AI in Healthcare Translations: Understanding the Risks and Implementing Safeguards

The importance of clear and accurate communication cannot be overstated in the ever-evolving healthcare landscape. This is especially true in our increasingly multicultural and linguistically diverse world, where the ability to convey medical information across language barriers effectively becomes crucial.

A study by Berkeley found:

There is growing evidence that clinicians and other healthcare workers use online services like Google Translate as a last resort when no other language services are available. In 2013, Turner et al. surveyed local health departments and found that almost a third had used Machine Translation to translate written materials because they lacked the budget or professional translations. Clinicians have also successfully used Machine Translation to communicate directly with patients, particularly in urgent situations when they had exhausted other alternatives.

As technology advances, artificial intelligence (AI) and tools like ChatGPT are emerging as pivotal aids in breaking down these barriers. However, with their rise comes the need to understand their capabilities and limitations in the context of healthcare translation.

The purpose of this blog is to explore the nuanced role of AI in translating healthcare information.

We aim to provide clinicians with insights into both the potential and the pitfalls of using AI for translation purposes. While AI can significantly enhance communication efficiency, it’s essential to recognize when and how to use these tools appropriately, especially considering the critical nature of medical information.

In this discussion, we will delve into the limitations of AI in translation, exemplifying through specific languages and scenarios where AI might not be the ideal choice.

The Limitations of AI in Translation

While AI and tools like ChatGPT have opened new avenues in language translation, especially in healthcare, their limitations must be acknowledged to prevent potentially dangerous misunderstandings.

The efficacy of AI translation can vary significantly depending on the language and context.

For instance, in New Zealand, healthcare professionals might find that while Maori and Samoan languages are relatively well-translated by ChatGPT, Tongan poses a more significant challenge.

An illustrative example of these limitations can be seen when considering the offerings of major translation services.

Google Translate, a widely used tool, does not support the Tongan language. Conversely, Microsoft Bing Translate and its associated API offer Tongan translations, but the accuracy and context-appropriateness in medical scenarios can be inconsistent.

This variability highlights the need for careful consideration when choosing the right tool for the job.

A Common issue

A striking case that underscores the limitations of AI in this field involves a misdirected translation attempt with ChatGPT.

When tasked with translating educational material about Diabetes into Tongan, ChatGPT, influenced by its training data, produced content more aligned with social media marketing than healthcare.

This error occurred because, as a large language model, ChatGPT predicts text based on patterns it has observed in its training data. If the model has been exposed to significant content about marketing in Tongan but not in healthcare contexts, it tends to generate predictions reflecting this imbalance.

The resulting translation, in this case, completely missed the medical context, highlighting a critical gap in the AI’s understanding.

ChatGPT struggles with Tongan. So we need other strategies.
(Image Source: wikipedia)

This example is a potent reminder of why checks and balances are necessary when using AI for translation in healthcare. Relying solely on AI can lead to significant errors, particularly in languages less represented in the model’s training data.

It emphasizes the need for healthcare professionals to be vigilant and to consider back translation (discussed later in the article) as a part of their quality assurance process. Such precautions are vital to ensure that the information conveyed is accurate, relevant, and, most importantly, safe for patient consumption.

Strategies for Handling Less-Trained Languages

In AI translation, especially in healthcare, a major challenge arises with less-trained languages – those not extensively represented in AI training data sets.

The disparities in translation quality between widely spoken languages and these less-represented languages necessitate specific strategies for effective communication.

Simplification Before Translation

A practical approach is simplifying the language and concepts before employing AI translation tools.

Medical jargon, which can be complex and nuanced even in widely spoken languages, often poses a significant challenge for AI translators when dealing with less-trained languages.

By simplifying the text, healthcare professionals can enhance the likelihood of a more accurate translation. This involves using basic, clear language and avoiding idiomatic expressions or highly technical terms.

Initial Drafts in English

Another effective strategy is to create the initial drafts of medical documents in English.

With English being one of the most extensively trained languages in AI models, it tends to yield more accurate and coherent translations. This step serves as a foundational layer, ensuring the core message is clear and precise before being translated into another language.

Utilizing Dedicated Translation Services

After simplification and drafting in English, the next step involves using specialized (and dedicated to purpose) translation services.

While ChatGPT can serve as a starting point, services like Azure or Google Translate offer additional layers of accuracy for specific languages. Each service has its strengths and weaknesses, and understanding these can help select the right tool for the translation task.

For instance, if a clinician needs to translate medical information into Tongan, a language not supported by Google Translate, turning to Microsoft Bing Translate could be a more effective choice.

However, it’s important to note that even these specialized services have limitations and should be used judiciously.

The use of these strategies not only helps in tackling the challenges posed by less-trained languages but also serves to enhance the overall quality and safety of the translated medical information. By carefully preparing the content and selecting the appropriate translation tools, healthcare professionals can better navigate the complexities of language translation in a diverse patient population.

Back Translation: A Critical Review Process

Back translation emerges as an indispensable tool in healthcare translation, particularly when dealing with AI and less-trained languages.

