The health care sector is increasingly integrating artificial intelligence to enhance patient interactions and streamline operations. A recent study from Cornell University introduces a framework aimed at embedding sustainability into these AI systems. Named Sustainably Advancing Health AI (SAHAI), this framework focuses on optimizing energy consumption and emissions associated with AI technologies in health care environments.
The research team, including Dr. Chethan Sarabu, director of clinical innovation at Cornell Tech, and Udit Gupta, assistant professor of electrical and computer engineering, emphasizes the importance of considering greenhouse gas emissions from AI applications. This includes emissions from AI-enabled patient messaging and the water required for cooling data center hardware. Their findings are detailed in the article titled “Sustainably Advancing Health AI: A Decision Framework to Mitigate the Energy, Emissions, and Cost of AI Implementation,” published on September 12, 2023, in NEJM Catalyst.
Health care systems in the United States are turning to AI technologies to alleviate the strain on their workforce. Automated responses to patient inquiries are a key component of this transformation, with the market projected to reach $187 billion within five years. Despite the advantages of AI in reducing workloads, the technology poses significant challenges to energy infrastructure and sustainability objectives.
The researchers illustrated the carbon emissions associated with AI tools through the example of an AI-generated messaging application implemented in a large academic health system. They estimated that operating this AI tool with 3,000 physicians responding to 50 messages per day for one year would produce around 48,000 kilograms of carbon dioxide (CO2), equivalent to roughly 2,300 “tree-years”. A tree-year represents the amount of CO2 absorbed by one tree in a year, which typically varies based on various factors.
In their modeling, the team utilized a lightweight generative pretrained transformer (GPT) that consumes less computing power than more extensive models. While this lighter model can handle directing patients or answering general inquiries, more complex tasks necessitate larger models.
Dr. Sarabu noted that health care providers must carefully evaluate multiple factors, such as energy usage and cooling water consumption, when implementing AI systems. “If you’re responding to a patient about routine follow-ups, small differences in model accuracy, such as the difference between 83% and 85%, may not be noticeable,” he explained. “But if you generate double the amount of emissions with 85% accuracy, that’s probably not striking a good balance.”
The researchers advocate for addressing sustainability during the design phase of AI systems rather than retrofitting them afterward. “We’re really in the early days of AI being implemented in health care,” Dr. Sarabu stated. “What happens in the next three years or so will become ingrained in the system. If we make energy-conscious decisions now, we’ll create a more efficient system.”
Gupta added that prioritizing sustainability does not necessarily mean compromising on performance. “Data centers located on renewable energy grids will significantly reduce operational emissions,” he noted. “Hospitals can choose to run AI workloads in facilities that operate on renewable energy.”
The research concludes that there exists a “window of opportunity” to align the extensive integration of AI with environmental considerations. “Although the impacts of climate change disproportionately affect vulnerable patient populations and the health systems that serve them,” the authors wrote, “emissions generation is primarily driven by affluent countries and high-resource health systems, which must acknowledge and mitigate their contributions.”
By recognizing the scale of impact and investing in strategies to reduce the environmental footprint of AI, the health care sector can make strides towards a more sustainable future.
