Where AI is Falling Short in Healthcare: Lessons from the Front Lines

Dipak Patel, CEO, GLOBO , May 1, 2025

Artificial intelligence (AI) is rapidly becoming a staple in healthcare.

New solutions are entering the market at a lightning-fast pace. In fact, the United States has over 4,500 startups that focus on AI in healthcare, 47 of which were founded since the start of 2025.

The emergence of these tools stands to create operational efficiencies, assist with diagnoses,  and improve patient care. But beneath this excitement lies a more sobering reality: AI is still far from infallible. In some critical areas, it’s falling short, sometimes with life-threatening consequences.

Trust is a key component of healthcare—and in order for AI to be successful, we need reliable solutions. Accenture's Technology Vision 2025 report emphasizes the critical need for trust in AI systems, which are built upon foundations of accuracy, predictability, consistency, and traceability.

What happens when AI fails our providers? And more importantly—how can we use those lessons to build better systems?

Here are three examples that should make all of us in healthcare approach implementing AI with extra care and consideration:

1. Prediction Gaps: AI Predictive Tools Missing Critical Patient Risks

A March 2025 study published in Communications Medicine tested machine learning models developed to predict patient mortality risk in hospitals. These systems were meant to flag early signs of patient deterioration—a tool that could, in theory, save lives. But the results were far from reassuring.

The study found that these models missed 66% of cases where a patient was at risk of death due to injuries and complications. In other words, AI systems failed to identify two-thirds of patients needing urgent attention.

For clinicians, this represents a serious concern. Predictive models are only as good as the data they are trained on—and when they fail, they don’t just cost money; they cost lives. If organizations are going to integrate AI into major care decisions, we need rigorous, continual validation in real-world environments—not just promising test data.

2. Automation Bias: How AI Can Undermine Clinical Confidence

What happens when AI causes highly skilled healthcare professionals to second-guess their expertise? This is known as “automation bias.” In a 2024 study, researchers studied this phenomenon by examining pathologists working under time constraints.

The research focused on how AI-assisted decision-making could unintentionally lead doctors to trust the machine’s output over their own better judgment. Surprisingly, pathologists who initially made correct diagnostic assessments were 7% more likely to override their decision in favor of an incorrect AI recommendation when under time pressure.

This is very concerning. This outcome suggests that when not properly contextualized or explained, AI tools can offset clinical judgment. In high-stakes environments, doctors need support—not distractions. To prevent this, we need to design AI systems with explainability and transparency, not just predictive power.

3. Built-In Bias: When AI Learns Our Flaws

Finally, we can’t discuss AI concerns without addressing the issue of bias. Several reports in recent years have noted that AI algorithms can inherit and amplify systemic inequities based on how the tools are developed and trained.

In a notable study of algorithmic bias, researchers found that an AI tool used across major U.S. health systems was found to recommend additional care for healthier white patients while underserving sicker Black patients. Why? The model used healthcare cost as a proxy for healthcare need, and historically, Black patients have received less care—not because they needed less, but because of longstanding disparities.

This is a prime example of why equity must be strategically included in AI development from the ground up. Diverse training datasets, continuous auditing, and ethical review boards are critical for creating equitable patient experiences and outcomes.

Human-Centered AI: Designing for Better Healthcare Outcomes

According to Julie Sweet, Accenture chair and CEO, ” ... unlocking the benefits of AI will only be possible if leaders seize the opportunity to inject and develop trust in its performance and outcomes in a systematic manner so businesses and people can unlock AI’s incredible possibilities."

AI is not the downfall of healthcare—it’s one of the most promising innovations to hit the industry in recent years. But we have to be honest about where it’s currently underperforming. The path forward isn’t blind adoption of new tools. Instead, it’s critical integration, driven by real data, cross-disciplinary collaboration, and an unwavering focus on positive patient outcomes.

Healthcare leaders and providers need to stop asking, “Can AI solve this?” and start asking, “Should it?” And if so, how can we do it better?

It’s critical for hospitals and health systems to partner with companies that have a proven track record in the fields their solutions aim to solve. These tools should be grounded in research and co-developed with healthcare systems in mind.

At GLOBO, we’re on a mission to transform patient communication. As a leader in the language services industry, we have over 15 years of experience reducing friction for healthcare providers serving multilingual patients. We envision a future for language access that leverages both human and technology-enabled solutions to improve experiences for healthcare organizations and their diverse communities.

Want to learn more about GLOBO's AI-enabled innovations? Request a briefing session with one of our language access experts here.

