Why health AI needs an environmental impact framework

Transcript - Why health AI needs an environmental impact framework

Manijeh Berenji: To be able to run these large language models, whether it's a GPT-3.5 or one of the Llama models, clearly, there is an energy footprint associated with that. And we're starting to see how it's impacting our overall electric utilization.

Kelsey Waddill: Hello, you're listening to Healthcare Strategies: Industry Perspectives. Coming to you from HIMSS 2025 in Las Vegas. I'm Kelsey Waddill, a podcast producer at Informa Tech Target. Training an AI model like GPT-3 can consume enough electricity to power 120 homes in the U.S. for a full year. According to a paper from Google and University of California, Berkeley researchers, this kind of high electrical usage can impact air, land and water quality. So, with higher numbers of healthcare organizations adopting artificial intelligence tools, will we have to choose between a healthy environment or healthy patients? I spoke with Dr. Manijeh Berenji, MD, PhD, associate clinical professor of environmental and occupational health at the University of California Irvine School of Medicine & School of Public Health -- that's a long title -- to find out the answer. Dr. Berenji, thank you so much for coming on to Healthcare Strategies.

Berenji: Absolutely. Thank you so much for the invitation, Kelsey.

Waddill: Before we get into, kind of, the actual topic that we want to address here, I'd just love to introduce you to our audience a bit. Can you just tell us a little bit about yourself, what you do?

Berenji: Sure thing. So, I am an associate clinical professor of medicine and public health at UC, Irvine, here in Irvine, California. And I really have the pleasure of working with our faculty, medical residents and students, really looking at the intersection of the environment and its impact on health outcomes. It's been a great ride so far. It gives me the opportunity to work with various different patient populations. I also do work at the VA, so I interface with veterans. So being able to understand how the environment impacts our day-to-day activities, how it actually has the ability to worsen conditions that many of our patients suffer from. Really trying to better quantify some of these exposures so we can do better when it comes to treating our patients.

Waddill: I am really intrigued to hear about how -- it sounds like you have interaction with a lot of different kinds of patient populations – so, I'm really interested to hear more about what you're seeing in terms of how the environment impacts different populations in different ways. So, to get into that, we have actually discussed on Healthcare Strategies a bit about this topic, but I think it's necessary to reiterate here, at the front end, why folks in the medical fields should care about what's happening in the environment right now and the impact that it's going to have on patients and is currently having on patients.

In recent years, it has become clear that we must account for the impacts of natural disasters, changing climate trends and global warming as we're assessing communities' health and wellness. And I'd just love to hear from you -- can you talk a little bit about the impact that our changing environment can have on patient health?

Berenji: Absolutely. So here in Southern California, we recently had the wildfires here in Los Angeles. I was personally affected. I live in metropolitan Los Angeles, and for a number of weeks this past January, many of us here felt the direct impacts with the air quality, just seeing the smoke, the particulates in the air. I personally had eye irritation, throat irritation, some mild cough for a number of weeks. But to see how it was impacting the patients that we treat, I had a number of veteran patients with asthma exacerbations, folks coming in with just general concern about their overall health, how it's impacting their kids and their grandparents. So really trying to navigate that was challenging because information was coming pretty quickly to us in the healthcare field. Being able to take in all this information from the National Weather Service with respect to the winds and the fires, getting up to date from the news, identifying where these fires were spreading because we had so many of them at any given time.

So, being able to wrap our heads around the changing dynamics of these fires really put things in perspective for me, both on the clinical front, but also for, kind of, my own health and well-being. So being able to disseminate information to patients, leadership, respective community members, making sure that the information is understandable, that folks know what to do when they get an alert, how do they protect themselves and their families, and ensuring that we can be a resource for our patients, especially if they're having an issue that requires medical attention.

So that was pretty much the first couple of weeks for me in January, trying to coalesce our messaging and making sure that we were taking care of each other during that time. But clearly, we're seeing more and more of these events at greater frequency, at greater intensity and trying to essentially integrate our response both from the emergency management side to the clinical treatment side and really trying to reconcile how we make determinations about where we should counsel our patients, how we should counsel them. It was one of those things where you just have to adapt and making sure that you're doing everything that you can to stay updated and be a resource to patients in a moment of crisis. But we got through it. We are LA Strong, and honestly, at the end of the day, we did take some lessons learned in terms of how we can better optimize our emergency preparedness and our overall resiliency in the light of some of these natural disasters that unfortunately are occurring at an increased frequency.

