Podcast: Will AI Assistants Make Your Doctor Better?

In this episode, we explore how AI co-pilots are equipping doctors with powerful tools to enhance decision-making and patient care and discuss how it could impact you - a potential patient - in the not distant future.

This podcast is sponsored by Mouser Electronics. 

Episode Notes

6:20) - Equipping doctors with AI co-pilots | MIT News | Massachusetts Institute of Technology

This episode was brought to you by Mouser, our favorite place to get electronics parts for any project, whether it be a hobby at home or a prototype for work. Click HERE to learn more about the role of AI in the medical world!

Become a founding reader of our newsletter: http://read.thenextbyte.com/

Transcript

What's going on folks? Welcome back to the Next Byte podcast. And just let me ask you a question real quick. Do you hate going to the doctor? Do you feel like they're just not paying attention to you? Well, you're not alone. Your doctor feels the same way. They gotta do all this work with administrative stuff, with taking notes, and it takes away from the thing that they spent years of their life learning to do, which is taking care of you. But all hope is not lost. There's a startup that thinks they can tackle this with AI. So, buckle up and let's see, are we gonna let AI take the wheel?

Daniel: Let's go!

I'm Daniel, and I'm Farbod. And this is the NextByte Podcast. Every week, we explore interesting and impactful tech and engineering content from Wevolver.com and deliver it to you in bite sized episodes that are easy to understand, regardless of your background. 

Farbod: Alright friends, as you heard today, we're going to be talking about America's favorite topic, the healthcare system. Because who doesn't love talking about medicine, doctor trips, and the absurd cost of healthcare in general. Daniel, when was the last time you went to a doctor?

Daniel: Honestly two weeks ago, but as a specialist because I hurt my foot. Before that, it's been like probably over two years.

Farbod: Yeah, I'm with you. I'm delinquent. We're supposed to go every year, but we don't. Now the last time I went, pretty much every visit breaks down into this form factor. I go in, they ask some questions, they talk, mostly take notes about what I have to say. I go home and then I get some memory of what happened and what we talked about, but most of it slips my mind until the following year. There's like all these, there's like my chart, all these other tools where the doctor's supposed to upload their notes to, but like, they're so splintered across the board, like from my last doctor to this new doctor, and then from doctor to specialist, because I had to see a specialist one time. It's like, there's too many sources of information that I just straight up don't access, right? Super annoying all around. But-

Daniel: Well, and my personal gripe, similar to yours, is like, not just disconnected systems, but I feel like I'm at the doctor's office and I pay for like an hour worth of visit. I feel like I spend like five to 10 minutes with the doctor. And then you spend the other 45 minutes talking about like, oh, like here's my insurance card. Can you scan this? Like, oh, wait here, go talk to this person about this. Go reiterate your health history to this person. She'll fill out a form.

Farbod: It's crazy.

Daniel: I could have done ahead of time. Like, I'm kind of leading to a statistic here, but I feel like three quarters of my time at the doctor's office is spent on not medical stuff and statistics support it. So, in surveys they took of doctors, doctors spend about three-quarters of their time focused on paperwork and billing, not patient care. So, if you can imagine like you spend hundreds of thousands of dollars and go into debt to go to school, medical school, and you do a residency and all this stuff, and then you end up spending three-quarters of your time doing accounting, I would be burnt out and upset too. And basically, the things that these doctors are most uniquely equipped to do well in our society versus like your random Joe Schmo off the street is medical care and they're spending less than 25 percent of their time doing medical care.

