intensive margin analysis shows people are working less time at their main jobs.

Iris: The paper gives us empirical evidence showing that AI's impact is measurable, even if it doesn't predict every single job loss perfectly because of the temporal delay in innovation.

Ruth: It’s a warning, though; this empirical foundation supports the need for proactive policy responses before we get clearer signs of disruption.

Warren: It really frames AI not just as a productivity tool, but as something that's actively reshaping how and where people work right now.

More episodes

← Home