The AI Policy Podcast
Episode: The Impact of AI on Labor with Harry Holzer
Host: Gregory C. Allen (CSIS)
Guest: Harry Holzer (Georgetown University, former U.S. Department of Labor Chief Economist)
Date: October 17, 2025
Overview
In this episode, Gregory C. Allen sits down with leading labor economist Harry Holzer for a comprehensive discussion about how artificial intelligence (AI) is intersecting with the world of work and labor policy. They explore historical perspectives of automation, current research on AI’s labor market impact, the challenges of predicting mass unemployment, and policy recommendations for supporting workers during the AI revolution. The conversation is grounded, data-driven, and leavened with personal anecdotes and cautious optimism.
Key Discussion Points & Insights
1. Harry Holzer’s Background and Motivations
[00:56 – 02:47]
- Holzer traces his interest in labor economics to growing up on a rural chicken farm as the child of Holocaust survivors, developing an early awareness of economic and social hardship.
- His career has focused on disadvantaged groups, especially low-wage workers and people of color, with the aim to create more economic opportunity.
“You get a sense that a lot of people have very, very limited opportunities. And if we can help them... that would be a good thing as well.”
— Harry Holzer [02:33]
2. Historical Roots of Automation and Skill Bias
[02:47 – 06:41]
- Skill-biased automation: Historically, most automation has favored higher-skilled workers—especially since the digital revolution—while replacing lower-skilled roles.
- Replacement & creation cycle: New technology often eliminates some jobs but creates categories of work that didn’t exist before (e.g., assembly line workers replacing horse-and-buggy craftsmen).
“We have a term for that in labor economics. We call it skill bias.”
— Harry Holzer [03:06]
“Some jobs disappear. Sometimes new job categories of jobs open up.”
— Harry Holzer [06:27]
3. Labor Market & Policy Crash Course
[03:40 – 09:50]
- Holzer explains how automation increases productivity, lowering consumer prices, boosting demand, and usually leading to more jobs in new industries.
- The Department of Labor’s primary federal responsibilities include regulation enforcement, labor market interventions (job retraining, unemployment insurance), and data/statistics.
“Automation makes the workplace more productive... As long as markets are competitive, that means... there's a lot more demand. And sometimes... new product categories, like smartphones that didn’t exist 25 years ago.”
— Harry Holzer [04:08]
4. Measuring AI’s Current Impact
[11:21 – 16:49]
- Two recent studies are discussed:
- Yale Budget Lab & Brookings: Finds no broad labor market disruption from AI (e.g., ChatGPT) after nearly three years.
- Stanford/Eric Brynjolfsson: In narrow job categories, some impact exists (e.g., in customer service and software), but evidence is early and not conclusive.
- High-profile CEO statements (e.g., Amazon) predict reductions in workforce due to efficiency gains from AI.
- Holzer cautions that it is still early days; much depends on how AI is deployed—whether to complement or replace workers—and whether companies invest in retraining.
“Employers and the AI developers are going to have a lot of discretion in terms of how they mold the product... It can be done in a more worker friendly fashion... or in a more substitution... automation as opposed to augmentation fashion.”
— Harry Holzer [15:09]
5. Friction, Culture, and the Pace of Adoption
[16:49 – 23:34]
- Adoption of new technology brings friction: workplace culture, business practices, and the need for process reengineering slow down impact, no matter how advanced the technology.
- Classic “Solow paradox” discussed—why productivity gains from IT were not immediately seen in economic data.
“It took a decade to work out... for employers to figure out how to best use these machines, which workers to let go, which workers to retrain...”
— Harry Holzer [21:31]
6. Predictions: Will AI Cause Mass Unemployment?
[23:34 – 33:53]
- Contrasting predictions highlighted:
- Dario Amodei (Anthropic CEO): AI could eliminate half of entry-level white-collar jobs within five years, causing unemployment to spike up to 20%.
- Jensen Huang (Nvidia CEO): AI will change jobs but not eliminate humans.
- Holzer is skeptical of apocalypse narratives, drawing on history:
- Technical capability ≠ immediate, perfect replacement.
- Past disruptions created new roles; much depends on the ability to reskill and adapt.
- A special concern with AI: Its speed and breadth could put more workers on a “treadmill”—constantly needing to upskill as AI advances.
“We've heard the song before... The worry with AI is that every year so you can make an adjustment, learn a new task, and then AI might take that over a year or two away.”
— Harry Holzer [27:10]
7. Automation vs. Augmentation; Elasticity & New Industry Creation
[33:53 – 40:13]
- Distinguishing augmentation (AI as a tool) from automation (AI as a replacement) is complex but both can impact jobs through productivity increases.