Advancing Patient Safety: How AI Can Improve the Quality of Patient-Provider Communication

Dipak Patel, CEO, GLOBO , March 10, 2025

Poor communication between providers and limited-English speaking patients is known to drive up the cost of care through increased readmissions, extended hospital stays, and preventable medical errors.

Despite the widespread recognition of this challenge, some healthcare organizations and their language service providers struggle to consistently evaluate and improve the quality of medical interpreting services.

With over 68 million people in the United States who speak a language other than English, ensuring accurate medical interpretation isn't just a compliance requirement; it's imperative for critical patient safety, experience, and clinical outcomes. But how do we know if the interpretation encounter is good or just good enough?

The Connection Between Quality and Health Outcomes

Communication barriers can severely affect the quality of care patients with limited English proficiency (LEP) receive. Without adequate language support, LEP patients are:

  • Less likely to have an understanding of their diagnosis
  • Less likely to understand treatment plans or follow-up care
  • Less likely to feel confident in the care they received
  • More likely to have a negative experience with their care providers

Medical interpretation can positively impact patient care. According to Jennifer Winters, assistant vice president of Value Based Care at Affinia Healthcare, utilizing high-quality interpretation from the start of care is fundamental. “Recognizing the patient's cultural and linguistic needs can create a greater level of understanding, lead to better outcomes, and reduce repeat visits, as it allows healthcare providers to see a more complete picture of the patient's health,” said Winters.

Plenty of research highlights that patients who receive care in their preferred language tend to have better health outcomes than those who do not. In one study examining readmission rates and the length of hospital stays for LEP patients, researchers found that over 24% of patients who did not have access to an interpreter during admission and discharge were readmitted within 30 days. Additionally, patients who used an interpreter had a significantly shorter stay than LEP patients without an interpreter (2.57 days vs. 5.06 days).

The Gap in Quality Measurement

Despite the connection between implementing language support along the patient journey and health outcomes, organizations face obstacles in assessing interpretation quality. While there are guidance and quality standards for interpreters, we see an opportunity to better define how interpretation encounters are evaluated. Current monitoring practices vary across health systems and language services providers.

Health systems serving diverse populations have higher interpreting service utilization rates—one of the largest health systems we serve, on average, uses over 100,000 minutes of video remote interpreting each month. With a high volume of interpreting sessions, humans can only evaluate a small fraction of these interactions manually. The limited sampling of calls makes it difficult to ensure consistent quality and experience across the board.

Live Quality: A Proactive Approach to Interpretation Assessment

At GLOBO, "good" isn't good enough when it comes to patient safety and experience for LEP populations. We are working toward redefining what it means to provide interpretation excellence. It begins with our new Live Quality tool, which uses AI technology to review 100% of GLOBO’s video and audio interpreting calls.

“We are thrilled to be the first in the language services industry leveraging AI to ensure all interpreter sessions are reviewed,” said Elizabeth Robeck, senior vice president of operations at GLOBO Language Solutions. “This is a real game-changer for providing greater insight into the quality of each interaction and delivering consistent experiences for patients and providers.”

Through the GLOBO Live Quality tool, we are able to quickly assess anything that might be disruptive during interpreting sessions—from backgrounds and attire to lighting and audio quality—to ensure a compliant and professional experience during every encounter.

By smartly leveraging AI technology, we can help our healthcare partners:

  • Gain a better understanding of interpretation quality across all encounters
  • Address issues before they impact patient care
  • Improve patient understanding and confidence
  • Enhance patient-provider experience and satisfaction

Advancing Toward a New Standard in Medical Interpreting

With the capability to evaluate every interaction, we are one step closer to better defining measurable standards for assessing the quality of each medical interpretation session. GLOBO Live Quality is more than just an improvement in language services—it's a crucial step closer to ensuring safer, more effective care for all patients, no matter what language they speak.

Want to learn more about GLOBO's AI-enabled innovations? Request a briefing session with one of our language access experts here.

GLOBO named Deloitte Technology Fast 500 High Growth Company

Dipak Patel, CEO, GLOBO , November 25, 2024

Financial performance reflects skyrocketing demand for translation and interpreting services, especially for the medical community

GLOBO Language Solutions, headquartered in Philadelphia, Pa., has been named a Deloitte Technology Fast 500™ winner, placing them among the 500 fastest-growing technology, media, telecommunications, life sciences, fintech, and energy tech companies in North America. GLOBO ranks No. 455 on Deloitte the list, now in its 30th year. 

With a consecutive three-year percentage growth rate of 253%, GLOBO’s extraordinary financial performance clearly reflects America’s strong demand for translation and interpreting services to serve the estimated 11 million people in the Deaf and hard of hearing community and the ever-growing limited English proficiency (LEP) population, which represents more than 25 million people or approximately 8% of the U.S. population.

