AI could help hospitals become smarter, but it may also make them less sustainable. Researchers from the EWUU AI Hub are developing a framework to reduce AI’s environmental impact across its entire life cycle.
Hospitals exist to protect human health. Yet healthcare also produces carbon emissions, consumes resources and contributes to environmental pollution; pressures that ultimately harm health. As hospitals adopt more artificial intelligence, this paradox becomes sharper. AI can improve medical imaging and make hospital processes more efficient, but it also requires energy-intensive computing, data infrastructure, water and specialized hardware.
The new EWUU AI Hub project RAINS, short for Reducing AI’s Negative Environmental Impact in Sustainable Hospitals, will examine how hospitals can benefit from AI without having a high environmental burden. Its goal is to develop a framework that reduces the environmental impact of AI in a hospital setting.
Balancing performance and sustainability
“AI is being used more and more in hospitals,” says dr Laurens Bliek, from the Department of Industrial Engineering & Innovation Sciences at Eindhoven University of Technology. “Besides its positive impact, AI can also have a negative impact on the environment. We want to reduce that impact and make the trade-offs more visible.”
“You can achieve a large reduction in energy use with only a negligible loss in accuracy. If we consider the whole life cycle, the potential gains could be much larger.”
dr Laurens Bliek (TU/e)
AI is often presented as an enabler of the transition towards a circular society. It can optimize resource flows, improve operational efficiency and support data-informed decisions. Yet its sustainability cannot be judged only by the savings it creates. A system that reduces waste may still rely on energy-intensive training, large data centers or newly manufactured processors. This use of resources also continues after development, when the model is deployed, used and maintained.
A life-cycle approach
RAINS will therefore take a life-cycle perspective, covering data collection and preparation, model training, hardware, deployment and maintenance. Rather than producing one technical tool, the researchers aim to create a practical decision framework. “How do you measure different types of impact?” Bliek says. “What do you need to take into account with data gathering, hardware and software? And how do those choices affect both environmental impact and performance?”
These decisions involve genuine trade-offs. Smaller AI models may consume less energy, but they are not automatically preferable if they are also less accurate. In healthcare, environmental gains cannot come at the expense of patient safety. The framework should therefore help decision-makers understand the consequences of different options.
The breadth of the project makes collaboration within EWUU essential. Researchers can combine expertise in algorithms, data, software and hardware with UMC Utrecht’s knowledge of clinical practice. “The advantage is that we can create an overarching framework that covers all the different layers of an AI system,” says Bliek.
The team will begin by reviewing existing approaches and consulting experts and stakeholders through focus groups. It will then examine three hospital applications: medical image processing, wastewater treatment and energy management.
From medical imaging to energy management
“Medical imaging will serve as the first use case because AI is already used in this field”, says Bliek. “Training and fine-tuning image-recognition systems can require large, diverse datasets and substantial computing power. Once deployed, the energy needed by a single application may be modest, but the total can still add up when the same system is used across many hospitals.” Wastewater and energy will also be examined. These cases show the double role of AI clearly. AI could help reduce water pollution or support more efficient use of renewable energy, while the technology itself still consumes resources. “In those cases, the positive effect can also be environmental,” Bliek says. “But our focus remains on reducing the negative impact of the AI system itself.”
Low-hanging fruit
The current seed project of the EWUU AI to the RAINS project will help the team refine its approach and prepare an application for larger research funding. Bliek believes substantial improvements may be possible. “In my own research, I already see low-hanging fruit,” he says. “You can achieve a large reduction in energy use with only a negligible loss in accuracy. If we consider the whole life cycle, the potential gains could be much larger.”
Research team: dr. Laurens Bliek (TU/e), dr. Ewoud Schuit (UMCU), dr. Nishant Saurabh (UU), Alessio Belmondo Bianchi Di Lavagna, MSc (WUR), dr. Qi Han (TU/e)