“This is a matter of life and death”: the hidden dangers of AI in translation

Nearly a third of translators have already lost work to generative artificial intelligence. Professionals in the sector, however, say the dangers of this technology go beyond job loss.

The Tower of Babel. Source: Alexander Mikhalchyk

At a time in which the threat of generative artificial intelligence looms forebodingly over countless livelihoods, the event seemed to be the ultimate manifestation of the battle between human and machine. An audience was seated in a dimly lit room before a panel of experienced translators, who were studying complex, previously unseen texts. A large screen showed their progress as they typed away furiously under the watchful eye of the viewers. It was up to them to decide which iteration of the texts were better – those of the humans, or their AI challenger.

“[One] big takeaway was that the humans were often faster, which really baffled people,” says Christophe Fricker, an Associate Professor in German and Translation at the University of Bristol and the event host. “Human translation is alive and kicking… it can do things that technology can’t do.”

The Human vs Machine AI translation slam made its first appearance in London, after Fricker began to be asked at parties whether working as a translator was “still a thing”. Since then, more events have been organised around the country. The endeavour to assert the value of human translators over AI ones is far from an easily achieved one, however. Companies are often more concerned about their bottom line than the quality of their translated material. Last year, a poll by the Society of Authors found that a third of translators have already lost work to generative AI.

“Over the past eight months, a lot of my agency clients have gone from human translation to MTPE (a form of machine translation with human input called Machine Translation Post-Editing),” says Thorsten Hendriks, director of HPD Translations, a company which takes on diverse projects including websites, games and manuals. “They call it MTPE, but you can see by the quality of it, it’s not machine translation. It’s AI rubbish.”

Hendriks describes himself as a generalist, not a specialist, covering everything from gambling sites to subtitles for Netflix and Warner Brothers. Translators who do not offer the same variety of services, he thinks, may be the worst hit in this new technological landscape.

Other translators, seeking to embrace the inevitable and integrate AI into their work, use a hybrid model in which they check and edit an initial AI-generated translation. For Lloyd Bingham, a board member of the Institute of Translation and Interpreting (ITI), this business model is unsustainable in the long term, as it allows companies to pay “a half to a third of what it would to employ a human translator from the outset.”

“By using AI for translation, you can double your profitability, at least in theory,” he added. “I just don’t know if the companies that are doing this are aware of all the risks. This technology gets stuff wrong all the time.”

He points out that translation markets that have traditionally been seen as “AI-proof”, such as terminology-rich legal, medical and financial documents, are increasingly being translated using AI – with potentially disastrous consequences. “In the medical sector especially, this is a matter of life and death. Do you really want to trust AI with that?”

According to Hendriks, this is not a risk that many of his clients are willing to take, challenging the claim that human translation will inevitably become redundant.

“My MOD (Ministry of Defence) client… wants the human input, because MOD manuals for military equipment are for soldiers to read. They need to be more than 100% correct, to the comma, to the [full stop], everything.”

“That’s why you would always need ‘a human in the loop’ – as much as we hate that term,” he said.

The legal implications of the use of AI do not just extend to the dangers associated with low-quality translations. For Fricker, the law must catch up to tech firms who have used stolen data to feed their AI software. Every other conversation around the issue, he says, is simply “a distraction” from the “rogue, Wild West expropriation of copyrighted materials”.

“The big elephant in the room is that almost everything that has gone into the training of most freely available tools has been acquired illegally. Everybody using most of the tools available is complicit in breaking the law,” he said.

“AI is not just another tool… We’re talking about something that is fundamentally different.”

Most recently, authors have condemned tech giant Meta for allegedly using a database of pirated books to train its AI software. A group of authors affected by the theft are now taking legal action against Meta in the US. Bingham believes that the vast influence of such companies is drowning out the voices of translators.

“It’s hard to cut through the noise, because it’s the tech giants that have the biggest say,” he said.

“They are the ones who are able to get their message across more loudly than any of us can, and they’re making promises that you can maintain or increase quality while cutting your costs. That’s not the case.”

Moreover, the industry has voiced concerns that AI could have a detrimental impact on language acquisition more generally. At the beginning of 2025, Cardiff university, a member of the Russell Group, announced that it would be cutting much of its course offering, including its entire modern languages department. Though the UK has consistently lagged behind its European counterparts in this area, AI may further accelerate the decrease in appetite for language learning in schools and universities, fuelling the decline of graduates seeking to enter the translation and interpreting sectors.

Sophie Hengl, a freelance interpreter, decided to change careers and set up her own counselling business in part due to AI’s impact on the industry. She says that while AI has been much slower to replace interpreters than translators, rapid technological advancements may spell the end for human interpreting.

“If you were my student, I would advise you against becoming an interpreter as I don’t think it will be viable in the future,” she said.

While she believes that top interpreters will survive, “others will either drown or will work for peanuts, or humans will control what the machine produces. There will not be interpreters in [the current] numbers in 30 years.”

“Interpreters are intelligent people who will be able to [retrain] into something else… they will become highly qualified linguists,” she added.

While the future may seem bleak for these professions, Fricker emphasises that there will always be a demand for humanity over automation. “The notion that people are yearning for more anonymity or yearning for more technology is nonsense. The opposite is the case. People look for other people.”