- Economic impact depends not on substitution per se, but on demand in the end market: If productivity increases dramatically without enough demand growth, jobs decline.
- General-purpose technologies (like AI) will affect more sectors but might spread disruption more broadly rather than localizing it (“Rust Belt” effect).
“Employers are going to have to make judgments, okay, who is, who does it make sense to reinvest in and who not.”
— Harry Holzer [36:40]
8. Thought Experiments: Superintelligence and Labor
[40:13 – 52:13]
- Historical context: Transition from agricultural (41% of U.S. workers in 1900) to 4% by 1970 was offset by new opportunities in manufacturing and services.
- Horses, by contrast, were simply rendered obsolete.
- If AI surpasses human abilities in all domains, would humans remain “employable”? Holzer is skeptical that machines will be universally superior, pointing to human judgement, emotional skills, and arbitrary preferences (e.g., watching humans play chess or sports).
- Some demand for humanness is arbitrary but meaningful and enduring.
“I just gotta believe that at the end of the day there is something about humanness that will distinguish it and continue to distinguish it.”
— Harry Holzer [48:20]
9. Policy & Data: Preparing for AI’s Labor Impact
[52:13 – 56:08]
- Holzer advocates for better, real-time data collection to truly understand AI’s impact on the workforce.
- Existing BLS/Department of Labor data insufficiently granular; both administrative and private-sector data partnerships needed.
- Surveys need to ask direct questions about tech/AI use on the job.
“We are concerned the government is not adequately prepared for the collection of high quality economic data that will inform policy to address the workforce issues AI creates.”
— [52:35 – referencing economists’ open letter]
10. Policy Recommendations for Worker Support
[56:08 – 65:59]
- Three buckets:
- Basic K-12/higher education as foundation
- On-the-job retraining and upskilling (need for employer buy-in)
- Support for displaced workers (community college, lifelong learning, adjustment assistance)
- Recommendations: Subsidies for retraining, better guidance for curricula at state level, partnerships with businesses, expansion of internships/apprenticeships as entry-level jobs shrink.
- Track record of retraining programs is mixed: Historically, disappointing, but models like Trade Adjustment Assistance (in its later forms) and community college partnerships show positive results.
“Maybe you could subsidize retraining, you could even tax worker displacement... There’s all kinds of ways in which government can do each of those things.”
— Harry Holzer [58:28]
Notable Quotes & Memorable Moments
-
On the fragility of "mass unemployment" narratives:
“I remain more optimistic than Dario, but... we should [not] discount the possibility of a lot of disruption, a lot of displacement, and workers might need a lot of assistance to adapt.”
— Harry Holzer [28:25] -
On why not everything automatable gets automated instantly:
“Adoption of new technology brings friction: workplace culture, business practices, and the need for process reengineering slow down impact...”
— Gregory Allen [17:29] -
On the arbitrary value of human work:
“There is only a market for watching humans play chess. There is essentially no market for watching AI play. Now that is an arbitrary preference, but it is clearly a real one...”
— Gregory Allen [50:10] -
On hope for effective workforce policy:
“There have been some [programs]... where you have to be certified that you lost your job to imports to get... living money over and above unemployment insurance, plus extra training money. Now, for a long time it looked like that program was a failure, but eventually it adapted... the last bit of research... said it was actually a much more successful program than it had been previously.”
— Harry Holzer [64:06]
Timestamps for Important Segments
| Time | Segment/Topic | |------------|--------------------------------------------------| | 00:56 | Holzer’s personal background and perspectives | | 03:06 | Skill-biased automation in historical context | | 04:08 | Why automation hasn’t killed all jobs | | 07:12 | Department of Labor’s core functions | | 11:27 | Evaluating AI’s impact using occupational mix | | 15:09 | Early signs: Is AI substituting or complementing? | | 20:59 | Solow Paradox: Productivity and technology lag | | 23:34 | Mass unemployment: Is this time different? | | 33:53 | Automation vs. augmentation and sector impacts | | 40:13 | Horses vs. humans: Superintelligence thought exp. | | 52:13 | The need for better labor market data on AI | | 56:08 | Policy recommendations for labor and education | | 62:34 | Retraining programs: Track record and prospects |
Conclusion
This episode provides a nuanced, historically informed, and policy-conscious exploration of AI’s impact on labor. Holzer is cautious, not alarmist—recognizing genuine risks but identifying historical and economic mechanisms likely to dampen dystopian outcomes. Concrete policy steps—especially around education, retraining, and data collection—can help ensure that the workforce is resilient as new AI-driven disruptions unfold.
Find more from Harry Holzer at his Georgetown University profile and associated Brookings papers.
