Advancing AI Capabilities and Evolving Labor Outcomes
Listen
Radio episode about this paper
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.
Institute for Ethics and the Common Good, University of Notre Dame · Keough School of Global Affairs, University of Notre Dame
econ.GN, q-fin.EC
Submitted: 2025-07-11
Updated: 2026-11-10
Comments: Added 2 figures
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 96/100
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…
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
Summary
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. p. 0 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. 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. p. 0 A first-differenced analysis was conducted comparing the period from October 2022 to March 2023 with the period from October 2024 to March 2025. p. 0
The findings indicate that higher exposure to AI is associated with reduced employment, increased unemployment rates, and shorter work hours. p. 0 These negative associations are more pronounced among older and younger workers (those below 30 and over 50), men, and college-educated individuals. p. 0 Occupations heavily reliant on complex reasoning and problem-solving experience larger declines in full-time work and overall employment as AI exposure rises. p. 0 Conversely, occupations involving manual physical tasks appear less affected, with some showing employment gains. p. 0 Overall, the results suggest that AI-driven labor shifts are occurring along both the extensive margin (unemployment) and the intensive margin (work hours), with effects varying based on occupational task content, demographic groups, regions, and industries. p. 0 Specifically, a 10-point increase in exposure is associated with a 5. p. 06 percentage point decline in total occupational employment when using ChatGPT-generated scores and an 8. p. 45 percentage point decline when using Claude-generated scores, alongside reductions in hours worked at the main job and a decrease in the share of full-time workers. p. 4
The paper contributes to the literature by providing real-time labor market analysis utilizing CPS data, developing dynamic occupational AI exposure scores that evolve with model capabilities, examining dynamic and heterogeneous effects across task content and demographics, expanding labor outcome scope to include work hours and part-time employment, and analyzing wage impacts. p. 8 The main results demonstrate that AI exposure is linked to a reduction in occupational employment, higher unemployment rates, and a shift toward part-time work. p. 2 This empirical foundation offers systematic academic evidence supporting these observations regarding the immediate impact of AI on the labor market, underscoring the need for proactive policy responses rather than waiting for clearer disruption signs. p. 2
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
- Dynamic Occupational AI Exposure Score Construction
The AI system can construct dynamic Occupational AI Exposure Scores (OAIES) by using prompt engineering with state-of-the-art models like ChatGPT 4o and Claude 3.5 Sonnet to assess what percentage of each task an occupation can perform at five distinct stages of AI development, from pre-LLM technology to agentic AI.
- Real-Time Labor Market Outcome Tracking
The system can track near real-time changes in employment status, work hours (main, other, and total), full-time status across occupations by linking the dynamic OAIES scores to monthly Current Population Survey (CPS) microdata using a first-differenced analysis between defined periods.
- Task Content Heterogeneity Modeling
The system can estimate how labor market consequences vary by task content by incorporating indices like the non-routine cognitive analytical index, routine manual index, and non-routine manual physical index into the econometric framework to document which tasks are most vulnerable or resilient to AI exposure.
- Demographic and Exposure Group Analysis
The system can assess heterogeneity by demographic groups (age, gender, education level) by estimating interaction terms in the regression model to determine how the relationship between increased AI exposure and labor market outcomes differs across these groups.
- Task Intensity Mediation Analysis
The system can estimate models that include indicators for occupations in the top quartile of different task indices to assess whether labor market consequences of rising exposure vary systematically with the intensity of specific types of tasks, revealing which occupations are most vulnerable or resilient.
Sources
- Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversations
- Embodied AI with Foundation Models for Mobile Service Robots: A Systematic Review
Related papers
- Huge If True: Learning from the Size of the Surprise
- The Rebundling of Freelance Work in the AI Age
- Dynamic investment and coordination in teams: Theory and experimental evidence
- Does Frontier AI Strengthen Cyber Defense? Evidence from Project Glasswing
- Generative AI at Work
- What is Work For? Labor, Politics, and Meaning in the Age of AI