Ronald Rodriguez, MD, PhD, Professor of Medical Education and Program Director, MD/MS in AI Program at The University of Texas Health Science Center at San Antonio, and Niklesh Akula, Fourth-Year Medical Student at Joe R. and Teresa Lozano Long School of Medicine at The University of Texas Health Science Center at San Antonio, talk about getting a dual degree in medicine and AI.
Interview conducted by Ivanhoe Broadcast News in 2024.
This story is incredibly intriguing on several levels, but the first of which you guys are embracing this whole thing. There’s all that talk about, it’s going to replace me. You guys embraced it wholeheartedly. What was behind that decision?
Rodriguez: Several things are behind that. The first was that we recognized, whether we like it or not, it’s going to move forward, and you can either move forward with us driving the technology, or be driven by the technology, and we would much rather be in front of it and making sure it’s used ethically, effectively with patient-centered concepts, and that’s part of it. Also, part of it is that for the most part, it’s being driven by large tech companies and the large tech companies have their own agenda, and their agenda doesn’t align with most health care professionals, and so we have grave concerns about the escalating costs of health care and the potential to escalate the costs with incorporation of AI could lead to a worsening of health care disparities so that some people will have access to this amazing technology and some people will not, and so one of the ways that we felt that we could try to level the playing field is to help drive the technology development. There have been several tech gurus who around 2012, made predictions that 80 percent of physicians would be replaced by AI within 20 years. I think some of them still believe that. They see, if you look at the large picture, that health care costs and services constitute one of the largest components of our GDP, and if you were a business person, being able to take a slice of that is probably one of the most attractive things you could do. The only reason that it hadn’t happened already was because our legislators in their wisdom many years ago, were very concerned about patient confidentiality and they passed multiple laws that protected patient confidentiality as electronic medical records are being developed. These laws like IPA, and because of that, access to that information became very hard for insurance companies and big tech companies, and so that fortunately has led to a delay in the use of health care information in driving the development of AI. But it’s now a wide-open opportunity for them.
Is the whole process analogous to big pharma and academia working together so that academia has some measure of control over what’s getting out there?
Rodriguez: I think it’s for us yet to be defined. This is nascent technology. The use of AI in health care, while it has been around for a while, has been around in a way that was not visible. Only now are we starting to see how it’s being used and the potential for its use in everyday medicine? Several of the large companies that drive the development and use of electronic medical records are incorporating AI technologies into their products. They have a business model in which they charge a certain amount per provider per year, and the amount that they charge is almost as much as they charge just to use the EMR, to begin with. There are these technologies that allow microphones in the clinic’s patient rooms and they can hear what the doctor says and they can hear what the patient says. They can transcribe all of that. They can then form a summary of that conversation in a format that’s compatible with most clinic notes as they’re written and they can provide that for you as you’re walking out of the room, so you can get a summary of the interaction in near real-time. That’s a great idea. From the doctor’s point of view, it often takes 20 or 30 minutes to write, and it can be done in a few seconds. It does take away some of the emphasis from the doctor’s point of view, so the doctor may feel certain aspects of the discussion or the most important and the AI may be more comprehensive and not give equal weight or give equal weight when it’s not really warranted. There are some potential issues with that. It should still be reviewed by the physician who ultimately should be responsible for that content. The cost of that type of interaction could double the overhead costs. Who’s going to have access to that except those who have really good insurance?
The one thing you brought up just a second ago is that AI is an intangible thing for most people. I can think of two cases, either Google or eBay use an AI to write up the summary or robotics in the operating room. What’s really behind this? And how do you tie that in with how you designed your program?
Rodriguez: The actual use cases for AI in medicine are far broader than that. You can use it to schedule patients, figure out who’s going to show up, who isn’t, and how long are they going to take. Are they going to take a 20-minute visit or are they going to take much longer or are they going to be very quick and easy? Are they going to require specialized expertise? Should they be sent to an expert in a certain area? Should they be going to that particular doctor who may not see very many patients in this particular area that patient needs and then reassigned to someone who has the appropriate expertise, all those things can be done behind the scenes in the scheduling with AI so that the clinic is much more efficient and the physician doesn’t spend his time waiting for a patient to show up or using a lot of time for a patient that don’t have the complexity that needs to be seen by that hypo physician?
When you were putting this program together, every story had a big wow moment. What is it for you?
