4 ms·

I wouldn't say everyone loses. Being able to use foreign language websites and read documents is incredibly useful, and they don't need the world's best transla

I wouldn't say everyone loses. Being able to use foreign language websites and read documents is incredibly useful, and they don't need the world's best translation - although I would say the quality has been pretty decent recently. I'm sorry for translators but they're vastly outnumbered by people who find machine translation useful. If the choice is between finding a translator and paying them to be able to read something at all vs immediate free translation with the occasional mistake, I know which one I find more useful.

onion2k 4y ago
If the choice is between finding a translator and paying them to be able to read something at all vs immediate free translation with the occasional mistake, I know which one I find more useful. Yes, and that's how people will think about everything that can be generated by AI, from images to writing to music to code. People are very happy to accept a cheap-but-worse alternative. That that thought and apply it to what you do, and you'll see what the problem is.
- Jensson 4y ago
  There isn't much of a market for shitty software though, if you are worse than Google people will just use Google etc.
  - onion2k 4y ago
    There isn't much of a market for shitty software though.. Are you sure? I mean ... gestures broadly at the software market
kaashif 4y ago
It's very, very surprising to me that that person has found a way to say that "everyone loses" with machine translation. If they wanted to look at a foreign website 30 years ago, what did they do? Call up a human translator and get them to translate? No, we just didn't read foreign language websites. For me, this is on par with arguing that English to foreign language dictionaries are bad, it's just ridiculous. Sorry if this comes across as too aggressive, I just come across this backlash against technology and labour-saving devices somewhat often, and it really confuses me.
- AgentOrange1234 4y ago
  There weren’t even web sites 30 years ago, but yes to all of this.
  - joefourier 4y ago
    30 years ago was 1992 and the web was already a year old, so there were already a few websites, including some in Hebrew and French (although most were physics-related). Usenet itself is going to be 43 years old next year!
    - AgentOrange1234 4y ago
      Ah, I thought HTML was 94? I could have it wrong.
- onion2k 4y ago
  Your post isn't aggressive at all. It's fine. If you look at machine translation in isolation you might actually be right. Perhaps we don't really need accurate translation at all, and translation jobs won't be missed, and having more-but-worse translation is a benefit of AI. But my point was more about using translation as an illustration of the impact of AI on an industry. In the case of translation there are maybe 5 or 10 percent of the translation jobs today than there were 20 years ago. Machine translation has decimated the translation industry. People are happy to have bad translations in order to save money. That is a pretty indisputable fact. My point is that this is likely to be the case for everything that you can generate with AI. If people will accept a slow, buggy, shitty app generated with AI because it saves them money then there is every chance that there could be 90% fewer devs in a couple of decades than there are now. It's hard to accept as a dev. I'd wager that if you'd told a translator back in 2002 that a computer will replace them in two decades they'd have had a hard time accepting it too.
  - flappyeagle 4y ago
    The difference with software is: the AI doesn’t need to write the software. It is the software. Rather than “write me a calendar app that has a reminders feature” it’s “remind me to pick up the kids at 5:00”
RandomLensman 4y ago
If the corpus of verified, good translations comes down hard then the ability of machine translation to improve much beyond its current state or to deal with language changes will be limited. The issue with things like machine translation is how to compensate the humans that fund and create the translations going into the machine - they are not redundant.
- Ekaros 4y ago
  Also the translation really depends on the corpus. EU legislation and such is big input. But I'm not sure how well that covers many other cases. Or prose for example. Chinese/Korean->English is readable, but actually pretty horrid for artistic works. Not something you really want to read for pleasure.

onion2k 4y ago

If the choice is between finding a translator and paying them to be able to read something at all vs immediate free translation with the occasional mistake, I know which one I find more useful. Yes, and that's how people will think about everything that can be generated by AI, from images to writing to music to code. People are very happy to accept a cheap-but-worse alternative. That that thought and apply it to what you do, and you'll see what the problem is.

Jensson 4y ago
There isn't much of a market for shitty software though, if you are worse than Google people will just use Google etc.
- onion2k 4y ago
  There isn't much of a market for shitty software though.. Are you sure? I mean ... gestures broadly at the software market

kaashif 4y ago

It's very, very surprising to me that that person has found a way to say that "everyone loses" with machine translation. If they wanted to look at a foreign website 30 years ago, what did they do? Call up a human translator and get them to translate? No, we just didn't read foreign language websites. For me, this is on par with arguing that English to foreign language dictionaries are bad, it's just ridiculous. Sorry if this comes across as too aggressive, I just come across this backlash against technology and labour-saving devices somewhat often, and it really confuses me.

AgentOrange1234 4y ago
There weren’t even web sites 30 years ago, but yes to all of this.
- joefourier 4y ago
  30 years ago was 1992 and the web was already a year old, so there were already a few websites, including some in Hebrew and French (although most were physics-related). Usenet itself is going to be 43 years old next year!
  - AgentOrange1234 4y ago
    Ah, I thought HTML was 94? I could have it wrong.
onion2k 4y ago
Your post isn't aggressive at all. It's fine. If you look at machine translation in isolation you might actually be right. Perhaps we don't really need accurate translation at all, and translation jobs won't be missed, and having more-but-worse translation is a benefit of AI. But my point was more about using translation as an illustration of the impact of AI on an industry. In the case of translation there are maybe 5 or 10 percent of the translation jobs today than there were 20 years ago. Machine translation has decimated the translation industry. People are happy to have bad translations in order to save money. That is a pretty indisputable fact. My point is that this is likely to be the case for everything that you can generate with AI. If people will accept a slow, buggy, shitty app generated with AI because it saves them money then there is every chance that there could be 90% fewer devs in a couple of decades than there are now. It's hard to accept as a dev. I'd wager that if you'd told a translator back in 2002 that a computer will replace them in two decades they'd have had a hard time accepting it too.
- flappyeagle 4y ago
  The difference with software is: the AI doesn’t need to write the software. It is the software. Rather than “write me a calendar app that has a reminders feature” it’s “remind me to pick up the kids at 5:00”

RandomLensman 4y ago

If the corpus of verified, good translations comes down hard then the ability of machine translation to improve much beyond its current state or to deal with language changes will be limited. The issue with things like machine translation is how to compensate the humans that fund and create the translations going into the machine - they are not redundant.

Ekaros 4y ago
Also the translation really depends on the corpus. EU legislation and such is big input. But I'm not sure how well that covers many other cases. Or prose for example. Chinese/Korean->English is readable, but actually pretty horrid for artistic works. Not something you really want to read for pleasure.