This process involves translating a document into the target language. This is then translated back into the original language (usually using a different translator or service).

The goal is to compare the back-translated text with the original to identify discrepancies and ensure accuracy.
Understanding the significance of back translation begins with recognizing its role as a quality assurance measure.

In healthcare, where precision is paramount, back translation safeguards against misunderstandings arising from incorrect or imprecise translations. It is especially crucial when dealing with critical information such as medical instructions, where minor errors can have significant consequences.

Implementing back translation effectively often involves using a different service or tool from the one used for the initial translation.

For example, if a medical document is first translated or written using ChatGPT, the back translation might be done through Google Translate or Azure. This approach allows for a broader range of linguistic patterns and interpretations to be considered, potentially catching errors or nuances the first service might have missed.

A practical example of this can be seen when specific medical terms or instructions are involved.

When translating dosage instructions or medical procedures, back translation helps convey critical information such as quantities, timings, and steps accurately. Any discrepancies found in the back translation process can be addressed by adjusting the translation or consulting with a human expert.

The case earlier mentioned, where an educational piece about Diabetes was incorrectly translated into a marketing article, highlights the importance of back translation. The error would have remained unnoticed if this text had not been back-translated, leading to potential confusion or misinformation.

This example underscores the necessity of back translation as a routine part of the translation process in healthcare settings, where the stakes are high, and the margin for error is minimal.

In conclusion, back translation is not just a recommended step but an essential component in the translation workflow for healthcare professionals.

By incorporating this practice, clinicians can significantly enhance the reliability and safety of AI-generated translations, ensuring that patients receive accurate and understandable medical information.

Navigating Cultural Nuances in Healthcare Communication

Addressing cultural nuances in healthcare communication is an aspect of translation that extends beyond the mere conversion of words from one language to another. It involves a deep understanding and respect for cultural differences that can significantly impact how patients receive and understand healthcare information.

This is where AI, despite its advances, often falls short, as it cannot fully grasp and convey the subtleties of cultural context.

In healthcare, where communication is about conveying information and building trust and understanding, recognizing and adapting to these cultural nuances is crucial. For instance, certain medical practices or terminologies that are commonplace in one culture might be unfamiliar or inappropriate in another.

AI tools like ChatGPT may accurately translate the words, but the cultural context can be lost, leading to misunderstandings or discomfort for the patient.

To effectively navigate these cultural nuances, healthcare professionals need to be proactive in understanding their patients’ cultural backgrounds.

This involves literal translation and considering local health beliefs, traditional practices, and societal norms. For example, explaining a medical procedure to a patient from a culture with traditional beliefs about health and illness might require a different approach than a patient from a more Westernized background.

Getting ChatGPT to write the initial text in the reader’s native language can often address this issue. However, depending on the depth of training, the content and knowledge presented may not be as deep. This is also a situation where back-translation is an essential step in the process.

The role of human translators can also be particularly important. While AI can provide a basic framework for translation, human translators can add the layer of cultural sensitivity and understanding that AI lacks. They can interpret the nuances of language and culture, ensuring that the translated information is accurate, culturally appropriate, and respectful.

Furthermore, involving bilingual or multilingual healthcare professionals can enhance communication effectiveness. These professionals can bridge the gap between different cultures, offering insights that might not be evident to someone outside of that cultural context. They can also provide a more personalized approach, adapting their communication style to better connect with patients from diverse backgrounds.

When Precision is Crucial

In healthcare, where the accuracy of information can have life-altering consequences, the role of multilingual human experts becomes indispensable.

Despite the advances in AI translation tools like ChatGPT, there are scenarios where the precision and understanding of a human translator are irreplaceable, particularly in mission-critical situations.

Human translators bring an essential layer of expertise, especially in contexts where nuanced understanding and interpretation are needed. Their knowledge extends beyond the linguistic aspects to encompass language’s cultural and contextual elements.


As you can see, the complexities of AI translation in healthcare are filled with challenges and opportunities.

As this blog has highlighted, AI tools like ChatGPT offer remarkable capabilities in bridging language gaps but are not infallible. Healthcare professionals must be aware of the limitations of AI translation, especially in less-trained languages and when addressing cultural nuances.

The strategies discussed, such as simplification, back translation, and the judicious use of AI and human expertise, are essential in navigating these challenges. They provide clinicians with a framework to leverage AI effectively while ensuring patient safety and care quality.

In healthcare, where every word can carry significant weight, the unchecked reliance on AI translations, particularly concerning pharmaceuticals, medicines, and dosages, is a risk that must be managed with diligence.

AI translations are best utilized for patient education, which can enhance understanding and engagement but should always be supplemented with human oversight.

As we advance in our use of technology in healthcare, it’s imperative to strike a balance. Combining the efficiency of AI with the precision and cultural sensitivity of human expertise will lead to the most effective and safest outcomes for patients. In this way, healthcare professionals can harness the power of AI translation while upholding the highest standards of patient care.

Reference: Reliable and Safe Use of Machine Translation in Medical Setings – University of California, Berkeley

Scroll to Top