Care Culture Talks · Episode

Inside the ER. How a US doctor sees the rise of AI in emergency care.

Guest:Dr. Aurelio Muzaurieta

HostEyal Heldenberg

Duration11:41

PublishedJune 23, 2026

Listen and watch

SUMMARY

Episode Summary

The conversation explores the cautious integration of technology, particularly AI, into healthcare, emphasizing the need for human oversight and the challenges posed by the complexity of patient interactions.

Takeaways

Key takeaways

AI adoption in healthcare must be calibrated against clinical risk, not commercial momentum. First do no harm applies to technology implementation as much as to treatment decisions.
AI hallucinations in clinical documentation are real, documented and currently fall under physician liability, making review of AI-generated notes a non-negotiable step.
Ambient scribe technology improves the relational quality of clinical encounters by freeing providers from note-taking, but the efficiency gains depend on note accuracy.
The value proposition of AI in clinical documentation is currently undermined in cases where note editing time exceeds note-writing time.
Human oversight of AI in emergency care is not a transitional state. It is the appropriate standard until the evidence base for AI accuracy in high-stakes clinical settings is established.

Quotes

Key quotes

First do no harm is a central tenant in medicine. The implementation of new technologies into healthcare must inherently be more conservative than in other industries.

Hallucinations occur. A surgery that did not happen in a patient comes up in the chart as something that happened and you have to read everything carefully to ensure accuracy.

The ambient technology allows me to spend more time face to face looking at the patient and actually hearing them. The problem is that now there is another task on the physician to make sure what the AI creates is true.

First do no harm is a central tenant in medicine. The implementation of new technologies into healthcare must inherently be more conservative than in other industries.

Hallucinations occur. A surgery that did not happen in a patient comes up in the chart as something that happened and you have to read everything carefully to ensure accuracy.

Chapters

What this episode answers

00:00 Introduction to Care Culture Talks
01:03 Navigating AI Hallucinations and Documentation Accuracy

‍

FAQ

Frequently asked questions

Q1.

Is No Barrier built for healthcare compliance or adapted to it?

There is a meaningful difference between a technology that adapts to healthcare and one that is built for it. No Barrier is clinical-grade and healthcare-native: tuned for clinical language, backed by expert clinical linguists and guided by medical advisors who practice on the frontline. Jeffrey Chen, MD brings emergency medicine perspective from high-diversity urban EDs. Ilan Shapiro, MD brings FQHC and community health experience from AltaMed. Nadav Shimoni, MD brings both clinical judgment and an investor perspective. No Barrier is HIPAA and SOC 2 Type II certified, with a standard BAA, end-to-end encryption and PHI deletion after 7 days. For health systems that need to go further, custom compliance layers are available from the first conversation, including human oversight architecture and Section 1557 alignment. Reach out to discuss your compliance requirements.

Q2.

What makes No Barrier's medical interpretation different from every other platform on the market?

No Barrier is not audio-only. No Barrier retranscribes the full conversation in real time so both provider and patient can read every word on screen as it is spoken. For hearing-impaired patients this changes the encounter entirely. No Barrier also surfaces images and charts for complex medical terminology so providers can illustrate concepts in parallel to the discussion. The interpretation quality standard is identical whether the encounter is a telehealth visit or a face to face consultation, which means a multi-site health system gets the same clinical-grade interpretation across every location, every shift and every modality. At the end of the encounter, No Barrier generates a bilingual discharge letter ready for the EHR, closing the language access loop from first word to follow-up care.

Q3.

How does No Barrier fit into a health system's language access program across every care modality?

Language access breaks down when a health system has one vendor for telehealth interpretation, another for phone and a third for in-person, each with different quality standards and audit trails. No Barrier is one vendor across every modality: telehealth, face to face, video and phone interpretation. The same compliant, auditable medical interpretation standard applies to every patient encounter regardless of how or where care is delivered. For CMIOs and compliance officers building a language access program that holds up across sites, shifts and care settings, that consistency is the operational foundation everything else depends on. No Barrier was selected for the 2026 NACHC/ScaleHealth Accelerator Cohort because it understands what that infrastructure requirement looks like at scale.

Q4.

How long does it take to deploy No Barrier and what does IT need to prepare?

No Barrier is a web app. No hardware procurement, no long onboarding cycle. It runs on devices the organization already has, which means deployment can move faster than any other clinical tool most IT teams have managed. That said, No Barrier's implementation experience across health systems shows that existing infrastructure is not homogenous. Internet connectivity, tablet availability and device coverage across care settings vary significantly between sites and those variables determine whether the platform performs at full capacity from day one. [Reach out to start the conversation.](https://www.nobarrier.ai/contact)

Q5.

