Insights
AI Translation and the Immigrant Wage Gap
On a construction site outside Seoul, a safety briefing is delivered in Korean. Within seconds, workers from Vietnam, Uzbekistan, Nepal, and Cambodia hear it in their own languages.
This isn’t science fiction. Two of South Korea’s largest builders — Daewoo E&C and Lotte Engineering — deployed AI translation platforms across dozens of construction sites this year. Daewoo’s system supports 180 languages with a construction-specific glossary. Technical terms. Safety protocols. Equipment names. Calibrated for an environment where a mistranslation isn’t an inconvenience — it’s a catastrophe.
But this story is bigger than construction safety
Across OECD countries, immigrants earn 18% less than native-born workers. Three-quarters of that gap isn’t about skills — it’s about access to better-paying jobs, industries, and firms. And the single strongest predictor of that access is language proficiency. Research across Europe consistently finds a 23–27% wage penalty for immigrants who haven’t mastered the host country’s language. A Federal Reserve study calculated that barriers preventing immigrants from reaching their productive potential cost the U.S. economy the equivalent of 25% of immigrants’ total economic contribution.
A translation problem
The OECD found that immigrant earnings gaps shrink by a third in five years and by half in ten. That timeline assumes the immigrant does all the adapting. Years of immersion. Formal classes. Cultural code-switching. The entire burden falls on the person with the potential — not on the institution that needs it.
AI inverts this
Instead of asking “How fast can you learn our language?” it asks “Why should language determine who gets to contribute?” A surgeon who speaks Dari. An engineer who speaks Tagalog. A welder who speaks Uzbek. Their competence never changed. The interface did.
We’ve spent decades measuring immigrants by their fluency. Maybe we should have been measuring our systems by their flexibility.