This conversation is part of a series of interviews in which JAMA Network editors and expert guests explore issues surrounding the rapidly evolving intersection of artificial intelligence (AI) and medicine.

Throughout her career, Tanzeem Choudhury, PhD, has worked at the intersection of wearable computing, AI, and health care. Over time, she came to see that tracking activity-related behaviors with wearable devices had the potential to support well-being before people became sick. Today she also thinks about how AI tools can be used to efficiently test and deliver new ideas for technology-assisted well-being in the real world.

Choudhury, who holds the Roger and Joelle Burnell Chair in Integrated Health and Technology at Cornell Tech and is the director of the Health Tech Hub at Cornell Tech's Jacobs Institute, got her start at the Massachusetts Institute of Technology (MIT) Media Lab before going on to cofound multiple health tech start-ups.

Harvard psychiatry professor Roy Perlis, MD, MSc, editor in chief of the newly launched JAMA+ AI, director of the Center for Quantitative Health at Massachusetts General Hospital, and an associate editor at JAMA Network Open, spoke with Choudhury about her early work on wearables and how she thinks researchers, entrepreneurs, and clinicians can work together to avoid the "valleys of death" that keep useful technology out of the hands of physicians and patients.

This interview has been edited for clarity and length.

Dr Perlis:I want to go way back to your time at the MIT Media Lab. How did you come to start working on health in the first place?

Dr Choudhury:It's like my academic childhood, right? My experience at MIT has deeply influenced my work in 2 ways particularly.

I wouldn't call MIT an interdisciplinary place. It was almost like an antidisciplinary place. It's a place where you tackled problems with innovative solutions and really didn't care about interdisciplinary boundaries. That has deeply shaped the work that I did at the Media Lab and continue to do, in thinking about the problems and research I want to tackle and how I bring ideas from different disciplines to solve those problems without being too focused on exactly which discipline I fall under.

The other way that Media Lab has deeply influenced me is in thinking about sharing ideas and communicating broadly to different individuals and different communities from the commercial sector to the research sector -- people who work in technology and people who know nothing about technology, but care about the problems. That really shaped my entry into health and tech because the ability to think about problems from different perspectives and communicate solutions in a way that is meaningful to people who are using the technology has been very important throughout.

Dr Perlis:You work across disciplines. There are a lot of things you could do. How'd you land on health care?

Dr Choudhury:I didn't start in health care. In the Media Lab when I did my thesis with some of the early wearables, I was always deeply interested in how people behave, their different personalities, and how they interact with each other. I came to the US from Bangladesh, and there is so much richness in who we are and our interaction and how it impacts our lives. So, that's where I started -- bringing together psychology and technology.

One of the things that drew me to health care is [the realization that] we think about the role our behaviors and interactions play in our health way too late -- when we're sick. I saw the potential for how we can go back a little bit before someone is sick to understand how we can think about the information that our lives provide in the outcomes of our health and how to design technology that could support well-being rather than only coming to solve a person's illness problems.

Dr Perlis:Give us an example of what specifically you were building at the Media Lab.

Dr Choudhury:I started working on wearables before it was a common term. It was basically a bundle of sensors that were packaged together that people could carry around and wear. My work was one of the earliest that looked at how you can take these sensors and measure what people do in the wild -- or in the real world outside of lab settings -- and how it could be used to understand the dynamics of people better.

I was part of a group called Human Dynamics, and our bundle of sensors would track your movement using accelerometers and gyroscopes, things that would understand how you are not just taking steps, but if there's a little bit of oomph to your step -- if you are more or less energetic.

My work looked at how we can robustly measure that as people go about their daily lives, not in the lab where you take 10 steps.

It was first, "Can we build technologies robust enough to track our behavior?" and then, "When we come together, how do we influence each other's behavior?" My thesis looked at measuring the physical world social networks and understanding if you and I are talking, are you having more of an influence on my style of talking or am I influencing your style of talking? One of the things that we looked at was how people, when they come together, influence each other's style and behavior and how that actually corresponds to their level of influence in the social network.

It's not only our own behavior that matters when we come together. How we intersect and influence each other's behavior also matters.

Dr Perlis:You were measuring social dynamics long before it was on most people's radar screens. How come that isn't part of routine clinical care now?

Dr Choudhury:This information is very dense, and we need to know how to act on it. This information needs to be digestible -- and not just digestible, but you as a clinician and I as an individual need to be able to take action. So, one of the things that we need to do better is to take this vast amount of information and really tie it to action.

It's also thinking about clinical care and patient-generated data, and how do they merge together to help in clinical decision-making? Finally, of course, it's a matter of how all these things get reimbursed or paid for. We have these islands of solutions that are still somewhat fragmented, but we're in a great place now where we are seeing more integration of multimodal data into our clinical care. We are having new types of payments and reimbursement models, and technologists are distilling this information to make it more actionable. I'm optimistic that now is a great time where we'll see more and more integration into the standard of care.

