AI on the Frontlines: Dr. Alexandra Rabotin on Ambient Scribes, Language Equity, and Real-World Workflow Challenges

Dr. Alexandra (Sasha) Rabotin is a family medicine resident at Mission Community Hospital in the San Fernando Valley, California, with a developing focus on clinical informatics. She holds an MD from the Hebrew University of Jerusalem and a Master of Public Health from Tel Aviv University. Her research centers on the evaluation of large language models and ambient scribes in real-world clinical settings, with particular attention to workflows, language equity, and safety-net environments.

In this CHARGE interview, Dr. Rabotin offers a frontline view of ambient AI scribes, highlighting challenges around language equity, interpreter errors, and deployment in safety-net environments. She also reflects on practical clinician guidance and emerging opportunities as scribes expand beyond transcription.

Read the full conversation below.

Q: What first drew you into healthcare AI and, specifically, ambient scribes? Was there a specific moment or patient encounter that made you realize these tools needed closer study?

I was initially fascinated by the rapid advances of ML in pathology and radiology, but as an aspiring family physician, I did not see comparable advancements in my field. Ambient scribes were the first tool to truly attempt to address a daily pain point in primary care.

The spark to study them came when a patient told me a doctor had called her “too fat because of eating avocados.” I immediately wondered: How would an AI scribe document that? Could a simple prompt make the note more inclusive? I realized we couldn’t assume these tools were safe or neutral, and it motivated me to test them.

Q: In a recent study, you showed that a single prompt could meaningfully improve inclusivity in AI-generated notes, underscoring the importance of AI literacy. What practical advice would you give clinicians who are new to ambient scribes and want to use them safely and effectively?

Keep the patient at the center. Know where AI struggles and stay humble about the unknowns. Take ownership of the note and review it intentionally. If something feels “off”, explore it further. These tools are new, and we have a key role in guiding their safe and equitable use.

Q: Your research on ambient scribes in bilingual clinical encounters received the Distinguished Poster Award at AMIA 2025. Can you briefly walk us through what you examined, what you found, and what those findings mean for real-world clinical practice?

We examined whether ambient scribes incorporated interpreter errors into the note, using simulated cases with planted errors. Our findings suggest that current models tend to document the interpreter's error rather than the original speech, primarily when interpreting the patient's words.

For clinical practice, this implies two things. First, we need testing and validation in multilingual settings. Second, and more exciting, we may be able to use these tools as safeguards to flag clinically significant discrepancies.

Q: Given that you train and practice in a safety-net health system, what are your concerns about implementing general-purpose LLMs in these settings?

My primary concern is the linguistic and cultural mismatch in general-purpose models, which may be heightened in safety-net settings. For instance, Zolnoori et al. have shown that automatic speech recognition (ASR) systems have significantly higher transcription error rates for Black patients.

Q: Time savings from scribes seem more modest than early vendor claims. In today’s financial climate, how should organizations realistically think about, define, and measure ROI for ambient scribes?

ROI for ambient scribes shouldn't rely on productivity and burnout metrics alone. A broader cost-effectiveness lens from a societal perspective (including documentation safety, patient outcomes, and physician turnover) will likely help us assess the impact more accurately.

Q: Vendors are now expanding scribes beyond transcription into chart review, coding support, and early decision-support features. Where do you see ambient scribes heading in the next 2-3 years, and what opportunities and risks come with that evolution?

I hope they evolve into active assistants, helping us apply evidence and guidelines directly in the workflow. The most significant opportunity lies in frontline clinicians guiding this evolution, ensuring it serves our patients and doesn't add to burnout. The risks are scaling up harm and worsening existing disparities.