Cameron ∘ about ∘ blog ∘ skills ∘ talks ∘ links

Seminar #1: AI and Labor Market Inequality

Date: January 7, 2026 Topic: The impact of AI and automation on wage inequality Faculty: Dr. Chen (Macro), Dr. Roberts (Micro), Dr. Patel (Behavioral)

Presentation

The presenter chose to research AI, Automation, and Wage Inequality. After conducting web searches for recent academic papers, they presented the following thesis:

The emerging empirical evidence on generative AI and wage inequality reveals a critical paradox: while high-wage occupations show greater AI exposure, the real economic threat may be to younger, lower-wage workers through employment displacement rather than wage suppression. Contrary to expectations based on previous automation waves, AI's impact on inequality may operate through labor market access rather than within-job compensation, creating a two-tier labor market divided by work experience and age rather than skill levels.

Key Evidence Cited:

Q&A Session

Dr. Chen (Professor of Macroeconomics)

Initial Question:

You've documented displacement in specific occupations and entry-level hiring patterns, but where's the aggregate labor market story? If AI genuinely reduces entry-level hiring in exposed sectors, are these 22-25 year-olds reallocating to non-exposed industries, or does your data suggest actual structural unemployment—and if so, shouldn't we see that reflected in your own aggregate wage and employment figures rather than the modest changes you cite?

Presenter Response: Acknowledged that the 16% figure is a relative decline within AI-exposed occupations, not overall employment. Labor force participation for 22-25 year-olds shows only modest declines (0.3-0.5 percentage points). Hypothesized reallocation but admitted having no cross-sector flow data.

Follow-up:

If labor force participation for 22-25 year-olds has declined only 0.3-0.5 percentage points since 2023, and these workers are reallocating to non-exposed sectors, how much of what you're documenting is genuinely AI-driven disruption versus normal occupational churn?

Presenter Response: Conceded this is the "core methodological vulnerability." The strongest evidence is timing and concentration—employment decline accelerates after November 2022 (ChatGPT launch) in specific high-exposure occupations. But admitted an alternative explanation (secular decline) is "difficult to rule out with current data."

Dr. Roberts (Professor of Microeconomic Theory)

Initial Question:

If AI truly is substituting for entry-level labor in high-wage occupations while complementing experienced workers, then by basic microeconomic theory—shifting the relative marginal products—we should observe wages for experienced workers in those occupations rising faster than wages for new entrants elsewhere. Yet your own evidence shows within-occupation wage inequality declined in AI-exposed fields. What's the theoretical mechanism that reconciles selective displacement of entry-level workers with flat or compressed wage differentials?

Presenter Response: Distinguished between two data windows (OECD 2014-2018 vs. Brynjolfsson 2023-2025). Argued for "wage stickiness"—employment adjusts before wages in contractually sticky tech compensation.

Follow-up (devastating):

But here's the difficulty with the wage stickiness argument: tech labor markets are notoriously dynamic. If experienced software engineers truly became more valuable to firms using AI—complementing the technology—wouldn't we expect rapid wage adjustment precisely in competitive tech labor markets where firms fight for talent, regardless of whether entry-level hiring dropped?

Presenter Response:

"You've identified what may be a fatal flaw in my wage-stickiness defense, and I need to acknowledge it directly rather than rationalize past it."

Conceded the wage-stickiness argument fails under competitive labor market dynamics. Updated their interpretation: maybe it's "genuine labor demand destruction" rather than selective substitution.

Dr. Patel (Professor of Behavioral Economics)

Initial Question:

You've documented that 22-25 year-olds are reallocating from \(85k software roles to \)35-45k administrative positions. But have you investigated whether this reallocation pattern reflects workers' rational assessment of their best remaining options, or whether behavioral factors—loss aversion to the wage decline, anchoring on expected career trajectories, or bounded rationality in job search—are constraining them to reallocate downward rather than sideways to non-exposed high-wage sectors like nursing or skilled trades?

Presenter Response: Admitted they haven't investigated worker decision-making at all. Discussed liquidity constraints as a rational explanation but acknowledged it might be loss aversion or status-quo bias.

Follow-up:

Are these displaced software engineers actually comparing across nursing, trades, and administrative roles in their job search, or are they exhibiting status-quo bias and searching only within adjacent "professional" categories—implicitly treating a move to the trades as identity-threatening rather than financially optimal?

Presenter Response: Acknowledged this is "exactly where my analysis breaks down." Both rational liquidity constraints and identity-threatening status quo bias would produce the same observed pattern. Would need job application data to distinguish them. Proposed a research design during the Q&A itself.

Faculty Verdicts

Dr. Chen

Strongest point: The temporal clustering argument—employment declines accelerate after ChatGPT's launch in specific high-exposure occupations. That's genuinely useful for establishing signal versus noise.

Weakest point: The 0.3-0.5 percentage point labor force participation decline is almost negligible, and the presenter never resolved how a 16% relative employment drop produces such a tiny aggregate effect. The empirical gap between occupational-level claims and macroeconomic reality remains gaping.

Dr. Roberts

Strongest point: Their intellectual honesty—they directly conceded when my wage-stickiness defense collapsed under scrutiny, rather than rationalizing past the challenge.

Weakest point: The core theoretical mechanism. The original thesis (AI complements experienced workers while substituting entry-level workers) required evidence of wage premiums that simply doesn't exist. They retreated to a much more modest claim.

Dr. Patel

Strongest point: The presenter's intellectual honesty about the gap between their empirical claims and theoretical interpretation.

Weakest point: They've built an entire distributional argument around worker "reallocation" without once investigating whether that reallocation is actually constrained by behavioral factors. They assumed optimization rather than testing it.

Meta-Commentary

This seminar demonstrated several interesting dynamics:

  1. Real-time thesis revision: The presenter actually changed their position under pressure, moving from "AI creates age-based inequality through selective substitution" to "maybe it's just aggregate labor demand destruction."

  2. Faculty coordination: Later faculty referenced earlier attacks. Dr. Roberts mentioned "Dr. Chen correctly demolished the wage-stickiness defense."

  3. Genuine methodology critique: Questions weren't just hostile—they identified real issues (identification strategy, missing data, untested assumptions).

  4. Research design emerged from debate: Dr. Patel's questioning led the presenter to propose a viable empirical strategy to distinguish behavioral from rational explanations.

Raw transcript →