Dr Perlis:You mentioned collecting data and not necessarily knowing how to present it. How do you think about presenting this kind of information to physicians? What goes through your mind in how to structure what they need to see compared with what patients need to see?

Dr Choudhury:They're definitely different. From a clinician's perspective, it's important to understand how a clinician is making decisions now, what their pain points are, and where they would need more information. I think that has really influenced some of the design of what we built.

In one of our early projects where we collected about a year of data from individuals with schizophrenia, we had all kinds of measurements. When we met with the clinicians, we're like, "Wouldn't you like to see all this data?" They're like, "No, please don't show me all the data because I'm already overwhelmed."

So, what information is key for care? "My patient is trending towards relapse" or "the medication isn't working." Those significant change points are important so that they can be proactive or know when something is not working. That level of design and what's needed comes from understanding what the process is now, what technology can provide, and what is a good intersection. That has always informed the work that we do -- how do you take this vast amount of information and what do you need to distill?

For example, in mental health, we looked at doing some assessment using PHQ-9 [Patient Health Questionaire-9], a mental health assessment tool. People are already capturing this information. So, do we need to automatically predict PHQ? Maybe, but we've also seen that these types of measurements are so diverse and manifest in different ways across different individuals.

It is much harder to predict accurately someone's PHQ score using sensors than it is to predict changes. I can say someone's mental health or PHQ level is significantly deteriorating or improving. That is a good intersection of what clinicians can use and what technology can robustly provide. We need to do more of that instead of always saying, "I want to get the best, most precise prediction of a score."

Dr Perlis:I'm always puzzled by the fact that we want to build things that will estimate what we can get directly by asking our patients. I like the point that you made about being better able to predict change because I can ask someone how depressed they are.

When you started collecting data with people with schizophrenia or other serious mental illness, was it a struggle to convince them that this was worth doing? Did people want to use these devices?

Dr Choudhury:What helped us move the research and technology forward was not only meeting clinicians who might act on this information, but also the patients. Understanding where the patients were struggling and what their concerns were was just as important.

When we did this work, we actually went in a clinical setting and met with individuals with schizophrenia. One of the things that we realized with serious mental illness, which is a lifelong illness, is that individuals are trying to live in the real world, manage their symptoms, and have a better quality of life. Figuring out how the technology can improve their life was very important. It could be just identifying that someone, over time, has symptoms flaring up and maybe they need attention. It could be suggesting some simple change in someone's behavior that could help them manage their symptoms. Or just that a medication isn't working because the effect is wearing off.

Understanding where the need is and designing technology that can support those needs has been very important. Then, you really can design something that is acceptable.

Where people feel the most resistance is: you give me data, I promise something good will happen in the future. That is where we as technologists can do a better job. As we do research and clinical trials, we are often trying to carve out a very clean experiment design and how we build things. But the challenge is that once we show some effect there, in the real world some of it doesn't carry through because we are not really thinking about it in the real world and what would make a patient want to use these types of technologies. How could it help them? It's a challenging space to work on: how do you make a scientific claim and also make it useful? But that's something that we should do more of.

Dr Perlis:One of the challenges in AI, as it relates to health, is that developers build things they are excited about and clinicians and patients have things that they need, and they don't necessarily talk to each other until it's too late -- until the thing is built. As someone who has a foot in both worlds, how do we fix that? How do we get everyone at the table early enough to design things that will be useful to everybody?

Dr Choudhury:We're already getting there through our failures. Technologists have tried to be very disruptive and build things that they're excited about that haven't been adopted widely in health care. Also from a clinical perspective, some of the solutions that use technology haven't leveraged the best of technology to make it engaging, useful, or nonburdensome to patients or clinicians.

The landscape is changing. Before, technologists had a hammer in search of a nail, and over time, clinicians and technologists started working together more where we got better at finding the right tool for the right problem, or the right hammer for the right nail. But I think, in this case, where we have a big need in clinical care and rapid innovation in technology and AI, if we come together early on and codesign solutions together, we can invent new tools all together to better target the problem that we care about. So, we need more early interaction and codesigning than even matching the tool with the problem.

Dr Perlis:Do you think that happens routinely now or do you still see that people build and then ask?

Dr Choudhury:I don't think that happens routinely. I do think people build and ask, and that gets back to a bigger problem as we function in this space. As researchers, entrepreneurs, or clinicians, sometimes we stay in our own community. Even if we think about research between a technologist and clinical researcher, how do you take that research into the real world when you are an entrepreneur? How do you think about the most evidence-based solution and still build a technology that can be profitable in the timeline that you need to be profitable?

