Considering Google Translate or ChatGPT for medical interpreting? Not so fast

While artificial intelligence tools are growing in popularity, especially for written translation of non-critical information such as medical appointment notifications, should you trust them to interpret live conversations between patients and medical providers? While it is certainly tempting to jump on the AI trend, here are some things to consider so you act in the best interest of patients – and comply with the law.  

Section 1557 changes

For starters, Section 1557 of the Affordable Care Act spells out when it is and isn’t okay to use AI versus qualified interpreters. If you are a healthcare organization receiving Medicare or Medicaid reimbursements, you must provide language access services to limited English proficiency (LEP) patients that are accurate, timely, free of charge and protect the patient’s privacy. Importantly, only qualified medical interpreters, translators and multilingual staff can provide language access services deemed critical to the LEP patient’s rights, benefits or meaningful access to your facility’s programs. 

If you are using machine or automated translation for speech-based and text machine translations — whether Google Translate, ChatGPT or any other large language model — the AI-translated materials must be reviewed by a qualified human translator in critical circumstances. If machine translation is used in non-critical applications (e.g., appointment reminders), patients must be warned that the translation might contain errors.

Interpreter roles

To adequately evaluate AI interpretation, it must fulfill four major roles that human medical interpreters currently provide:

1. Interpret speech – Professional interpreters are accountable for conveying spoken dialogue between people who do not share a common language. To do their job well, they must be fluent (i.e., speak, read and write) in at least two languages. In a medical setting where a clinician is speaking to a patient, interpreters must be familiar with the subject matter and maintain a high level of concentration to convey the message. If the communication is highly complex, they may take notes and work in teams to ensure the message is accurate and complete. They must also comprehend cultural references, including slang, which may not translate literally.

2. Clarify – Along with interpreting word for word, medical interpreters are often called upon to clarify or verify what is said. This position can be tricky. To ensure the communication exchange is clear to all parties, professionally trained interpreters will announce that “the interpreter is speaking.” They may explain or make word pictures of terms with no linguistic equivalent and check for understanding. 

3. Cultural broker – In this role, interpreters provide a necessary cultural framework for understanding messages that are being interpreted. Cultural differences can lead to a misunderstanding on the part of either the provider or the patient. An example might be a cultural or religious opposition to a certain procedure or process, such as a blood transfusion. The interpreter will check this observation with the patient to ensure that everything being said is repeated in both languages, without engaging in stereotyping or generalization. 

4. Advocate – When a patient’s health, well-being or dignity is at risk, the interpreter may act as an advocate, taking action on behalf of the individual that goes beyond facilitating communication. The National Council on Interpreting in Health Care (NCIHC) notes that this goes beyond the expectations of what healthcare interpreters can do, is not an easy task, and should never be considered a routine intervention. 

Meeting key measures

Along with fulfilling the same roles as qualified interpreters, AI interpretation must meet these four measures:

1. Accuracy – Conveying messages between both parties, completely and with all components, including tone, register and cultural context; this can sometimes be a problem for AI models that are prone to hallucinations (producing inaccurate information)

2. Realism – Accurately interpreting the clinical situation, including expressing empathy in emotionally-charged situations, an attribute that gives humans an advantage over machines

3. Latency – Reducing the amount of time it takes to transcribe the spoken word, translating it into the new language and speaking it back in the moment to ensure that patients understand; while AI is fast, it may struggle to replicate the context of a real-time conversation

4. Cost – Like all technology, you get what you pay for. The more accurate, empathetic and fast your AI technology is, the more it will cost. Users may need to make tough trade-offs in one or more areas to leverage AI technologies cost effectively. 

Creating standards

To properly evaluate AI for the dissemination of general medical information and more, the linguistic industry must establish standards for quality. Today, no such standards exist for human interpreters to deliver at an acceptable level, much less AI. Without a consensus-based, universally accepted definition for linguistic professionals, how are we to define what good looks like with humans? How are we able to determine if AI is on par, better, or worse than a human at interpreting? These are questions the industry must address before moving to a fully automation language services solution. 

Are you interested in evaluating AI for language solutions? Click here to learn more about how GLOBO can help.

Dipak Patel is CEO of GLOBO Language Solutions, a B2B provider of translation, interpretation, and technology services for multiple industries. Previously, Patel spent 20-plus years in corporate healthcare leadership roles. The son of immigrants, he understands the significance of eliminating language barriers to improve healthcare equity.