S7 Ep19: Which jobs will AI replace?

Episode Summary

Executive Summary: The episode compares early evidence on AI’s labor-market effects from the US and Europe. One paper finds AI adoption in US commuting zones reduced overall employment growth but raised wages/employment at the top, suggesting displacement and inequality. A European paper instead finds AI exposure is associated with higher employment shares, especially for skilled workers, implying complementarity still dominates so far.

Main Topics: Competing views of AI and jobs (Priority: 5/5): The discussion contrasts the optimistic productivity effect—AI raises worker and firm productivity, supporting employment—with the pessimistic view that AI will substitute for labor and destroy jobs. Measuring AI adoption and exposure (Priority: 5/5): Gino Gancia explains how AI adoption is proxied using employment in AI-related occupations and how exogenous exposure is built from industry specialization and historical commuting-zone composition. US evidence on aggregate employment effects (Priority: 5/5): The US study estimates that places more exposed to AI experienced slower employment growth, implying AI has already had a measurable negative labor-market effect at the local level. Distributional effects and inequality (Priority: 5/5): The US results show losses concentrated among low-skilled, middle-aged, and production workers, while the top 10% of earners see positive effects, suggesting rising inequality. European evidence on task complementarities (Priority: 4/5): The Europe-based paper finds AI exposure is associated with gains in employment share and relative wages for more exposed occupations, especially where digital infrastructure, human capital, and market openness are stronger. Policy response: training, targeting, and redistribution (Priority: 4/5): Both speakers stress policy should help workers transition, retrain, and build STEM skills, while also supporting complementary AI uses and redistributing gains from winners to losers.

Key Arguments: AI can raise productivity and employment through a 'productivity effect,' but it can also displace workers through automation. AI adoption is hard to observe directly, so researchers infer it from employment growth in AI-related occupations and construct exogenous exposure measures using industry specialization. US causal estimates suggest a commuting zone with no AI adoption would have had 0.6 percentage points faster employment growth over 20 years than an average AI-adopting zone. The negative employment effect is not uniform across workers: it is strongest for low-skilled, middle-aged, and production workers, while the top 10% benefit strongly. These patterns imply AI has likely increased inequality, even if the paper does not directly estimate wage effects. The European paper focuses on composition rather than aggregate employment, asking whether AI complements some occupations and raises their employment share and wages. Across 16 European countries from 2011-2019, more AI-exposed occupations generally gained employment share, suggesting complementarity exceeds displacement so far. Countries with better digital infrastructure, more open AI environments, more human capital, and less rigid labor and product markets show stronger complementarity effects. The European findings are descriptive rather than causal, but they suggest current AI resembles skill-biased technological change more than a job-destroying shock. Both speakers argue policy should promote training, STEM education, worker transitions, and AI applications that complement labor rather than merely cut costs.

Data Points: Employment growth difference: 0.6 percentage points - US commuting zones with no AI adoption would have experienced faster employment growth by this amount over the 20-year sample period, relative to the average adoption zone. Sample period: 2000 onward - Gino Gancia’s US study uses evidence on AI-related employment going back to the year 2000, before ChatGPT. European sample period: 2011-2019 - Juan Jimeno’s paper studies AI exposure and labor-market composition over this period. Number of countries: 16 European countries - The European analysis uses data covering 16 countries. Top earners: Top 10% - In the US study, the employment effect turns strongly positive for workers in the top decile of the wage distribution. AI-related occupation proxy: Employment in AI-related occupations - Used as the empirical measure of AI adoption in the US paper.

Pivotal Quotes: "The source of the anxiety is that there will be no new jobs, new tasks for people to perform once that artificial intelligence is fully developed." — Juan Jimeno: Explaining why AI raises fears of labor displacement and job scarcity. "An hypothetical commuting zone with no adoption of AI at all would experience faster employment growth by 0.6 percentage points over the sample period of twenty years." — Gino Gancia: Summarizing the main US estimate of AI’s effect on local employment growth. "What we are seeing so far is basically the type of skill-biased technological progress that we saw during the 80s and 90s." — Juan Jimeno: Interpreting the European evidence as complementarity and skill-biased change rather than widespread displacement.

Implications: Early evidence suggests AI is already reshaping jobs unevenly: it may boost productivity and skilled work while pressuring lower-skill workers. Policy should speed retraining, favor STEM skills, and steer AI toward complementing labor rather than replacing it.