Farbod: And what I was going to say is like in a lot of ways the medical world has advanced with technology right but then I think about it as a patient and this like the normal interaction piece it feels like it's stuck in the stone ages, right. And that's where I want to talk about today's sponsor, which is Mouser Electronics. Right? Mouser, we love working with them. They're one of the world's biggest electronics suppliers. And more often than not, they got these really interesting insights because of the partners that they work with. And they've written this article that's all about AI and medicine and what the future of medicine can look like. They talk about diagnostics. They talk about drug discovery, which we've touched on quite a few times at this point on this podcast. They talk about how doctors can use AI for early detection of different illnesses, which is super helpful. I remember reading about how Steve Jobs actually came up with the idea for the Apple Watch because he was frustrated that his cancer wasn't caught earlier on than it actually was. So, there's like all these amazing things that can potentially come from the integration of AI into the medical world, and I'm going to be linking in the show notes. But one of the ones that is super relevant to the pain points that we just talked about for the past two minutes is actually just patient care, right? Like it doesn't sound super elaborate. It doesn't sound super technical, but it would resolve a massive pain point. And for all the reasons you just listed, right? You have people who spent a good chunk of their adult life becoming doctors, doing non-doctor work. Why are they doing accountant work? Why are they logging into all these different patient portals to put in information? Like, how does that even make sense?

Daniel: Well, and that's what I was going to say is like, dude, you talked about the six ways. Mouser just mentioned the six ways AI is revolutionizing medicine. I'd say this is number seven. Like in addition to diagnostics, drug discovery, robo-surgery, all the stuff you mentioned. This feels like an area that is ripe for innovation because it's clearly not as good as it could be right now.

Farbod: And it's frustrating for both people, both parties. I'm frustrated as your patient because you're sitting there taking notes and not paying attention to me and maybe missing stuff because you're so caught up in getting all the administrative side of things figured out. And then I forget, I can forget stuff. And then you're frustrated because you're spending, and by “you” I mean the doctor, you're frustrated because you're wasting your time as well. And you know, you said it best. It's ripe for disruption. And that's exactly what's happening here with this company called, I don't know if it's Ambiance or Ambience. I think it's Ambience. That's kind of what makes sense to me.

Daniel: Yeah, I'll stick with you.

Farbod: So yeah, let's just stick with that. And Ambience has a fun story. You know how we like background stories and whatnot. With the two founders, we got Mike and we got Nikhil. And these two guys actually met up back in MIT and they both have like stories which directly relate them to the product that they developed. So, Mike was misdiagnosed for a back injury and put on the wrong plan. So that is something which would definitely inspire me to better dig into the system and figure out what's going on. Looks like the conclusion that Mike reached was that the doctor just wasn't able to give him the right kind of attention because they were caught up in other stuff. Nikhil on the other hand is an immigrant who saw his parents navigate through the complex healthcare system of the United States while he was having issues as a child and saw the inefficiency that we've just been talking about this entire episode. So that inspired him to tackle it. And when these two guys met up at MIT's entrepreneurship courses, that there was that spark, there was that connection, and here they are with this company called Ambience, whose main job is to alleviate those pain points. How do we integrate AI in a form that can allow both patients and doctors to make better use of their time?

Daniel: And I think it's even more important to mention. One, because I've got a soft spot for two guys meeting in college, finding a kindred spirit in entrepreneurship and similar interests and going out and starting something in the real world. For folks that aren't watching this, that are listening to this, you can't see the stupid grin on my face because that's me cluing you into that's Farbod and my story as well. But I also want to mention these two guys launched an AI health care company before this one called Remedy AI and they were focusing on like, hey, let's use AI for diagnostics, for remedy selection, like helping doctors diagnose the correct conditions and helping doctors select the correct medicines and or treatments. And that is a perfectly successful company all on its own. But in 2020, they're like, hey, there's still this portion of healthcare field that hasn't yet been automated and like around 2020 is when they founded Ambience, I think. And this is around the same time as like seeing the massive propagation of large language models, LLMs into common life around us. And obviously they've continued to expand since then, but if you're sitting there as an AI expert in 2020 and you're seeing large language models come about and you've got the insight that, you know, Doctors are spending 75% of their time on paperwork instead of doing doctor stuff. And we've got these large language models that are experts, um, in understanding language and then translating that into actions. Like I completely understand and appreciate that they take this as an opportunity to launch a new company called Ambience and the whole job of ambience is to help automate administrative work for doctors. And I think. We talked about the 75% figure a lot, like doctors spend too much time on paperwork and billing. I would say like your average large doctor's office where they've got a bunch of assistants and they've got an admin person up front, they probably spend less than 75% of their time on paperwork and billing, but they said, one of the bigger issues is in rural areas where there's smaller medical practices, like think about like a one- or two-person family practice that is the doctor for everyone in the town. Those are the folks who are spending even more than 75% of their time on admin paperwork and billing. And they either don't have the people nearby, like truly from a population density perspective to warrant hiring someone to help with administrative tasks. And so, they're saying that's a big barrier to getting sufficient medical care, especially in rural areas as well. So, this is, we're going to talk a little bit about their AI tool in a second, but just think about this as something that could massively impact everyone's experience with the health care process, but especially help level the playing field to get like state-of-the-art level care, right? Additional doctor attention toward patients in rural areas where right now economically that's not feasible.