What does a health system actually gain operationally when it implements No Barrier?

Language barriers create friction at every level of a health system: appointment delays, staff burnout, patient satisfaction gaps and compliance exposure around care equity. No Barrier's implementation team understands that friction because they have mapped it across hundreds of organizations. The operational results are measurable. Cost reduction of up to 50% depending on the organization. Patient volume up 30% for one provider that reports at Pacific Eye Associates after implementation. And a qualitative shift in staff and patient relationships that Community Clinic NWA described as "regaining the connectivity they had lost" with traditional interpretation tools. Implementing No Barrier is not just a cost decision. It is a decision to let operations flow at every touch point of care. Reach out to map where No Barrier can improve your flow and cost structure of interpretation.

Transcript

Full episode transcript

Aurelio Muzaurieta (00:10)
A central tenant in medicine and part of the Hippocratic oath that we take when we become physicians is first do no harm. And so the implementation of new technologies into healthcare inherently must be more conservative than in other industries where you want to first disrupt and shake up everything and break things. This is much harder in a healthcare setting because we're dealing with people's health.

And I think that that's appropriately that worry and those concerns are appropriately placed. With respect to bringing new technologies like AI into the clinical space, think, well, what's most important is that there continue to be a significant amount of human oversight on the, on the rollout of these, of these technologies and, and,

And I can tell you that in practice there continues to be a need for this human oversight because as the technology is able to help remove some of the administrative tasks, we want to maintain precision and accuracy in that documentation. And AI is still in its nascent state of trying to take, for example, an interaction between a patient and a physician in a noisy emergency department to...

be represented in a few paragraphs what that conversation was and hallucinations occur. Sometimes a surgery or a medical procedure that you think that did not happen in a patient is.

comes up into the chart as something that happened and you have to look at it, read everything carefully and ensure that what the AI has interpreted as your interaction with this patient is indeed accurate. And that still falls on the physician or the medical, medical provider to, ensure is, is true. And I've, and I've fallen, this, this Swiss cheese model, it's, have, I have made mistakes and said, that, was not a surgery that happened.

Then one of my colleagues will remind me so that's something that all medical providers who are starting to adopt these technologies should be aware of. And as...

ambient scribe technology and other tools come into the clinical space. The question always becomes who will take liability for mistakes that surround this and in the current system, it's the physician using it. So while it can help take away some of the impersonal interaction of taking notes, for example, next to a patient on your own,

and not looking them eye to eye. That's the traditional model right now. I love that the ambient technology is allowing me to spend more time face to face looking at the patient and hearing this. The problem here is that now there's another kind of task, I guess, on the physician to make sure that what the AI creates is in fact true.

Sometimes that can actually take more time than doing it yourself. So there's a bit of a value proposition there that needs to continue to be had. And as the technology improves, I think that will get better. I'm pretty optimistic around it, but as it states now, you know, if the AI writes that a surgery happened or that I did a particular part of the physical exam that I didn't do, you know, that falls on the provider, so.

That's one thing that I'm concerned about and people should definitely know is as we take on these technologies that have great potential. But in turn, and the other thing I think is data. And I will say that

what is critical that everybody must do from is informing all the parties involved that recording an AI technology is in fact occurring in this clinical interaction. And I will say in my experience, most patients are totally okay with this. I often start my interactions where I'm gonna be using AI scribe technology.

And the way I kind of approach it with patients is the following. Hello, Mrs. Jones. Good afternoon. It's wonderful to meet you. I see you have your husband here in the room. I use an AI scribe technology that records our conversations and helps me with my documentation so I can focus less on the computer and more on you. And I will say most patients are pretty happy to have the doctor have

more attention focused on them as a patient than on the computer screen, typing away and not really seeing and looking and feeling in that human interaction. So that's really helpful and I find it great.

Eyal Heldenberg (04:06)
Yeah.

Along those lines, one of the things that I think we see is that, as scribes come and basically say, hey, we're gonna help you with operational thing, Documentation. And in the medical interpretation field where we operate, it's kind of an existing thing, right? You have

human interpreter over iPad and now, you know, coming this AI medical interpreter. And sometimes, we think about what if the, the model performs better than human, like, or maybe kind of in a general statement. If we have a better workflow, theoretically, we have a better workflow, let's say whatever English to Spanish translation in real time that let's assume it's better, more accurate, timely and others than existing workflow that is 30, 40 years or so. Because many times I'm thinking like what health systems should, look at the specific existing workflow that maybe is not safe enough or not accurate enough, or maybe there is other way to get it more, more accurate. So
‍

Eyal Heldenberg (05:26)
I think it's a bit different from...

