For many years, the healthcare industry has been known as a major energy consumer. With the rollout of ever-increasing AI tools and technology, healthcare's energy demands are rising even further. That's why researchers have now looked at the environmental impact of AI in healthcare. Their advice? Deploy AI responsibly.
This is not a new discussion. We have been living in a digital, online world for many years. Where there used to be factories, there are now often mega data centers to host and keep all the servers and digital services accessible. Those data centers use an enormous amount of electricity. Although renewable and sustainable sources are increasingly being used to generate it, the share of fossil energy is still considerable. AI applications, think of the LLMs (Large Language Models) for generative AI services such as ChatGPT and machine learning solutions, also require a lot of energy. In fact, a lot more than you might think.
Responsible use of (generative) AI
More and more hospitals are introducing (generative) AI applications and machine learning. Among other things, for processing many thousands of electronic patient records, medical imaging, diagnostics and clinical trials. In fact, not a day goes by these days without mention of an AI-driven solution being developed, tested or deployed somewhere in a healthcare facility.
A team of researchers from the universities of Adelaide and Reading examined the impact of AI, and then generative AI tools (LLMs), on the carbon footprint of healthcare. They conclude, and advise, that it is important to use these applications responsibly. For example, using shorter prompts in AI tools deployed to summarize conversations with-and data from-patients. Consider tools such as Microsoft's Dragon Medical One.
ChatGPT consumes 15 times more energy than Google
Every day a patient is in a hospital, doctors, nurses and other hospital staff type up pages and pages about their health. By the end of a hospital stay, one patient's documentation may contain tens of thousands of words. Unlike busy healthcare personnel, LLMs like ChatGPT have the time to read through and process this information.
“With great processing power, however, comes great responsibility. A single AI query uses enough electricity to charge a smartphone 11 times. In addition, to cool the server, 20 milliliters of water per query is also consumed in the data center. ChatGPT uses an estimated 15 times as much energy as Google,” said research leader Oliver Kleinig. “Implementing LLMs and generative AI tools in healthcare can thus have very significant environmental consequences. The responsible managements of hospitals should therefore think carefully about where and when AI should be used in their organizations.”
CO2 emissions from hundreds of households
ChatGPT's daily CO2 emissions already equal those of 400-800 U.S. households. AI systems in healthcare would likely have an even larger footprint because they require more powerful models to process complex medical information and must be run locally for patient privacy.
In addition to the high energy consumption in data centers, another factor that comes into play is that these services require more computer systems, and thus must be produced. That alone could double the carbon footprint of AI operations.
Five key questions
To reduce the environmental impact of hospitals and other medical facilities, the researchers suggest five key questions that healthcare providers should consider before implementing AI systems:
- Does my organization need a large language model? Can existing technology suffice?
- Which LLM should I choose? Use the smallest possible model to reduce resource consumption - smaller, finely tuned LLMs can outperform larger applications.
- How can I optimize my LLM? Use smaller and specific prompts to reduce the carbon impact of applications. Concise prompts with refined information are more energy efficient.
- What hardware should my LLM run on? Preferably use hardware that runs on renewable energy.
- What data should I share? Maximize LLM efficiency by sharing data where appropriate.
The study suggests that AI could also reduce the healthcare industry's environmental impact in other ways, such as improving patient flow and reducing paper use. In general, especially in the Netherlands, an increasing number of healthcare institutions and healthcare manufacturers are working hard to reduce carbon emissions. The connection of the UMCG to a solar energypark, Philips' helium-free MRI scanners and the participation of more and more healthcare institutions in the Green Deal for sustainable healthcare are good examples.