Waddill: Thank you for sharing that. I'm very glad you're safe.

Berenji: Yeah.

Waddill: And I think for some of us, depending on where we are geographically, you can be sheltered from the realities of how the environment is changing. Thank you for sharing from your own story, and again, glad that you're safe. As a culture, recognizing the need to capture these environmental impacts in health data, we're also seeing parallel to that a dramatic increase in AI adoption by healthcare organizations. And these might seem unrelated, but I'm curious about what kinds of environmental impacts artificial intelligence can have as we're talking about greater adoption and more widespread environmental impact. How does AI factor into all of this?

Berenji: So here in the healthcare sector, we're seeing the rapid advancement and deployment of AI technologies. We're already using AI to help with documentation. There are a number of ambient listening platforms that take, essentially, voice information from the clinicians, from the patients and, essentially, creates a note, which has really transformed how we do our administrative tasks. So, we're already seeing the amazing benefits that this technology can offer, but on the flip side, we are also seeing an increase in energy utilization to be able to run these large language models, whether it's a GPT-3.5 or one of the Llama models. Clearly, there is an energy footprint associated with that.

And we're beginning to see some of the ramifications of how the need for more electricity to drive the functionality of these AI models and respective operational capacities, we're starting to see how it's impacting our overall electric utilization. And clearly, we have to look at, kind of, the full picture here. So, if we're able to use the technology to essentially mitigate a health event, if we can actually predict who is going to develop a certain disease and try to be on the front end of that before it turns into something that could potentially cost tens of thousands of dollars if not more, if we can use the technology wisely, efficiently to drive care in such a way where we can actually do better when it comes to prevention, we can actually make that justification for utilizing some additional energy utilization to be able to drive some of those models.

But, at the same time, again, just better understanding the landscape in the last few months. I feel like there are some things that we do in healthcare that, quite frankly, does not need an AI solution. I think it requires reimagining how we do our day-to-day operations in healthcare. So, we all love the shiny new toy. I'm one of them. I really love what this technology brings, but I think we just really have to use it intentionally, use it responsibly and make sure that we are using it the best way possible for our patients so that they do not need to have potential health issues accelerate. And it gets to the point where we can't do as much as we could have done. So that's how I have created this justification of the technology, being able to go through these trade-offs. But at the end of the day, really focusing on the ultimate good that this technology can bring and making sure that we can redesign some of our existing workflows to really leverage our human capital to the best of our ability.

Waddill: Yeah. Do you feel comfortable just sharing either some of the areas where you think AI use is worth the impact or worth the electrical use or the areas where maybe we are overdependent on it and it's causing more harm than good?

Berenji: I think it's really just about being intentional with how we use the technology and understanding our current needs.

Waddill: Okay.

Berenji: And our current barriers to delivering the best healthcare possible. And if we're able to do that, essentially use a framework to make those justifications, and then really being cognizant of what kind of energy expenditures are going to be required.

I think we really have to be able to amalgamate all of these different data sources so that we can make informed decisions. And a lot of times, like I said, we can actually do improvements in healthcare without necessarily resorting to the AI. The AI is really there to drive predictive analytics. It's there to help reduce documentation burden for clinicians so they can deliver better healthcare. We can actually use the technology in a way where we're still protecting our greater patient population and ensuring that we are doing everything that we can, leveraging all the tools that we have in our toolbox to get them the best outcomes.

Waddill: What advice would you give to healthcare organizations that want to achieve that balance about being more thoughtful about their usage of these tools as you've been mentioning? Are there any important first steps that you would advise them to take as they seek to strike what could be a pretty difficult balance, I think, between preserving the environment, assessing their environmental impact and also still making use of these technologies to even analyze how these environmental changes are impacting their patients and that kind of data? Just what advice would you give to that kind of an organization?

Berenji: So yeah, these are really pivotal questions that we're asking right now, and I've talked to healthcare leaders at various organizations about how we can actually start to find that balance. I think really what it boils down to is understanding current operational capacity. What kind of energy expenditures are we currently using? How much water are we currently using? And really just trying to understand what our baseline is before mass AI deployment. I think taking the time to do that work is actually going to be very informative because that way, as we start to scale the technology, we can start to identify priority areas.

