ChatGPT cut demand for both analytical and intuitive translation tasks, unlike GNMT
Hakan_Ozalp · x · 2026-11-04
MishaTeplitskiy shares a Management Science paper (Oct 2026), "AI-Driven Human Task Substitution: Evidence from Google Translate and ChatGPT" by Yilmaz, Naumovska & Aggarwal.
Using two sequential natural experiments in the translation industry and difference-in-differences on transaction-level data (2016–2017 and 2022–2023):
- GNMT (predictive, task-specific AI) reduced demand for analytical tasks (regular translation) by economically meaningful amounts, with no detectable effect on intuitive tasks (transcreation).
- ChatGPT (broad, generative AI) produced substantially larger and more uniform reductions across both task types — a marked expansion in the scope and magnitude of AI substitution.
- Key takeaway: conclusions about which human work is resilient to automation do not generalize from predictive to generative AI.
More from AGI Musings
- New paper: AI-flagged papers acknowledge 21% fewer informal collaborators — soumitrashukla9 · 2026-11-04
- AI Data Center Fossil-Fuel Buildout Set to Erase 15 Years of US Emissions Cuts — KyeGomezB · 2026-11-04
- "Automate the process, not the personality": dev warns against AI-driven sameness — jh3yy · 2026-11-04
- Memory without reasoning is just a filing cabinet: the case for an agent cognitive layer — jackalop211 · 2026-11-04
- "We automated the entry level" cartoon nails the broken AI career ladder — js_craft_hq · 2026-11-04
- Luis Garicano: AI economy's equilibrium hinges on whether compute, agent autonomy, or problems stay scarce — JMateosGarcia · 2026-11-04