While GLOBO serves multiple industry sectors, nowhere is the need for its language communication support more critical than healthcare, where a lack of understanding between practitioners and patients can literally mean the difference between life and death. Along with impacting quality of care, numerous evidence-based studies report unresolved language barriers and poor health literacy that can lead to regulatory fines, opportunity loss, medical malpractice and other expensive lawsuits, and preventable health expenditures, all of which can impact the total cost of care.

Sizing the challenge

Nimdzi estimates that the language services industry reached USD $67.9 billion in 2023 and projects it to grow to USD $72.7 billion in 2024. At a compounded annual growth rate (CAGR) of 7%, the market is expected to reach USD $95.3 billion by 2028. Hospitals and other provider organizations receiving reimbursement from Medicare and Medicaid are required by Section 1557 of the Patient Protection and Affordable Care Act to provide oral and sign language interpretation and written translation for Limited English Proficiency (LEP) patients.

Although the estimated cost of providing interpreter services is relatively low at $279 per person per year, the hidden cost of not doing so or only sporadically can be substantially higher. In a recent GLOBO webinar, Tricia Nichols, MSN, R.N., NEA-BC, CPXP, Patient Experience Director for Jefferson Health–North Region, noted that non-compliance quality penalties can reduce Medicare reimbursements by as much as 2%. She estimated that when a patient leaves Jefferson or any other healthcare facility due to a negative experience—such as language barriers or other factors—and chooses a different provider, the lost revenue from that patient over their lifetime of care can amount to an estimated $1.5 million.

Damages can be equally staggering from medical malpractice claims due to a failure to provide an interpreter, often resulting in communication breakdowns, misdiagnoses, and medical harm. A study conducted by the University of California at Berkeley, School of Public Health, pursuant to a contract with the National Health Law Program (NHeLP), analyzed the medical malpractice claims of a malpractice carrier in four states to identify when language barriers may have resulted in harm to the patient. In 35 claims, the carrier paid $2,289,000 in damages or settlements, and $2,79,800 in legal fees.

Harder to quantify are the additional costs of healthcare resources incurred by the LEP population due to communication barriers, some of which include:

  • 30% more emergency department (ED) visits, with 24% of LEP patients more likely to have an unplanned ED revisit within 72 hours of care
  • 40% higher rates of hospitalization
  • 1.5 days longer length of hospitalized stay
  • 42% more imaging
  • Higher likelihood of an adverse event causing some physical harm (49% versus 30%)

One historical cohort study based on data at the Mayo Clinic in Rochester, Minn., found that patients with LEP had higher costs during hospital admission to discharge, with an average difference of $3,861 versus $3,166, or $695 more per LEP patient. Carolinas Healthcare System reported saving $1.5 million annually by using remote interpreter technology to improve LEP communication.

Meeting demand

Closing equity gaps should be the goal for any organization wanting to truly meet the needs of its LEP population while controlling unnecessary costs at the same time. This begins by providing linguistic services at key touchpoints throughout the patient’s health journey, most commonly an ED visit and hospital admission, discharge or both. At GLOBO, we believe effectively supporting LEP patients necessitates a holistic approach, seamlessly integrating language services at every medical interaction.

The LEP patient experience

Imagine yourself, for a moment, being in the shoes of an LEP person. In a perfect world, you’d  receive your provider’s text messages translated into your preferred language. Scheduling options are translated for you as well as visit prep information. If transportation assistance is needed, you’re able to schedule a ride in your native language and, upon arrival, you are greeted and provided  translated documentation about your rights to an interpreter.

During your healthcare visit, an interpreter is present via live, audio or video to ensure that you and your provider are communicating clearly. During check-out, you receive a translated version of your treatment plan, discharge instructions and follow-up scheduling. Any medication instructions and test results are translated, as well as billing statements.

Creating broad coverage

While this may seem futuristic, these technology-enabled linguistic services are possible today. To create a broad-based, cohesive language solutions strategy, organizations need access to live and remote translation and interpreting resources whenever and wherever the LEP patient needs it.

Indeed, GLOBO’s network of live interpreters, which support 430-plus languages, are available 24/7/365 through audio, video, or on-site. Providers can schedule interpreters via the GLOBO HQ platform or the GLOBO Connect mobile app, which puts an interpreter in the pocket of providers to further extend support. The platform, which enables users to schedule interpreting sessions at 15-second connection speeds, integrates via a simple application programming interface (API) into an electronic medical record (EMR) system such as Epic.™ It also provides real-time access to linguistic data to monitor overall usage costs and analyze trends.

