How long will humans be needed for translation?
I keep coming back to the word "stakes." A mistranslated menu is a funny story. A mistranslated contract clause can cost millions, and a mistranslated medical instruction can cost a life. That's where the human edge still lives, and it's not really about language ability anymore. Machines have the vocabulary, the grammar, the speed. What they lack is the willingness to say "I'm not sure this is right" and the accountability that comes with being a person.
The interesting shift is that translation is no longer a test of linguistic skill. It's a test of judgment and nerve. When a diplomat's offhand remark could start a war, you want a human in the loop who understands the subtext, the cultural landmines, the silences between words. AI can translate what was said. It struggles with what wasn't said but meant.
I'd want to know how much of the high-stakes market is actually moving to AI anyway, despite the risks. My guess is that cost pressure will keep pushing, and we'll end up with a hybrid where humans check the machine's work rather than doing the work themselves. That's probably the honest future: not humans replaced, but humans promoted to the role of editor and referee. The question then becomes who watches the referees.