Who Checked? Why AI Translation Needs Certification Where It Matters
There is a lot of talk right now about regulating AI. Some proposals are poorly thought through; others address real risks. Even when the intentions are sensible, the resulting rules can fail in practice.
Take website transparency. Knowing what data a site collects when you visit it is a welcome idea. Yet the repeated need to dismiss cookie banners just to read a recipe shows how a sound principle can produce a frustrating experience. If we want AI regulation that people respect, we should remember that lesson.
A confession first: I am not naturally drawn to regulation. Standards and checklists can become exercises in compliance rather than something that really achieves its goals. But my personal preference is one thing; what protects the public is another. This article is about the second.
Where standards earn their place
Some products face demanding requirements for good reason. A medical device cannot simply be built, boxed and shipped. Its manufacturer must show that it is safe and performs as intended, manage its risks, and monitor what happens once it is in use. If something goes wrong, there must be a way to trace and correct it.
Compare that with something I build for a sailboat, such as a wind indicator or a cup holder. It still has to be safe, but nobody expects the same level of evidence. If it disappoints, the buyer can return it or choose another product, so the market does much of the work.
Both approaches make sense, because the cost of failure is different. A broken cup holder is an inconvenience. A faulty scanner could cause serious harm. The principle is simple: the greater the potential harm, the stronger the evidence we demand.
Regulated areas, regulated tools?
For everyday use, like translating a menu, a newspaper, a movie or a conference, AI translation is closer to the sailboat gadget. If the result is poor, the consequences are usually limited.
Hospitals, courts and police interviews procedures are different. These areas are already regulated, and the professionals who work in them carry formal responsibilities. A mistranslated sentence there can affect a diagnosis, a verdict or someone’s future, and the person affected often cannot spot the error.
In these areas, it could be useful for AI translation devices to have some form of certification, similar to what happens to many other devices too. I am not suggesting that every AI translation tool needs regulatory approval. But where society already requires safeguards, it is reasonable to ask for evidence about the technology being used.
What certification can and cannot do
Certification does not guarantee perfection. Even certified products can fail. Certification only establishes that a product was evaluated against defined criteria under specified conditions.
In language, perfection does not exist. Translation quality has many dimensions, depends on context, and is sometimes open to debate. Two excellent interpreters may translate the same sentence differently, and both can be right, or wrong.
What certification can do is guide deployers. Independent evidence on accuracy, completeness, latency, and robustness to accents and background noise, tested for specific language pairs and conditions, would give institutions a credible basis for decisions instead of relying only on vendor claims and demos. No more and no less than this.
Realistic, but not idle
We should be honest about where we start. Human translation and interpreting have never been regulated the way medical devices are. Qualifications and accreditation vary widely, and the profession has long relied on training, experience, reputation and institutional trust, with results ranging from excellent to poor. Existing certifications such as ISO 17100 are vague by nature: they cover processes, not the quality of individual translations.
So we probably cannot expect a strict standard for AI translation when none has ever existed for humans. However, that is no reason to have nothing. Between no oversight at all and medical-device-style regulation, there is a lot of room. A good first step would be a minimal, voluntary, independent certification that states which system and version was tested, on which languages, under which conditions and with what results. Such a certification would not claim that a product is safe everywhere. It would simply make information comparable. Professional associations, researchers, developers and evaluation bodies could define the basic criteria together, building on existing evaluation methods. Sector-specific assessments could follow later.
Certification will not eliminate errors or settle every debate about quality. But it could bring transparency and accountability that are largely missing today. We should start working in that direction now rather than wait for a perfect framework.
We do not need to certify AI translation because it is AI. We need evidence about its performance when people are expected to rely on it, especially in highly regulated segments.
When an institution deploys AI translation where mistakes matter, it should be able to answer a simple question:
Who checked?