The Potential of AI Language Translation Tools for Addressing Healthcare Inequities

Language barriers in healthcare can significantly compromise patient outcomes, leading to inferior comprehension, lower adherence, higher costs, and increased safety risks for non-English-speaking patients. While professional interpreters are the gold standard for addressing these barriers, new research is exploring how artificial intelligence tools might supplement these efforts, particularly for rare or less commonly spoken languages.

During a recent interview with JAMA, Dr. K. Casey Lion, MD, MPH, a research director at Seattle Children’s Hospital, shared her insights on the potential for AI-based translation tools to improve healthcare equity.

Insights From Dr. K. Casey Lion: Challenges in Language Equity

Speaking with Yulin Hswen, ScD, MPH, an associate editor for JAMA, Dr. Lion reflected on the inequities faced by non-English-speaking patients, drawing from her personal experiences as a pediatric resident. She recalled a Spanish-speaking family whose child suffered a delayed diagnosis of a brain abscess due to inadequate communication. Despite access to professional interpreters, the care team relied on their limited Spanish skills, resulting in a failure to identify the child’s worsening condition in time. This incident deeply influenced her focus on language equity in healthcare.

Dr. Lion noted that while the use of professional interpreters has become more accepted over time, barriers remain—particularly for languages of lesser diffusion, such as Chuukese or Marshallese. "These populations already face greater healthcare barriers," Dr. Lion explained. She added that without proper translation, AI models could further compound existing inequities, especially for these underrepresented groups.

Can AI Bridge the Gap?

During the JAMA interview, Dr. Lion discussed how AI-driven translation tools could address critical gaps in language access. She identified two primary areas of opportunity: real-time spoken communication and written translation.

Real-time communication, traditionally handled by professional interpreters, is a challenging space for AI due to the need for cultural nuance, contextual understanding, and the interpretation of nonverbal cues. However, Dr. Lion suggested that AI could be particularly beneficial for languages where human interpreters are often unavailable.

Written translation offers a more immediate opportunity for AI integration. Many patients currently face delays of days when waiting for human-translated discharge instructions, reducing the relevance and utility of this information. Dr. Lion’s team is testing AI-based systems that could provide faster translations, which are then reviewed by human translators. Their study focuses on four widely used non-English languages at Seattle Children’s Hospital: Spanish, Somali, Vietnamese, and simplified Chinese.

Why This Matters for Healthcare Providers

Dr. Lion emphasized that the successful implementation of AI translation tools must be guided by the preferences and needs of patients and families. Her team has engaged a parent task force to ensure patient-centered deployment and is gathering data to address cultural and linguistic concerns. Early feedback from families reveals a split preference—some prioritize immediate translations, even if imperfect, while others prefer to wait for fully verified versions.

For healthcare providers, AI translation tools could represent a significant step toward reducing inequities for non-English-speaking patients. However, Dr. Lion stressed that maintaining a “human in the loop” model—where professional translators oversee AI outputs—is critical to ensuring accuracy and trust.

While AI tools are not yet a replacement for human interpreters, they hold promise for addressing language disparities in healthcare. As these technologies evolve, balancing speed, accuracy, and cultural sensitivity will be essential to their successful adoption. Dr. Lion’s work, as shared in her JAMA interview, highlights the need for thoughtful, patient-driven approaches to integrating AI into clinical practice.