SCET Faculty Shomit Ghose publishes decision framework for sustainable Gen AI in healthcare
Clinical decision-making, customer support, technical management: uses of AI in healthcare are growing rapidly. But a large-scale shift requires massive resource input. Electricity consumption in data centers, for example, has increased by 116 trillion million kilowatts since 2016.
A new paper co-authored by SCET faculty Shomit Ghose analyzes the question: what trade-offs can health system leaders make to balance performance, cost, and resources?
from Lawrence Berkeley National Laboratory’s 2024 U.S. data center energy use report.
“Sustainably Advancing Health AI: A Decision Framework to Mitigate the Energy, Emissions, and Cost of AI Implementation”, published in the New England Journal of Medicine Catalyst this week, tackles this problem with the Sustainably Advancing Health AI (SAHAI) framework.
The framework was designed by Ghose and fellow authors Anu Ramachandran (Stanford), Chethan Sarabu (Cornell, Stanford Medicine), Udit Gupta (Cornell), and Vivian S. Lee (Harvard Medicine) to optimize AI’s utility with minimal energy consumption and emissions.
Considering model, data center, and governance costs, the authors found potential in mixed-quality models, with simpler models performing baseline tasks and more complex ones being reserved for high-performance tasks. They also called for a coalition between healthcare sustainability leaders for future development.
“From a competitive alignment perspective, climate-efficient AI is economically efficient AI,” the paper states. Moreover, the human health impact of sustainable resource management makes this issue especially critical to the healthcare sector.
Ghose has analyzed this discussion in-depth before, including in his article “Reducing AI’s Climate Impact: Everything You Always Wanted to Know but Were Afraid to Ask”. Here, he defines the below taxonomy for where AI energy interventions can be made, which is also used in the new framework.
Read the abstract of the New England Journal of Medicine article below:
With increasing pressures to deliver higher quality, safer, affordable care that is more equitable and accessible, U.S. health systems are looking hopefully at AI tools, including new and emerging generative AI capabilities, as a means of transforming medical care while alleviating workforce stresses.
These AI technologies require substantial energy and water consumption as well as other resources to develop, deploy, and maintain. When considered at scale, AI technologies have the potential to impact the energy utilization of health systems and their ability to maintain their sustainability commitments, including the 2022 U.S. Department of Health and Human Services’ Health Sector Climate Pledge.
Multiple factors determine the energy requirements of a given AI tool, and health system leaders will have a critical window of opportunity to align AI implementation with larger considerations of appropriate resource utilization, sustainability, and cost.
In this article, the authors offer a framework — the Sustainably Advancing Health AI (SAHAI) framework — for optimizing AI-related energy consumption and emissions in health care settings. Through an example of a generative AI use case — AI patient messaging — they calculate carbon emissions across various scenarios that could substantially affect the emissions profile of a major health system using such a tool. The authors discuss key takeaways for health systems implementing new AI technologies and offer concrete next steps for a coalition to advance health AI sustainably.
Clinical decision-making, customer support, technical management: uses of AI in healthcare are growing rapidly. But a large-scale shift requires massive resource input. Electricity consumption in data centers, for example, has increased by 116 trillion million kilowatts since 2016.
A new paper co-authored by SCET faculty Shomit Ghose analyzes the question: what trade-offs can health system leaders make to balance performance, cost, and resources?
from Lawrence Berkeley National Laboratory’s 2024 U.S. data center energy use report.
“Sustainably Advancing Health AI: A Decision Framework to Mitigate the Energy, Emissions, and Cost of AI Implementation”, published in the New England Journal of Medicine Catalyst this week, tackles this problem with the Sustainably Advancing Health AI (SAHAI) framework.
The framework was designed by Ghose and fellow authors Anu Ramachandran (Stanford), Chethan Sarabu (Cornell, Stanford Medicine), Udit Gupta (Cornell), and Vivian S. Lee (Harvard Medicine) to optimize AI’s utility with minimal energy consumption and emissions.
Considering model, data center, and governance costs, the authors found potential in mixed-quality models, with simpler models performing baseline tasks and more complex ones being reserved for high-performance tasks. They also called for a coalition between healthcare sustainability leaders for future development.
“From a competitive alignment perspective, climate-efficient AI is economically efficient AI,” the paper states. Moreover, the human health impact of sustainable resource management makes this issue especially critical to the healthcare sector.
Ghose has analyzed this discussion in-depth before, including in his article “Reducing AI’s Climate Impact: Everything You Always Wanted to Know but Were Afraid to Ask”. Here, he defines the below taxonomy for where AI energy interventions can be made, which is also used in the new framework.
Read the abstract of the New England Journal of Medicine article below:
With increasing pressures to deliver higher quality, safer, affordable care that is more equitable and accessible, U.S. health systems are looking hopefully at AI tools, including new and emerging generative AI capabilities, as a means of transforming medical care while alleviating workforce stresses.
These AI technologies require substantial energy and water consumption as well as other resources to develop, deploy, and maintain. When considered at scale, AI technologies have the potential to impact the energy utilization of health systems and their ability to maintain their sustainability commitments, including the 2022 U.S. Department of Health and Human Services’ Health Sector Climate Pledge.
Multiple factors determine the energy requirements of a given AI tool, and health system leaders will have a critical window of opportunity to align AI implementation with larger considerations of appropriate resource utilization, sustainability, and cost.
In this article, the authors offer a framework — the Sustainably Advancing Health AI (SAHAI) framework — for optimizing AI-related energy consumption and emissions in health care settings. Through an example of a generative AI use case — AI patient messaging — they calculate carbon emissions across various scenarios that could substantially affect the emissions profile of a major health system using such a tool. The authors discuss key takeaways for health systems implementing new AI technologies and offer concrete next steps for a coalition to advance health AI sustainably.