Artificial intelligence (AI) is transforming many industries, including language services, like interpreting, and the healthcare sector. Working at the intersection of these areas, we see our clients struggling to figure out how best to benefit from exciting developments in automation and AI-powered technologies without compromising interpreting and translation quality. At Avantpage, we’re supporting healthcare organizations by implementing AI carefully, safely, and transparently to optimize business processes and translation workflows.
Tools like machine translation can greatly enhance the work of human translators and there are ways that human interpreters can leverage AI to support their job as well. But while Avantpage embraces AI in certain aspects of language access, we encourage our clients to carefully consider the implications of services that offer to entirely replace linguists with AI interpreting technology. These offers of a service that’s faster and cheaper are understandably appealing when you’re on a tight budget. But are there more risks than benefits in a highly regulated industry where patient experience and outcomes are the top priorities?
The National Council on Interpreting in Healthcare¹ recently published their “Guidance for Healthcare Organizations Evaluating the Potential Use of AI-generated Interpreting,” where they highlight many areas for organizations to ask deeper questions.
We consider some of these here, highlighting five core concerns for health plans and hospitals considering an exclusively AI-interpreting model.
1. Language Availability for Interpreting
A crucial part of language access planning is understanding which languages you need to provide. The core languages spoken in your patient communities should be easy to identify. But how will you offer support to a patient who needs an interpreter in a less common dialect or a language of lesser diffusion? It is important to remember that AI models are only as good as the data used to train them. If that data is limited (eg. for indigenous languages or other less common languages and dialects), you cannot rely on the same level of accuracy as a commonly spoken language such as Spanish. If you are using an AI interpreting model, does your provider have a backup solution for languages that are not available through AI? And on the topic of Spanish, can the AI tool pivot to provide a Spanish speaker from Puerto Rico versus one from Mexico, to match the needs of an individual caller? These are important questions to consider when considering an AI interpreting provider.
2. Interpreting Quality Assurance
As mentioned above, it is important to consider whether an AI interpreting service has been trained using sufficient data to provide accurate output for your members or patients. But the quality assurance questions don’t end there.
Human interpreters are required to meet certain qualifying criteria. These include proof of subject-specific training, nationally recognized credentials, field experience, continuing education, and adherence to the National Code of Ethics. Years of experience (Avantpage requires a minimum of 3 years) have taught interpreters to self-monitor and to constantly look out for subtle misunderstandings between the parties. If a human interpreter identifies a communication issue, such as a cultural misunderstanding or a moment of unintelligible speech, they can intervene before it becomes a larger issue. How does an exclusively AI-generated interpreter guarantee the same level of nuanced expertise and service quality? Has the AI been trained with terminology that is specific to the healthcare industry? Does the service offer an opportunity to switch to a human interpreter if it becomes apparent that a quality issue is developing on the call?
3. Interpreting Quality Control
Distinct from quality assurance, quality control is about monitoring performance and improving a service to maximize quality and outcomes for patients. When it comes to written translation, tools with AI features learn and improve through detailed feedback from human linguists. If you are considering an exclusively AI interpreting model, find out how the tool will learn from mistakes. Will there be an opportunity for your staff, patients, or human interpreters to give feedback, and will that be incorporated into the model’s improvement? How does a user submit a complaint, how will errors be tracked, and who is ultimately responsible should a serious error occur? Will the service be spot-checked and monitored from the quality perspective, to be sure that patients and providers are receiving the level of quality they deserve? If you use a service like this, are you complying with federal regulations surrounding quality and AI, such as those outlined in Section 1557 of the ACA?
4. Data Security of Interpreting Platform
Anyone who works in healthcare in the United States understands the importance of the Health Information Portability and Accountability Act (HIPAA) and, in the European Union, the General Data Protection Regulation (GDPR). When a health organization contracts with a language services provider such as Avantpage, that LSP must show that they offer translation and interpreting services that align with these regulations. This includes HIPAA training for staff and interpreters, secure and encrypted phone lines and servers, and many other data security measures that are monitored through annual audits. AI interpreting models work by recording conversations, converting them to text, translating them, and then voicing them in the target language. If an AI interpreting service is recording patient data, where is it being stored and for how long? What measures are in place to prevent a breach? Are patients aware that they are being recorded, and is there an option to opt out of this and be routed to a human interpreter on a non-recorded line?
5. Building Patient Trust with Interpreting
Addressing the factors above is an important part of building patient trust, which is arguably one of the most crucial factors when it comes to providing quality care and improving health outcomes. You want your members and patients to feel that you care at every turn in their journey with your organization. As the National Center for Interpreting in Health Care stated in their guidance¹, “qualified human interpreters often play a crucial, but understudied role not only in ensuring effective communication but also in building empathy and trust in healthcare settings, aspects of care that AI will not be able to fully replicate.”
Where can you carefully implement AI within language services without eliminating the human approach? Can you be sure that your patients are not affected by unchecked racial biases, which can be unwittingly introduced in an AI interpreting tool by virtue of its limited training? If you do decide to offer AI interpreting services, is there an option for your patients to opt out and reach a human interpreter without any delay or other impact?
Supporting Informed Decisions
Avantpage is excited about the progress in this field and we have been incorporating AI-powered processes into our workflows for some time. By asking questions such as those suggested above, healthcare organizations can make informed decisions about the extent to which they are willing to replace human steps with AI. To learn more about how we provide meaningful language access by combining AI with human expertise, contact our team today.
Sources:
1. Guidance For Healthcare Organizations Evaluating the Potential Use of AI-generated Interpreting. (2024). https://www.ncihc.org/assets/documents/publications/NCIHC%20guidance%20for%20contracting%20AI-generated%20interpreting%202024-07-15.pdf
