It is the kind of dread that seeps. It’s not apocalyptic. Just the sort of thing that can eat up your livelihood.
An example. I am aware that my own work, both translations and non-translated writings, was used in the creation of Meta’s AI, without my permission. Like many colleagues, I searched and found the names of my own publications (in my case more than a dozen) on a database created by The Atlantic called LibGen. It’s free and available to anyone with Internet access. (Just search for key words “Atlantic,” “LibGen,” “pirated books”), and you’ll find it. You can also find a good article on the topic by Alex Reisner at The Atlantic. (OK, here’s the d-base, scroll down a bit, then enter your search terms. Reisner’s article appears to be behind a paywall. Sorry.)
A good short-hand way of understanding this aspect of AI is to recognize that AI’s impressive capabilities are based in part on the theft of millions upon millions of pieces of intellectual property.
Now, if I purchase, or am given, or otherwise acquire a good that I know has been stolen from someone else, that is considered a crime. Why wouldn’t this be the equivalent? That is, why wouldn’t our use of AI all while knowing that it was created on the basis of work stolen from other people, be a criminal activity? Consult your own legal experts. The folks I’ve spoken with suggest that it might, if the crime itself were more visible.
But what we’re dealing with is a very sophisticated erasure, in which there is no way to know for sure that, for instance, when I ask AI to translate a phrase from, say, Polish, into English, what it gives me is in fact the work of Antonia Lloyd Jones, or Bill Johnston, or Jennifer Croft—even though I know from the LibGen list that all three of my colleagues’ work was illegally used in the creation of the AI.
But here it’s as if all the license plates have been changed, and all the VIN numbers scratched out, and the stolen vehicles have been put into a huge parking lot with other cars that weren’t stolen, they were donated, or got so old that their owners forgot about them or died (that’s public domain), or they were just left on the street somewhere and picked up along with a whole bunch of the other random car parts (that’s social media).
So while it’s perhaps not technically illegal, or can’t be proven as such, at least not yet, though a number of lawsuits are in the works, the use of AI for translation, and other forms of creative work, leaves me basically with feelings of impotence, rancor, and betrayal—because I too use social media and I too publish my work, and sometimes it gets digitized or is published online from the start, and I signed contracts against its illegal use without my permission, but that doesn’t seem to have had any effect.
All this suggests that the system itself, especially with the introduction of generative AI into it, has been broken.
I tell you all this to make clear why one of the main attitudes that working translators and other creatives have expressed in the face of AI since its appearance has been aversion. “Get that thing away from me!” It’s an understandable response. One that I sometimes share.
At the other end of the response spectrum, of course, are the early adopters, those who look at what an AI can do and say, “Wow, that’s cool! I wonder what else it can do?”
To which my friends suffering from aversion respond, “Stop training the frickin’ AI!”
But just try telling someone that what they think is cool is in fact not cool, and see how far that gets you.
There are also aspects of this situation, in which a translator might take something that has been done previously and then massage it into shape, that are not at all new. Here I’m referring to the phenomenon often called “ghost translation” in English-language circles. Other cultures have different ways of describing this, and in some cultures it is common practice, especially where a big name author is concerned, to have a group of “under translators” do versions and then hand them on to the author with the famous name.
These terms have problems, I realize. Undertranslation is also a “decolonial translation” method or strategy, in which practitioners deliberately provide less comprehensive or faithful versions; while ghost translation can also refer to what one online source refers to as bridging “the gap between everyday language and the haunting whispers of the afterlife.” (This is not what translators usually think of first!)
The lack of a stable descriptor for the phenomenon is probably telling. I have in mind cases in which someone has created a translation already in some form, and we may or may not know the name of the person who did it. One of the more famous cases in English is the body of translation work (sometimes referred to as versions) of Rumi by Coleman Barks. One has to do a little work to discover that Barks did not know the source language and that these are based on earlier English translations by Reynold Nicholson and Arthur Arberry, and then prose translations by Persian scholar John Moyne.
There are many other such cases in which some unnamed person or persons played a key role. A partner, a parent, a “native informant,” scholars from previous centuries whose works are in the public domain. How is this different, I wonder, from using AI to create a version that gets a translator started?
Here I find myself in an in-between place. Setting aside those feelings of impotence, rancor, and betrayal I mentioned first, I’ve done some experiments with AI and translation, I’ve incorporated it into my translation teaching, and I’ve tried to engage my students in its responsible and careful use, particularly in the increasingly common industry phenomenon known as MTPE, which stands for “machine translation post editing.” (I’ll write later on how this differs from translating from scratch.)
Why I am doing this, and where I’ll turn my attention in later posts, has to do with where I personally find myself in the practice of translation, in translation research, and humanities research in general. It is where I believe we all find ourselves in fact at this moment.
###
