Seeing a lot of discourse around translations and accessibility and getting bricks thrown at me when I say AI tools could help. Sure the tech is in no way a replacement for a human editor but it sure might help streamline the process. Mind you, I definitely lean more towards being an AI hater than an AI lover, but god forbid we acknowledge it's use as a practical problem solving tool.
So like, on the one hand, you're absolutely right, and on the other hand, the fact that you're right is why capitalist structures have such an easy time misusing these tools abusively.
See, AI translations are good enough for really casual use, if you are translating to AND from one of the well supported languages (basically, written Chinese and English, plus or minus written Arabic and some other languages per model and source text availability).
I can translate tumblr posts from the languages my mutuals speak that I don't, and it's fine if I'm a little wrong, because getting the broad impression is what matters, it's a fucking tumblr post.
But it's NOT good enough for the kinds of expert use cases that are being discussed in the conversations you are referencing (eg those started by elbiotipo and imsobadatnames2, among others).
The problem is that because good enough for casual use AI assisted translation exists for the individual, companies get away with using that same translation software for stuff that it's really, REALLY not good enough for.
Stuff like, for example, translating medical documents for non-English speaking patients at English dominant hospitals. A thing I have personally witnessed multiple times, and even been asked to do by an employer in medical lab testing.
But also stuff like publishing a video game in multiple languages.
Now, as far as helping streamline the process?
Nope. Or, not in the way you're thinking.
Asking AI to translate something as a "first draft" fundamentally misunderstands the way fluency and translation work. The very order that they go in. The translator's first draft is the original document, and in the very act of reading it with an eye to matching concepts to the target language, they begin the process of translation.
The only way AI can be good enough is if the alternative is a complete lack of access to the original information. And ultimately, there aren't actually that many "work grade" cases where "I couldn't understand the original so I didn't read it" would be considered appropriate or acceptable translation of an original.
Look, Anglo to Anglo, I need you to understand something.
Until you are not only fluent in multiple languages, but then expected to be able to fluently transform them to each other without loss of data by subjectively correcting for differing context windows?
You don't understand how hard that work is, or the level of expertise it takes.
I mean, it's possible to learn to appreciate translation work without speaking multiple languages, but it takes a real and concerted effort because you have been lied to your entire life about how easy it is.
So. The problem is, our socio-economic system does not compensate or respect translation work appropriately to the skill involved.
And you coming in with a "solution" that further reduces compensation and respect while also not actually being anywhere near good enough for the task at hand...
It's really well intended.
But it reveals a deep misunderstanding of both translation work, and also the current state of AI to perform translation work.
Which sucks, because like. You do truly mean well, and if the scale of the problem were truly interpersonal, you'd be on the right track!
It's just that you're offering a solution to a different problem, and it won't work for this one.
Now, I know that's really disappointing to hear, so I'm going to tell you something.
If you want to use AI as an assistive tool for translation, there is a way.
Use AI as an assistive tool to learn a new language. AI isn't good enough to make you fluent, not at all. But it's good enough to let you build a basic, basic level of comprehension. Enough comprehension that you can start engaging with human made media in the target language.
This won't work for small languages, and it won't work for spoken languages, and sign languages, and a lot of other exceptions.
But like, if you want to learn to read written Chinese as fast as possible so that you can begin reading writing by actual writers in that language, AI can be a great little first step.
