Aggregate unemployment with no safety net

This is the economic risk where skepticism is best supported, which is why it is worth stating precisely: it is not that there is unemployment from AI today; it is that if the effect jumps from specific occupations to the aggregate, there is no distribution mechanism built to receive it.

The inputs exist. The IMF estimates that around 60% of jobs in advanced economies and 40% in emerging ones are highly exposed to generative AI ^[550]Gen-AI: Artificial Intelligence and the Future of WorkCazzaniga, Mauro; Jaumotte, Florence; Li, Longji et al. · 2024 · reportView source ↗archived copy onlyAccessed on 9 September 2026. The measured productivity effects are large: 20-60% in controlled studies and 15-30% in most experiments in real work settings ^[538]International AI Safety Report 2026Bengio, Yoshua · 2026 · reportView source ↗Accessed on 9 September 2026. And there is a distributional warning that is rarely cited: the impact on growth in advanced economies could be more than double that in low-income countries, and AI could reduce the incentive to offshore services, closing off the development route used by the countries that industrialised by exporting them ^[538]International AI Safety Report 2026Bengio, Yoshua · 2026 · reportView source ↗Accessed on 9 September 2026.

What this does not demonstrate. The link that decides everything is not visible today. Yale’s Budget Lab, with data to August 2026, holds that the occupational mix is still not changing in ways that align clearly with the introduction of AI and that usage measures show no connection with changes in employment or unemployment ^[169]Tracking the Impact of AI on the Labor MarketThe Budget Lab at Yale · 2026 · reportView source ↗Accessed on 9 September 2026. Humlum and Vestergaard, crossing adoption surveys with Danish administrative registers, estimate precise nulls that rule out effects larger than 2% two years after ChatGPT ^[535]Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AIHumlum, Anders; Vestergaard, Emilie · 2026 · preprintView source ↗Accessed on 9 September 2026. Ryan Nunn adds a fact that neither side tends to mention: before the pandemic, employment in AI-exposed occupations was considerably less cyclical than in non-exposed ones —it should have weathered the cooling better, not worse— and the main reason for weak payroll growth is the fall in net immigration ^[804]AI Is Probably Not (Yet) the Reason for Labor Market WeakeningNunn, Ryan · 2026 · reportView source ↗Accessed on 9 September 2026.