Rodriguez: Well, we started this more than four years ago, and so when we started it, AI was not a particularly prominent term, and we had a fair amount of difficulty getting it started and getting legs for it. We saw the use which at that point had already been pretty evident in biological research. It was being used for gene analysis and clustering and things of that sort. I had been used to identify imaging and pathology, X-rays. But the generative AI, the type of AI that we’re now seeing widely used by everyone across the country and Gemini and ChatGPT, and these similar chatbots. That hadn’t come to the front yet. When that came out in November of 2022, there was an inflection point in the National recognition, and that probably happened around March of 2023. When 60 Minutes did their- their story on these ChatGPT and at that time was called Bard, and I think there was this immediate recognition by the whole community that the potential here was far more than what people realized, and the power of these generative AIs to give insights and summarizations and provide information, not just like a Google search, but actual like a tutor summarizing information. That didn’t come out I think until they started to see the media wave around it. Within a few months, there were millions of users of ChatGPT and billions of visits per day at a scale that had never been seen in the history of the world.
How do you stay on the leading edge? How do you weed through what’s going to happen and what is not?
Rodriguez: That’s not the easiest thing to do. There are some really obvious low-hanging fruit for areas that we should investigate and work in, and we do pursue those. There’s a little bit of this forecasting that we have to do where I try to do my best to identify what I think are the most important things that are going to drive the technology. For me, fundamentally, that’s going to be the financial incentives behind this and I think as I look at it, what I see is a largely tech-driven process. The point of this has gotten so big that Microsoft Google Meta and X have laid off thousands of workers so they can purchase the AI equipment to put them at the forefront of this at the scale of billions of dollars.
Who’s driving this buzz?
Rodriguez: The potential and how it can be used is astounding. They recognize that, and they know that within three or four years, the person that has the best technology is going to be the one that wins. But there is a weakness to the tech companies that they are not so openly admitting but are fundamental. That has to do with the quality of the data that they have. When they skim the Internet for content, and they extract everything they can from it, a very large proportion of that content is wrong and is propaganda, and it’s inaccurate. They have to spend a considerable amount of effort in what’s referred to as preprocessing to eliminate things that are just flat, incorrect misleading vulgar or unethical. There are many things on the Internet that could be used for very malicious purposes, and they have to filter those things out. What they are starving for is high-quality proprietary information. If it’s a company, it may be the research that you’ve done to develop your product, which you never publish, and you don’t expose to the public, so it’s not out there except within the confines of that company because it allows them to function ahead of their competition.
They need to have doctors as advisors to tell them what to do, have you been approached by any of these high-tech companies because they have to hear about what you’re doing here and express some interest about content?
Rodriguez: I get approached probably five or six times a week, and I’m not always certain who was on the other end of it, and whether or not they’re intermediaries to some larger corporate interest. Often, I think they are. I have to be leery about that. I think that’s only going to continue. I do think that they are very open about their need for high-quality information. The type of high-quality information they need is our patient’s records. Those interactions that we have with patients provide valuable feedback to the tests that we do and the interpretations that we make. Being able to predict with higher accuracy requires understanding what you did wrong, as well as what you did right.
What do patients do in the meantime if you go in for a visit? How can a patient protect their information?
Rodriguez: There are laws in place that require that if we’re doing any research, the patient has to be informed of that. They also have to consent to it. If they don’t consent, we’re not permitted to use them in any study. In academic institutions, we’re particularly adherent to that. Academically based research is probably the most protective patient information, but it is also not the only way that companies can get that information. For some reason, if you’re a patient in a health care system, most patients are insistent that information is completely private. If they get a telephone app that is in their mind helping them do something, they often very willingly provide a level of detail that medical professionals often don’t even get. That goes to data brokers.
Is this part of what you’re teaching in this particular curriculum? How about the ethical considerations, so that they’re not disseminating bogus information?
Rodriguez: We have a curriculum that’s based on two campuses. We have courses that are developed and taught at UT Health San Antonio, and there are courses that are developed and taught at UTSA. The UTSA courses are formally more classical AI, and so they teach them the mechanics of machine learning, logistic regressions, neural networks, convoluted neural networks, deep learning, computer vision, autonomous systems, and things like that, but it they have relatively little in the way of clinical interaction. We do the opposite. We focus on the patient information that we have and the clinical problems that we face. Between the two, we try to marry the two, and we provide opportunities for collaboration across the two campuses. I think this fosters collaboration and also promotes the development of new ideas.
Is that part of that embracing that you’re talking about? What about going upstream so that this batch of students then gets recruited by medical schools, are you working upstream with people to continue it?
Rodriguez: We have put in several grants specifically starting at the high school level to try to reach out, with an emphasis on underrepresented minorities and economically disadvantaged groups. We reached out to the CAST-Programs, Cast-Med, Cast-Tech, and Cast-STEM, we’re working with them on a new grant application that’s going in June. We put in a grant application through the National Center for Education Research that went in March. We’re starting- and we’ve developed teams that are focusing on ethics, that’s focusing on how we can use AI in the classroom, how we can teach the students principles of AI and working towards the development of certification programs for them so that we start at the high school level and bring them all the way through. We’ve got some students now who will be entering medical school who came here specifically because we’ve developed this program. We’ve got a robust group of first years who have self-organized their own student interest group and are taking on new projects, largely self-initiated with a little oversight from my staff. There’s a lot of enthusiasm. They see the potential probably better than most of us who are older see it. But they see how important it is, and they recognize that they are at such a pivotal inflection point in time that they could become the leaders of the future of how these technologies are embraced and utilized. And it has such potential for good if it’s led by the right people and such potential for harm if it’s allowed to go to others for malicious purposes or personal profit.