Farbod: Yeah, definitely. And I just want to quickly point something out. You mentioned that this company was founded around 2020. That's usually when we saw these big LLMs taking off and being made available to the general public. Well, that's not like by accident either. One of the key pieces that enabled the LLMs that we use today is this transformer architecture, which was, I believe, found by a Google researcher. And that Google researcher just so happened to be one of the investors in this company. So, there was like a-

Daniel: In both Remedy and Ambience.

Farbod: Exactly. So, there was kind of like, you know, some home, was it Home Field Advantage? Yeah. Right? Yeah, Home Field Advantage there. And I tried to go as deep as I could into this transformer architecture to give insight on the sauce and what made it amazing in comparison to other stuff. Unfortunately, it's even a little too much for me, but I'll do my best on delivering the why this is so impressive to the best that I can. So usually, people were using recurrent neural networks. And this is a neural network that was extracting information sequentially. So, when I gave it a sentence, like, hey, can you please tell me the five most common fast-food options that Americans choose on a yearly basis, it would be able to take that sentence and try to relate the words to each other to understand what it is that I'm asking for, which is great. But the downside starts to happen when the sentences get bigger or there's actually multiple sentences and your ask is, let's say, a paragraph long because there's a decaying effect from the first word that you say to the last thing that you say, it stops being able to have a meaningful relationship in terms of what you're actually asking for. And that meant that you had to break the request down into smaller and smaller chunks. Well, what transformer architecture did is instead of trying to connect a word to the one that's in front of it or the one that's before it. It takes a word and tries to find its relationship to the entire context, the entire text that it's given. And that's what allows you to go into chatGPT and write an entire paragraph and for it to give you a pretty decent answer because it understands the context of what you're asking and how every element is associated with the bigger picture.

Daniel: No, I think you nailed it, dude. And I was also having trouble with transformer architecture. So, I did what a smart researcher does. And I asked an LLM to explain itself to me. And I appreciated the ELI5, explain like I'm five definition it gave me. So, I'm gonna share that as well. I think it sums up what you just shared. A transformer is a type of AI that helps computers understand language. It reads an entire sentence or paragraph at once instead of word by word, which is what you mentioned. So instead of going word one, word two, word three, word four to try and understand a sentence, it uses what it's called self-attention. So, it figures out which words in a sentence are related to each other. The example sentence that chat GPT gave me was the, for the sample sentence, she found her cat. A transformer AI model would know that “her” refers to “she”, and it wouldn't spend its time trying to understand what her and she are. It just looks at found and cat after that. So instead of reading, she found her cat, it goes, oh, she and her are related. So, it's about something that someone owns and it's a she. And then you can see found and cat. So, it actually does three steps there instead of four to understand a sentence. And imagine that economy of scale, like you're saying over large paragraphs. So, this chatGPT told me, transformers are way faster at tasks like translation or summarizing because they can see how all parts of a piece of text connect to one another. I liked your explanation more, but I think I understood more of that ELI5 explanation, if that makes sense.