Leveraging artificial intelligence (AI) support for translation and interpreting is clearly the next step, with applications expected to accelerate rapidly. Already, AI is in use today as prompts for routine intake tasks such as translated text messages for appointment scheduling. Expected to become ubiquitous, AI will fill in even more care delivery gaps, ensuring non-English speakers always have access to language support.   

Want to learn more about how to effectively integrate language solutions across your patient’s health journey? At GLOBO, we understand the complexity, scale, and importance of successfully managing a language support program. Request a demo here with one of our language access experts.

GLOBO Researches AI Models to Inform Healthcare Implementation Strategies

Dipak Patel, CEO, GLOBO , September 3, 2024

For healthcare administrators wanting to dramatically increase access to language solutions, Artificial Intelligence (AI) promises unlimited possibilities – especially if integrated across the patient journey safely and effectively.

Broadscale deployment of AI stands to deliver enormous benefits for limited English proficiency (LEP) patients and providers alike, ranging from greater convenience to lifesaving interventions that positively impact health outcomes, safety, treatment adherence, and more.

So, what’s holding up widespread integration of AI for medical translation and interpretation? To create an effective strategy, healthcare leaders must first know the strengths and limitations of AI technology today, and how to best apply it short- and long-term. Given the flurry of vendors entering the AI linguistics market, perhaps the biggest barrier is knowing where to start and who to trust.

GLOBO pilots LLM evaluation

To help providers evaluate AI capabilities, GLOBO conducted a three-month research study in 2024 to evaluate the performance of multiple large language models (LLMs) and their multimodal variants. What we found, quite simply, is that all AI technologies are not created equal. Our findings and observations are presented in our newly published research, “GLOBO AI-Powered Medical Interpretation Study: Insights for Health Leaders,”  which is available for download here.

Our research is designed to provide healthcare leaders with key insights into AI interpretation performance. This guide presents an executive summary focused on four key domains:

  1. Assessing the process of AI interpretation
  2. Evaluating how AI-enabled interpretation is measured
  3. Exploring the current state of AI tools
  4. Identifying where LLMs fall short with interpretation

 

Making AI Work in Medical Linguistics

Unlike human interpreters, AI requires a series of steps to successfully communicate from one language to another:

  1. The verbal message is transcribed into text
  2. Text is translated into the selected language
  3. The message is converted into speech

 

While this may seem relatively straightforward, each step requires a specific AI technology to perform the transcription, translation, and speech functions. When integrated, they must fulfill the same role as a human interpreter, meeting stringent quality measures for:

  • Accuracy – Conveying messages between both parties, completely and with all components, including tone, register, and cultural context
  • Realism – Accurately interpreting the clinical situation, including expressing empathy in emotionally charged situations
  • Latency—Reducing the amount of time it takes to transcribe the spoken word, translate it into the new language, and speak it back in the moment to ensure that patients understand
  • Cost – Like all technology, you get what you pay for. The more accurate and realistic your AI technology is, the more it will cost. 

With no out-of-the-box solutions optimized specifically for medical interpreting, healthcare organizations must be prepared to configure and fine-tune AI models to ensure they meet all predefined quality measures, which ultimately will increase efficiency and reduce costs.

AI Pitfalls

When assessing the limits of available LLMs, we identified common pitfalls during each step of the AI interpretation process. In the transcription phase, some models simply couldn’t handle transcribing multiple languages at a time. They also had difficulty transcribing short statements and were essentially incapable of assessing uncertainty, lacking a common-sense filter to ask for clarity, which a live interpreter would instinctively do.

For translation, we found that LLMs may refuse to translate critical information, mistakenly flagging a text as harmful or inappropriate. Worse, LLMs may hallucinate, duplicate or completely leave out details of a translated text, and incorrectly translate a word based on the context of the conversation.

Speech synthesis models also produced hallucinated speech, repeating syllables or unintentionally creating audio artifacts such as yawns. Speech LLMs also had difficulty with the pronunciation of short statements and generated non-native-sounding speech in some cases.

Keeping Up with Rapidly Evolving Technology

All of these challenges aside, it is important to acknowledge that AI tools are evolving rapidly, with the current language models continuously learning and being fine-tuned to create better outputs and user experiences. For instance, OpenAI, which launched its first version of ChatGPT in 2018, released its fourth generation in 2023, building on previous releases to enhance its output and ability to generate more advanced responses. As OpenAI and other technology companies continue to expand language models, tasks once deemed impossible are now becoming a reality.

It is incumbent on healthcare leaders to keep pace – or get left behind.

