Working Papers
Job Market Paper
How do workers decide how much to invest in AI? I survey a global sample of online platform workers and document a puzzle: adoption and reported productivity gains are high, yet investment in AI tools and learning remains low. The pattern cannot be entirely explained by income or AI exposure, and it persists among workers who report large gains. I develop a model of AI investment along two margins, the money workers spend and the time they commit, and test it in a randomized controlled trial. I compare an experiential Effective Use arm, which gives actionable guidance on better using AI, with two arms conveying labor-market information framed as disruption or opportunity. Across both margins, workers' investment intentions increase as expected returns rise, and money responds to returns more strongly than time. The Effective Use arm further raises willingness to pay for AI tools whereas information about labor-market disruption or opportunity does not produce comparable effects. No arm raises intended training on average, but this null masks heterogeneity. Intended time investment rises among workers predisposed toward AI, through experience, lack of skepticism, or a perceived stake in it. AI adoption may be widespread, but investment depends on prior work use, trust, expected returns and how workers learn about them.
The cross-border outsourcing of codifiable cognitive work to developing countries has been a resilient engine of services trade and a pathway to export-led development. This paper examines how generative artificial intelligence (AI) reshapes service outsourcing and redistributes its gains. Using granular transaction data linking clients, workers, and 2,700 skills on a global digital-services platform from 2022 to 2024, together with a novel capability-based measure of skill-level automation exposure, we trace adjustments on both sides of the market. We find that underneath the aggregate contraction of outsourcing — concentrated in developing countries — lies significant restructuring in how buyers and workers trade. Buyers transact less frequently while consolidating into fewer, larger, and more complex contracts. Workers complete fewer contracts but earn more per engagement, narrowing the skills they supply while winning work in AI-complementary and interaction-intensive tasks. Adaptation is highly uneven: experienced, higher-rated workers reallocate successfully while others contract, and within-skill concentration rises by 14 percentage points as those best positioned to adapt capture a growing share of a shrinking market. A task-based trade framework rationalizes these patterns: by compressing the cross-border wage–productivity arbitrage in routine tasks, generative AI shifts comparative advantage and redistributes the gains from digital-services trade across countries and workers. Because the technology's benefits and its adjustment costs accrue to different labor markets, the findings bear directly on whether AI narrows or widens global human-capital and income gaps.
Work in Progress