Machine translation has become common in many fields (not to mention everyday life). But does it hold up in all of them? For instance, what can AI do for health and pharma translation - and what are its limits?
Nowadays, most professional translators use machines to help with their work. These are usually in the form of CAT (Computer-Assisted Translation) tools that help translators do things like check the uniformity of translated terms. But actual translation comes from the human translator, who understands the nuances and subtleties of language, not to mention any legislation or regulations around particular translation jobs.
In the life sciences specifically, AI can also be used for language-related tasks like transcriptions, which take up a lot of doctors’ time during consultations, as well as automated tasks like billing. AI has also been helpful in certain aspects of medical research, like data analysis.
And in an emergency situation where it’s impossible to contact a medical interpreter, a machine translation tool could be invaluable, even though there’s always a risk of mistranslation. A number of sources put the rate of AI translation errors at around 10%, an impressively low amount, all things considered - but still very worrying when you consider that 8% of these translation mistakes can result in serious complications for patients.
When it comes to other aspects of health and pharma translation, though, a human touch is definitely needed. For instance, instructions for outpatient care require knowledge of things like the patient’s culture: How are instructions usually presented? What customs and practices apply to their particular culture (for instance, how often and when do people have meals, what are typical hygiene products, etc.)?
This cultural knowledge is also essential when it comes to medical interpreting. For one thing, different cultures approach and discuss illness and treatment differently, something AI wouldn’t be aware of.
Machine-generated pharma translations also come with some risks. In a field full of regulations - many of which can change based on markets and new legislation - any error can have outcomes like financial loss and risks to patient health. For instance, we previously reported on Chinese pharmaceutical company Guangdong Zhanjiang Jimin Pharmaceutical Co. being blocked from releasing a product to the US market because of a mistranslation of one of its components.
On the patients’ end, we also recently covered a medical machine translation error where AI failed to recognize the medication Coumadin and translated a doctor’s instructions to stop taking it as “Do not take anymore soybean.”
There’s another issue with machine translation in the health and pharma field that many of us might not realize: The lack of confidentiality.
Most AI translators aren’t programmed to protect the material they’re given to translate. This is because most AI learns by gathering data, which it will be able to use in the future. This means, as AI expert Lance Eliot explained in a recent Forbes article, confidential information concerning patients or products could end up being shared somewhere else online.
According to medical fax company WestFax, for instance, most AI, including ChatGPT, is not HIPAA compliant, something that can easily be spotted by reading terms and conditions, which specify that information and conversations could be collected and monitored.
It’s possible to work around this issue. Many experts suggest setting up extra security methods and logins. Eliot also says that an organization could create its own AI that wouldn’t save this kind of information or would only keep it within the company’s database. But he acknowledges that most organizations lack the funding and resources for this.
Machine translation and other AI applications have been helpful to the health and pharma field in certain ways. But they aren’t able to take on every translation task. When it comes to confidentiality, compliance, clear patient instructions, and compassionate healthcare interpreting, humans are still the best - even, in many cases, only - solution.