Code & Cure
Decoding health in the age of AI
Hosted by an AI researcher and a medical doctor, this podcast unpacks how artificial intelligence and emerging technologies are transforming how we understand, measure, and care for our bodies and minds.
Each episode unpacks a real-world topic to ask not just what’s new, but what’s true—and what’s at stake as healthcare becomes increasingly data-driven.
If you're curious about how health tech really works—and what it means for your body, your choices, and your future—this podcast is for you.
We’re here to explore ideas—not to diagnose or treat. This podcast doesn’t provide medical advice.
Code & Cure
#19 - AI That Tames Your Health Data Deluge
What if your health data spoke in one calm voice instead of twenty buzzing ones? In this episode, we explore an AI “interpreter layer” that turns step counts, sleep stages, and alerts into fewer, smarter signals that nudge real behavior—without the anxiety spiral. Vasanth (AI researcher and cognitive scientist) and Laura (emergency physician) bring lab insight and frontline reality to a problem most dashboards ignore: humans have limited working memory, serial attention, and a knack for missing rare but important events. More data isn’t always better; often, it’s just louder.
So what does “useful” look like? Clear summaries in plain language. Patterns stitched across streams—workouts linked to calmer moods, dinner timing tied to glucose swings. Personal baselines that ditch one-size-fits-all thresholds. Instead of a raw feed, imagine a tight weekly brief that surfaces the top two trends, why they matter, and one small experiment to try—aligned with your clinician. That’s the shift from charts to choices.
Trust and safety stay center stage. We unpack sensor accuracy, false arrhythmia flags, and the risk of AI hallucinations. The answer isn’t blind automation; it’s human-in-the-loop oversight, transparent provenance, and user controls to set goals, define “normal,” and mute the rest. We also show how primary care can ingest concise, standardized summaries instead of five pages of logs—making visits more focused and collaborative.
If you’re ready to trade a 24/7 body ticker for meaningful insights you can act on, this conversation offers a realistic blueprint. Subscribe, share with a friend drowning in metrics, and leave a review telling us the one metric you actually use—and the one you’d happily hide.
Reference:
Do we need AI guardians to protect us from health information overload?
Arjun Mahajan and Stephen Gilbert
npj Digital Medicine (2025)
Credits:
Theme music: Nowhere Land, Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 4.0
https://creativecommons.org/licenses/by/4.0/
We've never had more health data at our fingertips. So why do we still struggle to feel healthier? Today we explore whether AI can turn numbers into knowledge.
SPEAKER_01:Hello and welcome to Code and Cure, where we decode health in the age of AI. My name is Vasant Sarathi, and I'm an AI researcher and cognitive scientist, and I'm here with Laura Hagopian. I'm an emergency medicine physician. Alright, so we found this article on um potentially having AI help us sort through all the data that people collect on healthcare stuff, right?
SPEAKER_00:Or just like well, wellness and well-being stuff. Like the stuff that you track every day or may not track every day. Some people don't. I do. I mean, do you have a do you have a device like a Fitbit or something?
SPEAKER_01:Oh, absolutely. I mean, I love I love technology in all its forms. And um, being an AI researcher, data is sort of a really critical part of everything. And I'm kind of excited about the fact that we can track all these things. Like I I went for a run this morning, I was tracking, you know, the usual stuff, heart rate and so on, but also it was tracking all kinds of other things, right? There's an ECG thing, there's uh um Pulse Ox, all kinds of things on my smartwatch. And and I tend to track all of that. And by tracking, I mean I see it all in my screen and I get very excited to see looking at the graphs and pictures.
SPEAKER_00:And that's it. Like there's there's so much out there that we can track at this point on our own with these devices that we can purchase, right? Whether it's your heart rate, your steps, your uh, you know, active zone minutes. Um, a lot of people use continuous glucose monitors and check their, you know, blood glucose levels. There's just a ton out there for us to track. And so then you get all this data and you're like, well, what do you do? What do you do with it? What do you do with it when you have all this data? What do you do?