To effectively integrate AI interpretation into care settings, healthcare leaders must focus on finding a solution that can deliver fast, accurate, and empathetic interpretation. Collaborating with a trusted partner is one way to ensure AI technologies are tested and configured to meet your organization’s language interpretation needs at all points of care interaction across the patient’s health journey. 

Since 2010, GLOBO has been a leader in interpreting and translation services, providing an innovative approach to communicating with diverse patient populations. Our team of experts is dedicated to testing and designing the right AI-enabled tools to help your hospital, health system, or medical practice communicate with multilingual patients when it matters most. 

Don’t allow the complexities of AI to hinder your goals for enhancing and expanding linguistic services to your healthcare staff and patients. Download The GLOBO AI-Powered Medical Interpretation Study: Insights for Health Leaders here and let’s discuss how we can help your organization leverage AI interpretation tools to better serve your non-English-speaking population.

Evaluating AI Beyond Accuracy: The Art and Science of Translating Human Expression

Dipak Patel, CEO, GLOBO , July 1, 2024

Anyone who has experienced artificial intelligence (AI) for rudimentary translation capabilities undoubtedly knows that it may not translate the spoken word entirely accurately – even when the AI-generated words are a literal match.

Human variables such as tone, emotion, cultural nuance, race, and gender all play into how well or poorly the meaning of a conversation is truly captured and conveyed.

This is why American poet Amanda Gorman carefully vetted translators for her inaugural poem, “The Hill We Climb,” now translated into 17 languages. In Gorman’s case, she went to great lengths to find a like-minded translator who would reflect the spirit of her poetry – choosing, in one case, a resource known for being outspoken on issues including gender equality and mental health.

The debate, of course, is that translating human expression is both an art and a science, especially in emotionally charged conversations where gaining understanding is more complex than conveying what is actually said. While AI holds great promise for the translation and interpreting industry over the long term, it is important to understand the full range of capabilities required to be effective, especially when deciding where it best fits in healthcare today. To do so, one need only look at the four defined roles of live interpreters:

  1. Serving as a conduit for verbatim translation
  2. Providing clarification
  3. Acting as a cultural broker
  4. Advocating for the patient’s best interest

 

Breaking down the roles

Perhaps the most straightforward and simplest role is that of a conduit in the translation of verbal words from one language to another. This task is most definitely about getting the words right. For instance, a hospitalized patient might ask, “What’s on the lunch menu?” In this case, AI may be a reasonable solution to simply list the various choices.

However, suppose a physician tells a patient, “You’re hypertensive,” and there is no similar word in their native language. In this case, it is incumbent on the translator to provide clarification, explaining to the patient that the doctor means the patient’s blood pressure is too high. If the patient still doesn’t understand, a highly skilled translator must be ready to apply the right knowledge to explain what causes elevated blood pressure and what the patient must do to control it.

Being a cultural broker takes this one step further. Going back to the menu example, the patient may be of a nationality that commonly adheres to a vegan diet. Helping to identify selections that are in keeping with their dietary and cultural preferences may be in order. On a more critical level, a patient may be diagnosed as diabetic and prescribed insulin injections. In some cultures, use of needles is associated with addiction, and, without proper understanding and agreement, this treatment could result in patient non-compliance and medical complications.

In the case of the diabetic, the interpreter may be called to assume the role of advocate to explain the compliance challenges and even advocate for a different treatment approach, if possible. Interpreters may also serve as advocates on issues related to social determinants of health (SDoH), including food insufficiency, lack of transportation to keep medical appointments and other challenges.

Choosing the right vendor

With an explosion of new AI vendors emerging on the healthcare market, provider organizations must thoroughly evaluate capabilities, ensuring their services can address all levels of translation and interpretation complexities patients may need. This is where working with a knowledgeable partner is especially valuable. Just some of the things to look for in a language solutions partner include:

  • Deep healthcare experience to fully understand demographic and cultural requirements
  • A progressively improving database and machine learning technology to evolve AI beyond the conduit level addressing tone, empathy, and other variables
  • Ongoing support to continue refining and introducing solutions specific to patient needs

Today, it is safe to say that most AI vendors are solidly at the conduit level. While AI advances are underway to address variables such as tone and empathy, which mimics humans, live interpreters are still best suited to deliver difficult diagnoses and other emotionally provoking information. In some cases, trained AI bots are being deployed to respond empathetically; however, these approaches are relatively unsophisticated and best reserved for conduit applications.

Want to learn more about how to effectively integrate language solutions across your patient’s health journey? At GLOBO, we understand the complexity, scale and importance of successfully managing a language support program. Request a demo here with one of our language access experts.