Aurelio Muzaurieta (05:27)
I think this is a great point. I think what's interesting here is we're, I would say the current workflow around medical interpretation in most healthcare systems is safe enough, but not ideal and certainly not optimized to be the best that it can be. And I think

for what it's worth, it's wonderful when we have in-person medical interpreters for particular interactions that are difficult with patients, hard of hearing patients, the elderly can be difficult to use technology. We also have iPads that call in people live. And the quality of these interpretations can really vary depending on the day, the time of day, the clinical setting. And so...

All of the folks, and I'm very thankful for all medical interpreters who take their time, whether it's in the virtual or in-person space, to provide this invaluable service. My excitement around AI interpreting is the possibility of raising the...

quality and raising the bar for many of the interactions that little bit.

less important that it be in person for the day to to relieve the burden on the interpreting services that exist to be more focused on the more difficult cases and that. And when we're looking at AI, I think that there is a lot of data now that's showing that they can outperform what our current workflow is in particular clinical scenarios and settings, as we know, and there's many different types of clinical settings.

Settings, there's ambulatories, clinic settings, there's the emergency setting, there's out in pre-hospital setting. I all of these different opportunities that where patient communication is so key and that's that workflow as it states is, we always, I think we owe it to our patients to be honest with the quality of the interaction that's occurring provide them with the best service that exists at the most reasonable cost for all parties involved. I think that anybody who, who would deny technology that is better from

accuracy and precision perspective than what we currently have to a patient for a less expensive cost would not be doing right by that patient and by the health system. So I'm pretty optimistic about where we're going. I don't think we're 100 % there yet and we have a lot of things to move through, but overall that's what I see as opportunities around improving this interaction with patients with limited English.

Eyal Heldenberg (07:55)
Maybe llast question, your thoughts or projection on 2026 and let's say AI in healthcare in 2026. Where are we going with that? Are we going all in? Still kind of cautious? How do see next year?

Aurelio Muzaurieta (08:10)
Hmm. I think that, well, AI and healthcare in particular, I think is a little bit behind from AI and the rest of the different, workforce. and overall there's a larger, maybe a bubble is soon to be bursting around the excitement around, AI and a lot of sectors.

I think we've barely scratched the surface when it comes to opportunities existing within the healthcare sector and particularly medical services as a subsection of healthcare that we, that opportunities I think are very, very high. One of the reasons is because our healthcare system is extremely convoluted and difficult to navigate. And so as it stands now, any incremental improvement is actually a huge value add in this. So ⁓ I'm optimistic that the companies that are trying to enter

into the healthcare sector to improve workflows, particularly when it comes to the patient-physician or patient-provider interaction, are going to see more and more people see the value of AI and integrating that into their medical practice with their patients. It's happening on the daily basis already.

And one of the things I'm very excited about is the adoption of, we are very lucky at large academic centers that have a ton of resources from research and clinicians and people at the top of their game to have access to

so much knowledge to be able to give our patients, but we forget about all of the other venues that patients receive care that aren't necessarily as well-resourced and what AI has done from open evidence to AI scribes is allowed healthcare systems that aren't as well-resourced or

whether they're public or private and in the smaller kind of clinical scenarios to have access, more easily available access to knowledge and to better patient care. So there's this huge amount of folks that if they don't have access to the top academic centers in the United States can still get better care because of what AI technology has provided for.

That's provided for the physicians and the providers that are working with that technology. So I think that that's already begun. And now it's kind of, because there has been this rollout at Yale, I will say that there's already so much more of a willingness to adopt these technologies and to try them out because so much good has already come out of them.

Eyal Heldenberg (10:46)
Amazing. I'm sharing this optimism for 2026. I'm kind of thinking along those lines where different environments would probably have more, I would say, benefits from implementing those kind of latest and greatest tech to help them to manage the workflow, give a better patient care, and hopefully do some cost saving and cost-effective workflow I would say. So I totally share this optimism. All right, so thank you very much, Aurelio It was really fun to talk to you this time.

Aurelio Muzaurieta (11:22)
Thanks for taking the time. Until next time.

Eyal Heldenberg (11:24)
Thank you.