New guidance helps healthcare reduce the environmental impact of AI use

Key takeaways

Healthcare’s growing use of AI carries an environmental cost that reaches beyond electricity and water into hard-to-quantify Scope 3 supply chain emissions, in an industry that already accounts for close to 9 percent of US carbon emissions. A new procurement guide from Health Care Without Harm and Practice Greenhealth gives buyers two sets of five questions, covering vendors’ emissions commitments and data center operations, plus a framework for scoring the answers. The approach follows a familiar playbook, since purchasing power has already pushed suppliers toward safer chemicals and greener products, and the guide frames AI sustainability as a community health issue.

As healthcare broadens its embrace of artificial intelligence, AI is flexing its muscles in use cases ranging from managing the revenue cycle to improving health literacy to automating administrative processes.

That utilization comes at an environmental cost. AI’s energy footprint is well documented, as training models and running inferences requires much electricity and water. Less known, noted a Health and Technology paper, is AI infrastructure’s need for metals and plastics that aren’t easy to produce sustainably, often go into “products designed for planned obsolescence and inevitable upgrades,” and themselves contribute to pollution.

The answer could be a sustainable approach to AI procurement modeled on the industry’s approach to improving energy efficiency and reducing exposure to harmful chemicals. 

AI’s potentially substantial impact on healthcare emissions

Providers in the United States aren’t strangers to the sustainability debate. The nation produces nearly 25% of all global healthcare greenhouse gas emissions, and the industry is responsible for close to 9% of carbon emissions in the United States. 

Most of that stems from indirect emissions. Applying the Greenhouse Gas Protocol, a NEJM Catalyst paper pointed out that just 7% of healthcare’s carbon emissions can be considered Scope 1, or direct emissions from owned sources such as boilers and medical equipment. Another 11% are Scope 2, or indirect emissions from purchased energy. Emissions from supply chain partners (Scope 3) account for the remaining 82% of the industry’s emissions.

Though AI primarily contributes to Scope 2 emissions, the paper noted that “the broader supply chain of resources to develop and maintain complex AI tools” has an impact on Scope 3 emissions that’s both substantial and difficult to quantify. Emissions related to day-to-day use “rapidly accumulate when considering the frequency of use, the number of models deployed simultaneously, and the size of the health system.”

Building sustainability into the bidding process

One solution, according to the paper, is “thoughtful AI implementation” that accounts for applying AI in appropriate scenarios, using less complex models, and running applications in more energy-efficient data centers.

The third recommendation is at the heart of a sustainable AI procurement guide recently released by nonprofits Health Care Without Harm and Practice Greenhealth. The document spells out how technology buyers can bring sustainability requirements into the bidding process for “products or services in which AI is integral to the core functionality,” whether it’s an AI model itself or a product with AI running in the background.

The guide includes two sets of five questions, along with a framework that helps procurement teams rate vendors’ responses.

• The first set explores a vendor’s commitments to using renewable energy and reducing emissions, estimating per-task emissions, optimizing models to lower environmental impact, and partnering with healthcare organizations to devise their own sustainable strategies.
• The second set focuses specifically on data center operations. They include providing updates on power and water use, energy sources, green building certification, and community or legal complaints. 

Minimizing the impact on community health

Organizations wary of pushing partners on sustainability should remember that healthcare has been here before. The movements to reduce reliance on toxic chemicals and increase use of environmentally friendly products got their start when procurement teams showed they were willing to spend their money elsewhere. Such an effort ensured that more than 70 million square feet of flooring across Kaiser Permanente’s vast hospital network is free of harmful chemicals while remaining resilient and easy to clean.

Healthcare faces a similar imperative with AI, the guide noted. The technology’s environmental impact contributes to air, noise, and water pollution, as well as energy instability that can make heat or air conditioning harder to come by, all of which disrupt health in the communities that providers serve.

“By asking about key environmental factors that contribute to health, healthcare organizations can better understand AI’s impact and seek vendors committed to environmental stewardship and a healthy environment for all”.

Brian Eastwood is a Boston-based writer with more than 10 years of experience covering healthcare IT and healthcare delivery. He also writes about enterprise IT, consumer technology, and corporate leadership.