WEBINAR RECORDING · 2023-03
Artificial Intelligence and Trade
In this conversation: Dan Trefler and Javier López González; moderators Ben Shepherd and Hanna Norberg.
Dan Trefler and Javier López González examine AI as both a driver and a beneficiary of trade. Mobile-app research supplies concrete evidence, while discussion moves to measurement, privacy, standards, and organizational constraints on technological change.
Session information draws on the recording and its original description. Affiliations and policy context refer to the time of the webinar.
FROM THE CONVERSATION
Three takeaways.
What they mean in practice.
Prepared from the session recording using automatic transcription and editorial review. Follow the timestamp links for the original wording and context.
Specific markets make effects measurable
Watch 18:13 ↗Trefler studies international mobile-app use, linking descriptions to firms’ AI patents through text similarity. He reports increased usage, new products, and displacement of older products associated with AI, with an instrumental-variable approach intended to address reverse causality. This is a concrete setting, not an economy-wide forecast.
Policy interpretation: Use application-specific evidence before extrapolating claims about aggregate benefits or dominance.
Trade policy determines access to AI
Watch 31:29 ↗López González maps AI inputs across hardware, digital services, human expertise, and data flows. Restrictions on any can impede diffusion. AI changes products, logistics, and productivity, while international access helps firms acquire inputs needed to use it.
Policy interpretation: Examine the whole supply chain when assessing adoption barriers, rather than domestic algorithm development alone.
Regulation requires explicit tradeoffs
Watch 50:30 ↗Trefler favors tractable cooperation on particular applications over a sweeping solution. López González asks for evidence on costs as well as benefits, including privacy, security, and distribution. Both emphasize organizational change and established production clusters as constraints on rapid transformation.
Policy interpretation: Assess benefits, legitimate regulatory objectives, and implementation constraints together; unrestricted data movement or disruption is not automatically optimal.
From the discussion
At 44:12–46:24, Shepherd asks how confidently patent-text matches identify AI actually incorporated in apps. Trefler describes plausibility checks and says formal validation would help: semantic similarity is a constructed measure, not direct observation of code. At 56:28, a comparison with unexpected 3D-printing evidence prompts caution against forecasts based on imagined capabilities rather than adoption research.