Like I said, I think if we can actually use the technology to help us predict who is going to develop a particular health outcome, and doing everything that we can to mitigate that, to me that would be a high-priority indication for developing those AI models and respective tools.

Documentation burden. Clearly clinicians continue to have high rates of burnout. How can we actually leverage the technology to be able to help us in our day-to-day operations but still tapping into existing non-AI tools to help achieve that perfect balance between AI, non-AI and human capital? I think that's really going to be the exercise that healthcare leadership and clinical department leadership need to undertake to really do that deep dive, to do that needs assessment, to better understand our existing systems and identify those barriers in our current day-to-day operations. And actually being intentional with how we can leverage the technology to the best of our ability, being mindful of our energy and resource capabilities.

I think this is going to be something that, as we continue to deploy this technology and looking at how much energy is going to be utilized, I think we really have to understand ways where we can streamline a lot of our clinical workflows and essentially having access to all of the tools. AI is just one tool. There are many other tools out there that we can actually incorporate and do a better job of integrating those tools into our existing workflows. And being able to ultimately create the best value for the patient. We want to be able to give them what they need so they can take care of themselves and their families.

Waddill: Right.

Berenji: And AI is a great tool to help us achieve that. But at the same time, we have to make sure that we don't forget our core values, our core mission to our patients.

Waddill: Yeah.

Berenji: And making sure that we're doing everything that we can to be mindful of that.

Waddill: I wonder, too, if there's opportunity for healthcare organizations to communicate those priorities to vendors who could respond with tools that are perhaps more conservative of the environment. Obviously, there's not a huge amount of incentive there, perhaps from a financial standpoint for the vendors, but I wonder if it would be enough to just have healthcare leaders be more vocal about: we're looking for a middle way.

Berenji: Yeah.

Waddill: It would be interesting to see what that could do.

Berenji: I think those conversations are already happening, and I was very fortunate to present at HIMSS, just raising awareness of this issue. I think, to me, that's the first step.

Waddill: Yeah.

Berenji: And then as we start to understand, from a resource allocation perspective, what the energy requirements are going to be, what the water requirements are going to be, how much energy are we going to have to use to be able to run these large language models?

And starting to see the impacts of that with respect to our electric bills, how we're going to actually source energy. I think these are conversations that are now being had with our colleagues across the hospital system. So I think really creating safe spaces for these conversations, ensuring that we are prioritizing patients' health and leveraging the technology to the best of our ability, understanding our existing resources that we have and making sure that we have contingencies in case there is a natural event or something that may potentially create barriers to how we utilize a technology like AI.

I mentioned the wildfires earlier. How about if there's a flooding event or a hurricane? I think we have to be prepared for those types of scenarios and making sure that we have backup sources of energy, electricity to be able to run the core operations that need that level of energy expenditure so that we can still do our day-to-day hospital operations. But again, making sure that we are doing everything within our scope to leverage the technology and really understanding how we have to pivot, especially in times of crisis like a natural disaster. So, I think this is going to require a framework.

I know the Coalition for Health AI and some other consortiums out there are really developing these frameworks for using AI responsibly. I think being able to create some additional guidance around the environmental impacts and creating that fundamental understanding of how we can continue to scale. But again, every hospital is going to have different resource needs, so really taking the time to understand the resource needs, being able to understand where our electricity is sourced from, being able to develop those contingencies in case there is a natural disaster or other type of event where there could be downstream impacts to the grid and the electrical needs of the hospital to run a lot of these technologies, including AI.

So, I think, really just having that time to do the exercise will really yield benefits to the greater organization as a whole. And I think vendors are clearly cognizant of that because they, too, understand that natural disasters are happening more and more often, and being able to have those plans in place so that we actually are in a better position to deal with that type of situation as they arise.

Waddill: Thank you so much, Dr. Berenji, for your time today and for sharing this with us, and hopefully I'm excited to hear what our listeners take away from this and just in the hopes of moving towards a healthier and safer environment both naturally and clinically.

Berenji: Absolutely.

Waddill: So thank you so much for your time today and for coming onto the podcast.

Berenji: Of course. Thank you so much, Kelsey. I really appreciate it.

Waddill: Listeners, thank you for joining us on Healthcare Strategies: Industry Perspectives. When you get a chance, subscribe to our channels on Spotify and Apple and leave us a review to let us know what you think of this series. More industry perspectives are on the way, so stay tuned.

This is an Informa Tech Target production.