TITLE: Study Warns AI Translation Tools Endanger Patient Safety in GP Care

Artificial intelligence translation tools are increasingly replacing human interpreters in general practitioner surgeries, posing significant risks to patient safety, according to a new study by researchers at the University of Limerick.

The research, which reviewed international studies from 2017 to 2024, found that apps like Google Translate are being used as impromptu solutions during consultations, particularly with refugee and migrant patients. However, these tools have not been tested for reliability in medical settings, leading to potential errors that could result in misdiagnosis, incorrect treatments or serious harm.

Doctors often turn to their smartphones when professional interpreters are unavailable due to time constraints, limited resources or scheduling issues. While this approach addresses immediate language barriers, it undermines the precision required in healthcare communication, where accurate conveyance of symptoms, medical history and treatment options is essential.

The Role of Interpreters in Healthcare

Trained interpreters provide impartial, culturally sensitive support that facilitates clear dialogue between patients and providers. They ensure that nuances in language, tone and context are preserved, allowing patients to actively participate in their care decisions. In contrast, AI tools operate on algorithms that prioritize speed over depth, often failing to capture the bidirectional flow of clinical conversations.

The study emphasizes that healthcare differs fundamentally from casual interactions, such as those at a salon or repair shop, where minor misunderstandings may be tolerable. In medicine, even small errors can have life-altering consequences, especially for vulnerable populations like refugees and migrants who may already face barriers to care.

Evidence of AI Translation Limitations

Analysis of the reviewed studies showed consistent concerns among doctors about AI tools' performance. Common issues included inaccurate rendering of medical terminology, such as terms related to congestion, gestation, reproductive organs and feeding. The apps also struggled with pronouns, numbers, gender references, dialects and accents, sometimes producing substitutions that confused the intended meaning.

More alarmingly, researchers documented instances of 'hallucinations' — where AI generated plausible but entirely incorrect information. These flaws were particularly evident in multi-turn conversations, where context builds over time, a scenario AI is ill-equipped to handle without specialized training.

No evidence emerged from the literature that Google Translate or similar apps have undergone rigorous patient safety evaluations in general practice environments. This absence of validation contrasts sharply with the stringent testing required for other medical devices and technologies.

Impacts on Refugee and Migrant Patients

Refugee and migrant advocates, as noted in the study, strongly favor human interpreters, especially in maternal health and mental health services. These areas demand empathy and cultural understanding that machines cannot replicate. Patients expressed unease about AI use, including questions of consent and data privacy — concerns over how personal health information is stored, processed and potentially shared.

The reliance on untested AI risks exacerbating health disparities. For instance, mistranslations could lead to overlooked symptoms or misguided prescriptions, disproportionately affecting non-native speakers who comprise a growing portion of many healthcare systems.

Recommendations for Safer Practices

To mitigate these risks, the researchers urge healthcare providers to prioritize access to professional interpreters via in-person, video or telephone services. Protocols should be established in every clinical setting to enable swift arrangement of such support, ensuring it is as routine as other essential procedures.

The study concludes that AI tools not designed and validated specifically for medical interpretation should be discontinued in clinical use until they meet safety standards. This shift would prevent the normalization of improvised solutions that compromise care quality.

Attempts to obtain comment from Google on these findings were unsuccessful.

The research underscores a broader tension in healthcare: the push for technological efficiency versus the imperative of patient-centered safety. As AI integrates further into medicine, rigorous oversight will be crucial to protect those most reliant on accurate communication.