When it comes to medical translation, how much can we rely on AI?

Updated: Nov 17, 2022

Nowadays, though we may not realize it, machine learning and artificial intelligence (AI) are a part of most people’s daily lives.

A 2018 Gallup poll, for instance, found that 85% of American adults use at least one type of device or service that features AI.

Most of our interactions with AI are subtle. Take those computer-generated product recommendations that are displayed on screen when you’re searching for or purchasing a particular item. Or when you listen to music with a streaming service, the new songs or artists that are suggested based on your tastes. These recommendations are all generated by AI.

As the Gallup report notes, AI is also used in other common devices and services like navigation apps, voice assistants, and ride sharing apps. And of course, whether we wanted to or not, many of us have interacted with AI in the form of customer service or medical chatbots.

The world of translation hasn’t escaped the influence of AI, either. Notably, as we’ve previously discussed, most translators (a whopping 88%, according to this report!) use at least one CAT tool. CAT tools are machine learning programs that can do things like regularly plug in a standard translation for a specific term.

AI often makes things easier. But are we becoming too comfortable with it?

Just about anyone who’s interacted with AI in some way has probably also experienced its limitations. Maybe a song or film it suggested had nothing to do with your tastes. Maybe the directions your navigation app gave neglected to account for a detail that only a human being familiar with the route would know. Still, despite these issues, we do seem to put a lot of faith in machines.

This seems to have been the case for a team of researchers from the University of Pittsburgh and their colleagues.

The group conducted a study to see if AI they created could fix a problem that’s long plagued the medical community: translating medical terms into language that can be understood by laypersons.

Most important machine translations today are checked by humans; the research team hoped that their new system would take away the need for human verification.

Machines are known for their limitations when it comes to language. Things like nuance, figurative language, or unusual use of terminology usually won’t compute. Still, you could argue, what the researchers set out to do in this case merely required AI to recognize terms that it was already familiar with, and simplify them in a way it was programmed to do.

And yet, Slater’s Rocío Txabarriag reports that while the AI used in the study was frequently able to replace a medical term with one better suited to laypersons, it still had “issues [with] sentence completeness, fluency, readability, and terminology.” Additionally, it sometimes adapted terms that didn’t need to be changed.

The study’s results show that we still can’t rely on machines for some things. It may be everywhere, but AI’s not ready to completely take over.

Personally and professionally, many of us rely on AI to some extent. In many cases, like recommending songs and videos on streaming services, the results don’t involve life or death matters, so if a 'bot misses the mark, it’s not the end of the world. But it’s important to be wary when it comes to serious matters like communication and medicine. For now, both of these areas will still need to see AI as a tool, not a standalone solution.

Original Image Source

Contact Our Writer – Alysa Salzberg