SPEAKER_01:Well, I was actually going to ask you the question right back. I was like, you know, in your practice, do you see people coming in and saying, I need, you know, I saw this in my watch told me this and this, or I saw this information here, you know. Well, that must be, I mean, the information that they get is pretty scattered. And that's how I feel, at least when I look at this, is what do I do with the data? And I'm sure that going to my doctor and talking about it is a way to just like offload some of this burden onto the doctor. Of course, doctor doesn't have all the context, but that said, well, sure, that must be difficult too.
SPEAKER_00:It's an interesting question, right? Because it imagine you show up. There's lots of scenarios, but imagine you show up to the ER and you're like, hey, my smartwatch said I have an arrhythmia. And then they have to figure out, did you actually have, did you actually go into atrial fibrillation, for example? You may have, you may not have, right? It's not like uh you know, the the most accurate that kind of thing. But then on the flip side, you could bring in your data to your, you know, primary care provider also. And now you have what, like sheets and sheets worth of data for them to go, oh, this is how many steps I took, and this is what my glucose level was, and here's what my worsting heart rate has been. It's like a lot to parse through. And a lot of times, you know, your visits with a PCP or a specialist, 15 minutes, 30 if you're lucky. So this is that's a lot of data to try to parse through. Yeah, yeah.
SPEAKER_01:And it it's one thing, the one thing good about all this data, which I like a lot, is that it's data in the wild. So it's collecting, it's not artificial in a lab or in a clinic. It is data that about your health as you live your life. And I love that aspect of it.
SPEAKER_00:Okay, but I have a question for you. Have you ever had data come back that you're like, I don't know if I believe that? Because I have on my on my um on my Fitbed, I've had data come back that was like, you slept great tonight. You got eight hours of sleep. And I was like, no, I was up seven times with the children. There is no way that I slept well this evening. Um maybe you detected, I don't know, me being still for a long period of time, but that does not mean that I am well rested.
SPEAKER_01:Yeah. I mean, some of the data definitely is is is uh probably, you know, so some of it is wrong, but some of it is also just catching things that maybe are at a level of granularity that you don't really care about.
SPEAKER_00:Yeah, I mean, it yes, exactly. And then the question is like, how do you aggregate it? How do you learn from it? How do you notice patterns from it? Because it's one thing to track something, and it's another thing to like change your behavior because of it. Like, say you notice, oh geez, like I only walk 4,000 steps a day. I I really do want to walk more than that. I'm gonna make sure I take a 30-minute walk. That's something that's like actionable. You're gonna do something about it. Whereas if you're like the baby woke me up and my my sleep was crappy all week because the baby was sick. What are you gonna do about that? I mean, I've definitely for the baby to get better, I guess.
SPEAKER_01:Well, I've definitely done this thing where at the end of the night I'm under my step goal and I do a bunch of jumping jacks or walk around the house just to get the steps there, which is in some ways that is at least actionable, right? I've done something with that piece of information, but that was one metric.
SPEAKER_00:It was so many that people are tracking now, right? And so it's like to some extent, there's information overload there. Like there's just so much information there. And you're like, what do I, what do I do with it? Yeah. And have you ever felt like overwhelmed by all of it? Or do you like, do you look at it all, or do you ever ignore some of it? I mean, there's good evidence that like tracking can help you make a behavior change, but like, is there such a thing as like tracking too much? Do you ever get like tired of it or anxious from it? I mean, I think you you you can.
SPEAKER_01:Yeah. And I think there's a lot of research also on the cognitive burden that one experiences and generally cognitive overload. I mean, I think most of us have heard of this notion of having a working memory and having the idea that you can keep three or four things in that working memory at any given moment. So not 17? No, not 17. So, first of all, so you get when you get dozens of metrics that you receive simultaneously, you're not going to be able to keep all of that in your brain, in your memory, and some of it is going to get dropped, right? That's one issue. Um, another issue is that your attention uh is somewhat serial and not all that parallel. So um, you know, there's a What does that mean in uh plain English, please? No, I I mean I think that one of the issues is that with attention, you're doing a thing and you're attending to it, but your brain kind of blocks out everything else. So this is there's that famous uh experiment that um you can look it up, and and maybe we can link it up in the show notes. But it's um, you know, people are passing a basketball around and uh and you're supposed to count the number of passes. And while you're doing the task, other things happen in the scene and and you don't even catch it. And 40, 50% of the people don't even catch it.
