Is AI coming for everyone’s job?

It is not just translators who are wondering whether AI is coming for their jobs. A recent BBC article asked: “How long will humans be needed for translation?” Meanwhile, mathematicians have been having a remarkably similar conversation about their own profession.

OpenAI recently announced a $5 million contribution to a new Simons Foundation initiative. Its purpose is to help mathematicians examine, understand and build upon results produced by AI. I was rather struck by that. AI systems can now produce mathematical work at a speed and scale that humans may struggle to keep up with – but mathematicians are still needed to work out whether the results are correct, genuinely new, properly understood and ultimately useful. I heard a piece on BBC Radio 4’s news programme (today, 19 Oct)  where the mathematicians interviewed were pretty upbeat and confident about their future – and were still encouraging students into the profession. 

Because producing an answer is not quite the same as knowing whether it is a good answer. Those of us working in translation may feel this sounds rather familiar.

Apparently, we are all being replaced

Over the past few years, I have read countless headlines announcing that AI is about to replace translators. Before that, machine translation was going to replace us. Before that, translation memories were going to replace us. Atlas Translations was founded in 1991, so we have seen a fair few technological revolutions come and go. Some have genuinely transformed the way we work. Some have been useful but rather less revolutionary than advertised. Others have arrived accompanied by trumpets, predictions of imminent human redundancy and, usually, an expensive subscription. Technology has undoubtedly changed translation. It will continue to do so – and at a FAST rate.  But translators are not uniquely under threat. Accountants, lawyers, designers, writers, doctors, programmers, teachers and now mathematicians are all having versions of the same conversation:

If AI can produce something that looks like my work, where does that leave me?

It is a perfectly reasonable question – but I’m not convinced that “AI or human?” is the most helpful way to frame it.

What is the content for?

A better question is:

What level of quality, risk and accountability is appropriate for this particular content?

If you want to understand the general gist of an online review, a quick automated translation may be perfectly adequate. If you are translating a patient information leaflet, a legal agreement, evidence for court, medical-device instructions or a pharmaceutical submission, “perfectly adequate” needs considerably more unpacking. The consequences of an error are not the same. A slightly odd phrase in an internal email may cause nothing more serious than a raised eyebrow. An ambiguity in a consent form, contract or dosage instruction could have legal, financial or safety implications. That is why the right translation workflow should be determined by factors including:

• What the content is
• Who will read it
• What they will do with it
• Whether it contains confidential or personal information
• Whether it is regulated
• How long the translation will remain in use
• Whether it will be published
• What could happen if something is misunderstood
• Who is prepared to take responsibility for the result

This is the conversation we increasingly have with clients at Atlas. Sometimes the appropriate answer will be professional human translation followed by independent revision. Sometimes specialist legal, medical or technical review will be needed. Sometimes machine translation with professional post-editing may be suitable. And sometimes a client simply needs the gist. Translation is not one single product. Nor is every document equally important.

Human-first does not mean anti-technology

Atlas is proudly human-first, but that does not mean we sit in the office muttering darkly at the computer. We use technology throughout our work. It helps us manage terminology, maintain consistency, analyse files, coordinate multilingual projects and avoid wasting time and money by repeatedly translating identical content. We also offer machine translation post-editing when a client specifically wants it and when it is suitable for the material. What we do not do is quietly put a client’s content through an AI translation tool without their knowledge. Our translators must not use AI translation unless it has been specifically requested and agreed.

That distinction matters.

Clients should know how their content is being handled, particularly when it contains confidential, commercially sensitive, legal, medical or personal information. They should also understand what level of human checking is included- and what is not.

“AI was used” does not describe a complete workflow any more than “a human translated it” guarantees that the right human was chosen, the terminology was researched or the translation was independently checked. The detail matters. It generally does.

Where humans still earn their tea

A good professional translator does far more than replace individual words. They make decisions about meaning, tone, context, terminology, culture, ambiguity and audience. They recognise when the source text contains a mistake. They ask questions when something does not make sense. They notice when a beautifully fluent translation says something subtly, but importantly, different from the original. They also understand that the correct translation is not always the most literal one. And, crucially, professional translators and language-service providers can be questioned. They can explain their decisions, respond to feedback and take responsibility for the work.

The same principle appears to be emerging in mathematics. AI may generate a result, but humans are still required to test it, interpret it, place it within an existing body of knowledge and decide whether it can be trusted. That is not a minor supporting role. It is the difference between output and understanding.

The conversation is changing

For translation companies, I think the commercial conversation is moving on. It is becoming less about defending human translation against every new tool and more about helping clients choose the right process.

The question is no longer simply:

“Should we use AI or a human?”

It is:

“What combination of technology, professional expertise, checks and accountability is appropriate for this content?”

That is a more useful conversation, and a more honest one. AI will continue to become faster and more capable. Some types of work will change significantly. Some routine tasks may disappear, while other roles will develop around checking, advising, validating and making sense of what machines produce. But when accuracy, confidentiality, nuance, safety or reputation matter, human judgement is not an optional decorative flourish added at the end. It is part of the service.

So, is AI coming for everyone’s job?

Possibly. It certainly appears to be making the rounds.

But it may be more accurate to say that AI is changing everyone’s job – and making it even more important to be clear about where human knowledge, judgement and responsibility are still needed. Now, if you’ll excuse us, the translators and mathematicians may need to put the kettle on and compare notes.

Further reading

OpenAI: Sharing AI progress in mathematics

BBC: How long will humans be needed for translation?

Clare Suttie is founder and director of Atlas Translations.