Farbod: I was gonna say, I liked your explanation more because it really hit the mark there. And there's a key piece of what you just said, which is it's a lot more efficient because it reduces the number of steps that it takes. Essentially in parallel, it's able to process that she and her are one thing, and now let's just focus on the found on the cat instead of going sequentially. That unlocked a very big thing, which led to some shortages that you and I personally felt in GPUs. Because GPUs are very good doing tasks in parallel, whereas RNNs were sequential before. So now that they switched over to this transformer architecture, they could fully utilize the power of these GPUs, which then led to the GPU shortage and now the GPU craze with the Nvidia stocks skyrocketing over the past two or three years. So, some context there, I guess.

Daniel: Yeah, and then now let's talk about how this relates to healthcare, right? So, this company, Ambience Healthcare, start up from these MIT colleagues slash students. They're using AI to automate routine tasks that doctors and other healthcare administration professionals do before, during, and after patient visits. They say it's like copilot, but for doctors. And so, some of the main features here automatically records patient visits. So instead of the doctor spending all their time taking notes like you're saying Farbod or when I last time I went to the doctor, the doctor had an assistant in there with him. It was like medical scribing as fast as he possibly could. Does insurance coding. So picks the correct billing codes, follows all the complex rules, understands the documentation of which billing code to choose for which condition with which insurance provider picks the right billing codes the first time. So, there's not billing issues, not back and forth with the insurance companies. Also does after visit summaries. So, it gives you the notes, Farbod again, like you're saying. It's hard to find notes from your doctor. Sometimes you can't understand them. Sometimes they're not easy to access. They're saying it can send really, really clear summaries to the patients. Big bonus there, they are able to translate it into multiple languages pretty reliably as well. And then, I would say the last big notable point here is it's already used in 40 plus health systems across 100 different medical domains. So, it's not like this is a fledgling idea with a proof of concept. This is a real deal startup with already real-life adoption in the real world.

Farbod: And just to add on to that, based on the feedback they've already gotten, it's saving about two to three hours a day for every doctor that's leveraging this tool, which is a lot considering, let's say, an average eight-hour workday, right? It's incredibly impressive. But Daniel, you always say it. We don't always wanna be cheerleaders. We wanna play devil's advocate a little bit. And there was a post I saw this week from one of my favorite AI commentators on LinkedIn, and that would be Professor Missy Cummings from our alma mater, George Mason University. She has the hottest takes on AI. I love watching her feedback on the most recent trends because this is a domain where everyone's super excited and looking to the future, and she takes a bit more of a, I would say, realistic and cautious approach.

Daniel: Yeah.

Farbod: So, she shared this post where an oncology department that was using an AI tool for transcription was just straight up hallucinating things that were not even said. Now, when I think about critical applications of AI and places where there should be a lot of control and quality assessments and whatnot. I would say oncology departments or Medicare in general, medical care in general would be one of them. Not everything's going to be perfect right at the bat. Tools have their own ups and downs. So, you know, you have to make adjustments. But I think it's worth noting that although the promise is pretty exciting, that these tools before being deployed at scale do need to be monitored and tweaked accordingly. And I think even one of the founders said that, they were like, we were slow with the release because we've been trying to fine tune it as much as we can, which is great to hear, but I just wanted to put that out there.