SPEAKER_00:Is this the one where like the gorilla comes in?
SPEAKER_01:Yes, there's a big gorilla that walks. I mean, people are wearing white and they're passing this basketball around, and there's this brown gorilla person in a gorilla suit walking right through that group. And um, about I mean, when I did this test the first time, I definitely did not see the gorilla. I was busy counting the the the the uh the basketball. But the whole point is that when you're attending to something, your brain is very good at filtering out everything else and focusing on that thing. Now, when you have lots of different things, you're gonna not be able to do that. You're gonna focus on a certain aspect of it. And again, it relates to how much attention you can actually place on something. Um, in addition to all of that, you also have just like monitoring things, like sustained monitoring degrades over time. And so um there's been some studies about this too, where people have noticed that you know your accuracy decreases, um, especially when you watch for rare signals. I mean, this is called vigilance decrement. And this shows up in one of the challenges, for instance, in airport security and TSA, where you're looking for the weird thing in the bag, but that doesn't happen very often. And it happens so rare that it's really hard to pay attention the entire time to look and wait for those rare moments. And so that's just another thing here, which is your most of your health data probably is fine, is uninteresting. And your interesting events come up from time to time. And so generally the problem is that we have too many alerts and we tune a bunch out, you know, and we get this alarm, alert fatigue from all of this. And that's also a documented problem. And we start to get desensitized and start ignoring our missing really important things. And really, I think the the broad sort of cognitive science takeaway is that biologically, we're built for a few important signals at a time, not a 24-7 stock ticker of our own body. And I think that's the big challenge here.
SPEAKER_00:So then the question becomes like, can we reduce that data volume using AI, right? Can we have some way to sort of like filter or visualize or consolidate all this information into a format that is more digestible, that is more actionable, that has some insights with it, so that there's not this huge burden on each individual user to basically figure out what their health data means.
SPEAKER_01:Yeah. And I think that that's kind of the point of this paper is that we want to go beyond dashboards, right? Dashboards just show us all these charts and all the data and numbers, but we want to do more. Maybe we want to filter some of that, choose some that you know are important and others that are not. Maybe we want to add some medical context to that whole picture.
SPEAKER_00:Right, exactly. Some sort of clinical oversight that's like, hey, does this matter? Or what can we do differently because of this, or what's the limit that we should put of what's abnormal and versus what's normal for a heart rate or a glucose, for example? Right, right. And that's what this paper looked at. It's like, hey, can we figure out what patterns there are? Can we take that information and make it more user-friendly? Can we develop out key insights or identify patterns? And one example they gave was like, hey, if you have mood scores and sleep data and med adherence and whatnot, you can figure out what patterns there are. And they gave an example of, hey, like maybe your anxiety scores were consistently higher on Sundays over the past month. And so then you have a moment to reflect and be like, hey, why is my anxiety high on Sundays? This is an interesting example, I think, because I'm like, do you actually need an app to tell you that? Do you need AI to tell you that you're anxiety, that you have the Sunday scaries? Or do you just like inherently know that Sundays always feel a little bit crappy?
SPEAKER_01:Yeah, I think my take on that is, yes, I might know inherently, but it is sort of nice to hear that it's in it's validating to hear, uh, to hear that because it's coming from data. Your data actually shows the thing that you intuitively believe already. Um and so there's some I to me, there's some value in that. Um, of course, I think the the question is, what is the uh what is sort of the the different types of things that we're tracking? Like that example is to me seems like a form of um summarization, right, of the data or some kind of pattern that you'd identified. But there's so many other things, so many other functions that we would need because when you have a bunch of data, first of all, they're all different formats, they're all different, they all mean different things. Some of them have jargon. So exp summarizing and succinctly saying something in simple language, that's a valuable piece of information. For sure.
