The emergence of DeepSeek has had a dramatic effect on the global artificial intelligence (AI) scene. This Chinese alternative to ChatGPT, made for a few billion less, has not finished spilling ink, just like AI in general. While it has made inroads into many fields, translation and interpreting is no exception, and language barriers have never seemed so easy to overcome. Yet behind this apparent simplicity lie major ethical and practical issues.
Are you a professional translator or interpreter, or just a user? Here are 5 tips to help you keep your head above water when it comes to AI.
We all know that AI-based translation tools are impressive and constantly evolving, but they remain fallible tools.
Our ability to conceptualize and feel a wide spectrum of emotions sets us apart. In the same way, translation and interpretation professionals bring their own unique expertise, based on their own sensitivities and choices.
AI can therefore save time on simple linguistic questions, but any demanding project deserves a qualified translator or interpreter.
Generative AIs rely on algorithms that have been “fed” millions of data points to predict the most likely translations.
The first pitfall is results that lack cultural nuances, specific contexts or linguistic subtleties.
A new paradox is also emerging: after years of absorbing available human data, AIs are now capturing texts that they themselves have generated.
This is leading to an impoverishment of language, as well as an upsurge in hallucinations, those misleading responses presented as proven fact. As researcher Marc Watkins observes, “There's a new ‘Gold Rush’ to get quality human-generated data.”
One of the risks associated with the use of AI in sensitive translations is the possibility of leakage or exploitation.
Some AI publishers store information entered by users in order to improve their models. This happened to Amazon, which had to instruct its employees to stop disclosing confidential information to ChatGPT.
Providing as little information as possible to the AI is therefore strongly suggested. Particularly if sensitive, medical or legal translations are involved. In those cases, make sure you use platforms that guarantee confidentiality, encrypt data, or go for locally hosted solutions that don't require information transfer.
AIs are biased “by nature”, because their systems are based on existing data that may contain cultural, gendered or ideological biases.
UNESCO points out that, in response to the question “What are the great historical figures?”, AIs give almost no women's names, for example. An automatic translation can therefore reinforce racist, sexist or homophobic stereotypes.
Even more alarmingly, if you ask DeepSeek about the events in Tian'anmen Square or the working conditions of Uighurs, the AI asks you to move on with a blank “Let’s talk about something else”.
So it's important to remain critical of the proposed results, and not hesitate to adjust the translations so that they are as inclusive and respectful to the original message as possible.
Unlike human translations, which are protected by copyright if they are original, those generated by an AI do not benefit from this protection: a machine is not recognized as an author.
What's more, data is often collected without the authors' consent, raising licensing and plagiarism issues. The famous New York Time has filed a complaint against ChatGPT, accused of “regurgitating” content from the newspaper.
It is still possible to contact most platforms and ask them to remove your texts, but you need to be aware of this and be able to prove it.
Finally, what about legal responsibility? If a machine translation leads to errors, or even serious consequences, whose fault is it? The human who used the machine translation tool, of course.
The AI landscape is evolving faster than it can be written, so it's important to understand its limits, maintain an ethical practice and value human expertise.
Conferences and roundtables, UNESCO recommendations, best practices... there are many resources available to help us keep abreast of the ethical debates surrounding AI.