One of the biggest problems I see is that there are gaps between communities that are working on solutions. Researchers and clinicians are looking at futuristic solutions. Implementation scientists are thinking about how to take solutions and implement it into a care setting. Entrepreneurs are trying to think about how new solutions can be really "productized" and serve a specific business or clinical need. Those gaps are almost valleys of death, which we see but assume that the bridges will be built.

Now with the tools and technology being where they are and the real clinical need, there are opportunities to more easily blend those together. Before, it was harder to take the latest technology and deploy it in a clinical context, but now we often have AI tools and infrastructure where it's much easier to take a new idea and test it out in the real world. We need to have more support and processes to quickly take research into the real world, and test it and reevaluate it, and redesign and iterate.

Dr Perlis:That's a huge priority for us -- understanding the technology's great but show me that it works in the real world in the ways that we, in medicine, are accustomed to seeing. So, whether it's a randomized trial or big naturalistic study, show me that this actually changes outcomes. Speaking of which, in 2021, you stepped sideways from academia to join Optum Labs. What was it like to go from a more traditional academic role to a place where it was more about real-world deliverables?

Dr Choudhury:It was a fascinating experience, and I learned a lot. One of the things as a researcher that I never thought about was the payment side of health care, the scale of health care delivery, and the needs. I've worked with a lot of clinicians and in clinical research doing randomized controlled trials, but one of the things that was always a little bit fuzzy to me was, what does it take to take a solution that we clearly know has some clinical benefit and be able to scale it in a way that will have wide-scale adoption and also be affordable.

At least in US health care, that understanding is important: How do you reduce the tension between a patient-doctor interaction and a health system or financial side? And how do we design solutions that can potentially bring them together? That was really important for me in thinking about what are the ways where we can bring solutions that really reduce that tension and can be acceptable for all the stakeholders in health care?

I can give one example. I had been involved in a start-up that was taking some of the [mental health] research out into the real world in terms of product. In thinking about digital biomarkers and thinking about the context of where they could be useful, there are big needs in primary care in understanding and supporting mental health. It is a place where there is an infrastructure and also potential reimbursement models that can support integration of mental health. We already know that this is having a huge impact in terms of people's outcomes, both in mental and physical health. Someone with depression and diabetes will have worse outcome longitudinally than someone with just diabetes alone, and that will cost more. So, thinking about how there's already a clinical model and collaborative care that supports that, how can technology amplify that and scale it and make it more accessible and make it cheaper? So understanding how these things come together has shaped the way I think about research solutions now as well. It's been very valuable to learn things from different perspectives.

Dr Perlis:So with those perspectives, we build a lot of these technologies that everyone agrees should exist. It's extremely helpful presumably to be able to track someone's status over time, but the problem always becomes who's going to pay for it? Did your time at Optum shift how you think about that?

Dr Choudhury:There are new ways people are paying for health care. If we think about just wearable sensors, we've seen a huge change in growth in remote patient monitoring and reimbursement. We are seeing remote therapeutic monitoring that can help in therapeutic intervention. So, I do think that there are new types of reimbursement models that are coming into play. From a technology perspective, as technologists, we are not often aware or get involved in showing the potential of technology and how it could support the needs, but also reduce the cost, without compromising outcomes. So, I do think more involvement from all sides is needed.

We need to be more involved in thinking about the whole picture. Particularly, from a research side, we build that innovative solution and then sometimes assume that other people will figure out the rest. I think we're realizing we can't do that. We can't do it when it comes from ethics. We can't do that in terms of thinking about privacy issues. We can't take that stance when we think about payment and reimbursement and how much it's going to cost. I also think that on the entrepreneurial side, we often start with a big innovative vision and then fall back into the standard models of reimbursement or models of care just to be financially sustainable and [offer] the return on investment that's needed to grow a start-up.

As we think about investment and financing of innovative solutions and research, what are ways where we can move the field forward as a whole, as a collective community, and bring people together? And what type of funding model is needed? We have pure research funding, we have investor funding, we have philanthropic funding. One of the things we can collectively do is to bridge and create glues to carry the innovation forward into practice and support that growth. Right now, often we have these islands of growth and things in between fall through the cracks.

A lot of money is spent through NIH [National Institutes of Health] to do clinical research. How do we take it forward? We are seeing more and more efforts toward trying to get that forward, and I think we need more.

Published Online: October 18, 2024. doi:10.1001/jama.2024.21212

Conflict of Interest Disclosures: Dr Choudhury reported being the digital health senior vice president of UnitedHealth Group from 2021-2023 and being an equity holder and cofounder of Health Rhythms and Dapple Health. She reported receiving speaker honoraria from Oxford University, Dartmouth College, the University of Michigan, and the Addiction Health Services Research conference and receiving research funding from the National Science Foundation and the National Institutes of Health.