SPEAKER_00:Oh, seeing overlap between things. Yes. Like, oh, hey, um, you know, the days that you worked out, your anxiety scores were less, for example. That could be something that's useful because you're like, hey, that means that if I'm feeling anxious, maybe exercise will help me. We know that's true, anyways. Yeah, yeah, yeah. Um but long-term patterns, right?
SPEAKER_01:It gets at longer-term patterns that you might not be tracking. You might not realize that you might realize something about a particular day or a particular couple of days, but if you did you might not realize that it was a repeated thing that happened over the course of two months. Then you can then you can start to see that. So data helps you with that. But again, bringing that to the surfacing, that sort of detail is potentially useful. Um, you might be able to explore why something is the case. I mean, this you might need a doctor for it, so somebody else to help you with it, but to some degree, you might want to just like take the next step and think about the data.
SPEAKER_00:Um, I mean, they gave an example here where it's like, oh, if you're using a glucose monitor, you know, and your glucose spikes after dinner, there were a few episodes over the last week where this happened. Maybe there's like a dietary change that you might want to make along with your dietitian, for example. Yeah. Um, and so there's a ton of different data that could be used here, whether that's like heart rate activity, um, you know, vital signs, um, you know, mood scores, symptom diaries, um, even clinical notes and documentation, like summarizing that, converting the medical jargon um into something that people can understand. Yes. There's definitely opportunity here because you can take this like very dense information and figure out, you know, what interpretations can be made from it. And that's where the AI can come in and do that. I still don't, I'm still not convinced I need AI to tell me that I have the Sunday scaries. I know you like to see that sort of validated in the data, but um, I think there's definitely some opportunities here because there's so much information. And then you're like, what do I do with it?
SPEAKER_01:Yeah. I mean, to me, the long-term patent patterns and expose and exposing, sort of surfacing those is super interesting because that gets at something that I'm not even tracking on a daily basis. And that also takes me out of a place where I'm worried about spikes and adjustments every day. Like I don't have to, you know, I don't get anxious about that. And and because if there's a larger frame, if there's a larger framing of it, then I it's more calming to realize that wait a minute, it's not you don't have to worry about, you know, your your your your weight that might change every day a little bit. That's not the that's not the frequency at which you should be reviewing this, anyways. Over the long term, is there a general trend? What are the general trends? I think to me that is super interesting.
SPEAKER_00:Um But I want to point out that you just said, like, oh by tracking in the first place, maybe I would get anxious.
SPEAKER_01:Yes.
SPEAKER_00:And that's that's like one of those things where I'm like, well, are we over are we overdoing it? Are we tracking is good in many scenarios, but like there are times where it it could be detrimental.
SPEAKER_01:Um oh, I think what I'm saying is I I I I don't mind the tracking, but I don't want it to be visible to me all the time. I think having this AI layer in between uh is helpful uh because I I I I it would need all that tracking to be able to come up with the patterns in the first place, but I don't want to be the one making those patterns because then I have to deal with both the cognitive burden and sort of the anxiety that comes with reading all the data. And instead, now that's taken off my hands by the AI system, and now I'm presented with something that's either a actionable and is a form of a coach of some sort, or is just you know giving me some insight into my health, you know, my longer term health trends.
SPEAKER_00:Yeah. And and this is where you clearly need some sort of governance and oversight, right? That you need medical expertise, you need someone to go in there and be like, hey, this this is what this pattern means, or like this is the actionable insight that we can take away from this pattern. So I think I mean we talk about human in the loop a lot. It's like you you need to have the clinical context added in. And if there are sort of like next steps from it, that needs to be added in as well.