Daniel: Yeah, no, I always appreciate some wisdom. Always appreciate some wisdom from Professor Cummings because she's one of our favorite, I don't know, thought leaders in this space, let's say. I also want to mention episode, I think it was 192, where we interviewed Karsten Rocher, department head at Fraunhofer Institute, basically AI safety expert, leading AI safety expert in Europe. One of the things he told us is like, hey, I'm not scared of applying AI in a realm where the consequences are small and there's not any potential impact on people's safety. But when there starts to be impacts on people's safety, he's not saying stop AI development, he's actually saying quite the opposite, focus this area for AI development, make sure that it's safe, make sure that it's reliable and keep humans in the loop until you're sure that it's a hundred percent reliable. So, I think we can take some guidance from both Missy there, like a cautionary tale, like, oh, if you turn this thing loose and trust it blindly, you're going to be messing up, but I would also take as a counterpoint, some guidance from Karsten as well, which is like, you know, maybe doctors should use this co-pilot is a good term. Doctors use it as a co-pilot, but it shouldn't let go of the, let go of the wheel, let go of the yoke and let, let AI fly the plane on its own. Right. It's got to trust and check and verify everything that it does use it as a tool, not use it as a replacement for their own efforts.

Farbod: Yeah. AI take the wheel, right.

Daniel: Interestingly, I had a boss who is aggressively anti-religion and instead of saying Jesus, take the wheel, he'd always say universe, take the wheel.

Farbod: Maybe that's where we're headed. Who knows? But I think this is a good place to start wrapping it up. What do you think?

Daniel: Yeah, let's do it.

Farbod: All right folks Have you ever gone to a doctor's office and just been frustrated due to the fact that they might not really be paying attention to you, you know, they got their head down. They might be doing administrative tasks stuff like that. Well, you can't really blame him apparently in America a lot of doctors, they're spending somewhere between 70 to 75 percent of their time on just administrative tasks between taking notes about the meeting they had with you, the paperwork on the back end, insurance stuff. It really takes up a lot of their time and they don't love it either because they went through all that effort to become doctors and care for their patients. Well, AI might just save the day here. We got this cool new company. It's a startup called Ambience that's built on top of the same technology that you use on your favorite LLM. Think BART, think chatGPT. With the goal of most importantly, transcribing what you're talking about with your doctor so that your doctor doesn't have to do that. Giving you both summary reports of what went on. It can translate to different languages pretty accurately and all the other backend stuff that I just talked about. The administrative stuff, the insurance stuff, it handles that too. What does that mean? It means your doctor can now just focus on you and hopefully that means better healthcare for the average American. So, let's get excited for this.

Daniel: I love it.

Farbod: I'm already on board. I want it. I hope that next two years or so, my doctor can start using it so that I'm not so avoidant of going to see them anymore.

Daniel: No, I'm with you dude. And I will tell you just another note. Since we're on it, talking about people we love from George Mason University, a mentor of mine, Paul Singh, dude is like a serial entrepreneur investor in the tech space, CS grad from George Mason University. Dude told me like two years ago, he's like, I see this insight where like, I think he was hacking on like some natural language processing tool, NLP tool to do like Google trends, but for podcasts. I think I texted you about this at some point.

Farbod: Probably, it sounds familiar.

Daniel: And he's like, oh, like I can translate live conversation to text very, very reliably. And he went from like trying to hack around and like make like a cool tool for podcasters, like Google alerts to acquiring an electronic health records company focused on physical therapy and raising like double digit millions of dollars to apply AI in this space. And it makes me think very, very similar to this, right? They mentioned that this tool, Ambience can interact and directly integrate with health record systems, EHRs, which is what my friend Paul's working on. So interesting time to be an AI in healthcare and hopefully they're cleaning all the red tape up because it's awful. The state of the union right now is not good. So, I'm looking forward to this, man.

Farbod: Yeah, man, me too. Folks, thank you so much for listening. And as always, we'll catch you in the next one.

Daniel: Peace.

As always, you can find these and other interesting & impactful engineering articles on Wevolver.com.

To learn more about this show, please visit our shows page. By following the page, you will get automatic updates by email when a new show is published. Be sure to give us a follow and review on Apple podcasts, Spotify, and most of your favorite podcast platforms!

--

The Next Byte: We're two engineers on a mission to simplify complex science & technology, making it easy to understand. In each episode of our show, we dive into world-changing tech (such as AI, robotics, 3D printing, IoT, & much more), all while keeping it entertaining & engaging along the way.