How do you see everything with AI happening?
Akula: I think the best way to put it is if you take a patient who comes in with internal bleeding or other complications, especially if they’re a minority group, they’re not as efficiently diagnosed in the system. I’m not trying to say it as a knock on how we run things and how we do things. There are many well-equipped and capable physicians in health care, but sometimes there are some cracks in the system that need to be patched up, and I think AI can really help with that.
If there is a language barrier or how you explain something as a patient that the doctor doesn’t understand, are you talking about using AI as a translation tool?
Akula: That’s one way of using it. The way I have thought about it up until now is you take the data that we’ve collected across years of delivering medicine and feed the data to a machine learning model. The way it works is that it’s trained to look for the patterns that are innately built into that data and try and figure out exactly what is the connection between all these different factors like blood pressure, age, ethnicity, gender, along with a ton of other data points. The model is able to generate a formula that functions as an equation as to why a person has a diagnosis of XYZ.
You’re talking about extracting data from a data collective, but extracting different data based on the patient?
Akula: Yes. You’re essentially looking for the patterns between the data that aren’t very apparent to us as humans, but a machine learning model is built for catching those nuances in the data very easily.
What has changed for you since you’ve been in this particular program?
Akula: My awareness about AI and medicine has really taken off in the past year and a half. When I started to look into more of these AI applications, it’s gone from basically an AI tool studying an image to different numerical data to now even live analysis of video. You can even use it in the operating room where you have a surgeon going in with a camera and then the AI model can analyze what it sees on the camera feed.
Your approach to technology and medicine compared to somebody who is not in this program and doesn’t have as wide of an understanding. Have you noticed that?
Akula: Sometimes the lack of understanding of AI technologies has been apparent to me when speaking with different professionals and even students in health care. If I’m being honest, the AI specialist will also tell you that there is a black box to how the model works where it takes in data and then starts just basically spitting out different results, and part of the challenge is being able to explain it better for the wider acceptance by the public and political officials as well because this technology would require approval before being deployed to health care providers.
Do you feel like this is going to be a great attribute when you are looking for fellowships, for example?
Akula: I do. For residences and fellowships, it offers new ways to expand research horizons so that we can advance technology and the way we deliver health care. It would be an effective tool towards bridging the gap and delivery of health care for marginalized groups, such as ethnic minorities and low socioeconomic groups, in our society.
You’re going to become an interventional radiologist, how do you envision technology, when you’re all done with all the school?
Akula: I think by the time I start practicing as a physician, that’s going to be about six, seven years down the road. At that time, it would be interesting if you’re able to take a live feed image, feed it to the model, and then it tells you what the diagnosis is and what is the standard approach for the given diagnosis as you do your procedure in the OR. Surgeons are required to type up a post-operation report, and some use voice dictation to write up the report. It’d be interesting if using the findings that the model gets from looking at the image feed, it’s able to generate a report for you automatically, and all you have to do is just go in and then quickly make some edits.
Go into that a little bit more about how it will help the position in the OR.
Akula: Within the OR, I think what’s nice about AI is that it has a way of standardizing all the data that it’s getting. That’s how it’s been trained, and so that’s how we expect it to perform in the future with new patients that come in. Using that standardized method, while also minimizing bias, you want to use your models to analyze any patient that walks in. You can essentially go about delivering health care in a way that’s equitable to all the patients that come through your door and it helps to essentially standardize the results that you get after treatments as well.
Do you like the program? What’s the best part about it?
Akula: It has pushed me in a lot of ways that I didn’t even know that I was capable of achieving, I guess. Primarily, I come from a background that does not involve any coding or Python language. When I started, I was a little nervous about that and I spoke with Dr. Rodriguez. I mentioned my concerns about lacking a computer science background because I was worried how things would play out for me. But he made the effort to work with me and figure out how he could help me bridge that gap and shore up some of those shortcomings that I had, and I was able to get some extra help before I started the program. These preparation steps helped me be more capable for the start of the year.
END OF INTERVIEW
This information is intended for additional research purposes only. It is not to be used as a prescription or advice from Ivanhoe Broadcast News, Inc. or any medical professional interviewed. Ivanhoe Broadcast News, Inc. assumes no responsibility for the depth or accuracy of physician statements. Procedures or medicines apply to different people and medical factors; always consult your physician on medical matters.
If you would like more information, please contact:
Eileen Teves
Sign up for a free weekly e-mail on Medical Breakthroughs called First to Know by clicking here
