What is the meaning of language, in an AI era?


I’ve been thinking a lot lately about a slightly strange question:

If AI can translate everything for us, why do we still need to learn other languages?

It’s especially strange for me to think about because I’m building a company whose entire premise is that kids should learn another language. And meanwhile, translation is getting absurdly good. We’re probably not that far from a world where you can put in an earbud, hear someone speaking Mandarin, and understand them in English almost instantly.

You could look at that future and conclude that learning another language is becoming less useful.

I think the opposite is happening.

I grew up in a Chinese-American family where my parents spoke a mix of Chinese and English. I also went to Chinese school every weekend. On paper, I had a lot of Chinese around me.

But I couldn’t really speak it.

I remember meeting Chinese relatives or older Chinese people and realizing that I didn’t know what to say. Sometimes I’d resort to hand gestures. There was this weird gap where the language was technically part of my life, but I couldn’t actually use it to know people.

I could have used Google Translate. Today, I could probably have a much better translation tool open on my phone.

But that’s not really the point.

There’s a difference between understanding what someone said and speaking their language.

I don’t want to talk to my grandparents through an AI translator. I don’t want a machine sitting in the middle of every joke, story, or tiny interaction and handing the meaning back to me in English. And I don’t think knowing the English translation of a Chinese sentence gives me the same relationship to Chinese culture as actually understanding the sentence.

The more capable AI becomes, the more obvious that distinction feels to me.

AI is making information incredibly cheap. Translation, writing, summarization, tutoring — things that used to require another person or a lot of time can increasingly happen instantly.

But information was never the only reason language mattered.

Language is also how you belong somewhere.

It’s how you talk to your grandparents without thinking about it. How you understand why something is funny before somebody explains the joke. How you make a friend in another country. How you hear a story the way it was originally told. It’s the difference between being able to access a culture and feeling, in some small way, like it’s yours.

And I think that’s where language becomes leverage.

Not because every bilingual kid is going to get a better job someday. Maybe they will, maybe they won’t. That’s actually one of the least interesting arguments to me.

It’s leverage because knowing another language gives you access to people, relationships, and ways of seeing the world that you otherwise experience secondhand.

AI can make the secondhand version extraordinarily good. But it’s still secondhand.

This has also changed how I think about language learning itself.

For a long time, we’ve treated learning a language strangely. We give kids vocabulary lists. We make them memorize characters. We give them flashcards and quizzes and apps where they tap the correct picture.

I did plenty of that.

And then somebody actually spoke Chinese to me and I froze.

The missing piece wasn’t more information about Chinese. It was having hundreds and hundreds of normal, imperfect conversations in Chinese until speaking stopped feeling like a test.

That’s one of the ideas behind Phoenyx. We’re not trying to give a child every piece of Spanish or Mandarin content they could possibly consume. There is already effectively infinite content.

We’re trying to give them more reps.

Tell Phoenyx what you drew today. Argue about whether a dinosaur could beat a shark. Ask what he ate for breakfast. Get a word wrong. Try it again tomorrow. Slowly get to the point where answering in another language doesn’t feel like translating something in your head first.

I think AI makes that kind of practice possible at a scale that wasn’t realistic before. A tutor can be amazing, but most three-year-olds aren’t going to have a Mandarin tutor following them around the house whenever they feel like talking. Software can now be there for those tiny ten-minute interactions every day.

That’s the part of AI and language learning that excites me.

Not using AI so that children never have to learn another language.

Using AI to help them actually speak one.

Maybe in twenty years, none of us will need to know another language to order food, navigate another country, or understand what somebody is saying. AI will probably handle those things just fine.

But I don’t think that’s the same as making language less valuable.

If anything, when everyone has access to perfect translation, the meaningful distinction might become whether you can understand someone without it.

And whether you can speak to them as yourself.

-Kimberly Wang, Co-founder of My Phoenyx