SPEAKER_01:Yeah, I mean, I think it would be interesting if those patterns that we that it identifies are surfaced and shared with your primary care, for example. Well, what kind of reaction would they have? Your primary care presumably knows you pretty well and knows your history. And if, you know, this one tells me that I, you know, I have my anxiety on Sundays or whatever, is that going to be consistent with my my primary care's view about me? And if not, you know, are they going to raise, you know, could they be a check? Could they be the human in the loop uh in this process? I don't really know where this how this would fit into the healthcare system as a whole, but it seems like you're right. There needs to be some oversight, there needs to be some accountability for for it's I don't know, what it's telling you.
unknown:Right.
SPEAKER_00:Yeah. I mean, say you made a goal with your primary care or maybe with a coach or someone to say, like, hey, I want to take more steps. And my goal is 7,000 steps a day, and I'm gonna use this tracker for it. And then you can bring in basically like you know, you could bring in this is how many steps I got on every day for the last three months, or you could bring in some sort of summary that shows, shows that information in a more digestible way. Yeah. And that's easier than having like, you know, a five-page printout of how many steps you took each day. You want to see like the summary, the visualization, the average steps per day. Um, if there were a couple of days where you didn't get a lot of steps, maybe maybe there's some sort of pattern to be identified there that you could work on. But I I think this could definitely be used as a tool there.
SPEAKER_01:Yeah. And there's obviously questions about accuracy and as these you know AI systems tend to hallucinate. So there's an issue about that. Like, are these patterns truly patterns? Um, or you know, are they not? And so you have those issues of trust and oversight and those kinds of issues. So that comes up. So I think it's worth thinking about do we want these things, right? Do we want an AI guardian? Do we want this AI layer as I I sort of conceive of it?
SPEAKER_00:Yeah, I think if there's human oversight, it could definitely help with the information overload, but you need to know like what was filtered, why was it filtered, what were the settings put on it. Um and you'd want people to be able to understand it within that context.
SPEAKER_01:Yeah, and some degree of user controllability.
SPEAKER_00:Like you want to have some dials where the user tells you this is the kind of things I want right now, or this is what my baseline heart rate is, and it might be different than someone else's, for example, or like this is what my step count goal is and it might be different from someone else's. Yeah, right, right, exactly. Yeah. So I I I think there's definitely a lot of promise here. Um and I think there's a problem here too, to be solved, right? Is that we have so much data at our fingertips and it's like too much at this point because you can track basically anything and everything. And it's like, you know, it could be all in one app, it could be fragmented, but it's like, well, who's going in and like looking at every single piece of data and deciding what to do with it? Like, probably nobody. I know I I don't. I yeah, like the only thing I use on my fitness tracker is like active minutes and steps. That's it.
SPEAKER_01:Yeah. I mean, not to sound cynical, but the companies that uh like Fitbit and so on that are making these things are you know are looking at the data and giving you uh adding features in their apps to to help you with supporting some of this stuff, right? So they're looking at it directly and putting those in. But again, I think having uh more user controllability and not just have the companies decide what's relevant to you, but sort of you deciding what's relevant to you is also really important, I think, in this regard.
SPEAKER_00:Yeah, and I think it could if patterns come out lead to um and you and you like know what to do with those patterns. And actionable it could lead to you changing your behavior, and that's that could potentially be a good thing, right? If you uh move in the correct direction, if the insight is like appropriate and actionable.
SPEAKER_01:Yeah. I mean, there's obviously so much more to talk about this issue. I think we can we can end it there, but it's there's so much here and so many layers here. So if um, you know, we can definitely get to it in in future podcasts. Uh, but it's very exciting to see that there is now a possibility that we can actually make sense of all this data.
SPEAKER_00:Absolutely. I definitely agree with that. And I think it could really help people get over that information overload if there was a way to really like filter and contextualize all the information that they're receiving on such a regular basis. And it could help providers too, right? Yeah. Because there's so much to filter and parse through that the raw data is like just too much. You need some way to like summarize and interpret and filter and consolidate it in a way that you can actually have actionable insights from it. Yeah. So we will end there and we will see you next time on Coding Cure.
SPEAKER_01:Thank you for joining us.
