While health practitioners may now gain minutes or hours from AI scribes used in medical consultations, this comes at a cost that\u2019s not clearly visible. The environmental impact of AI is significant, and this will also impact human health.\u00a0<\/p>\n
Generative AI, which includes large language models such as ChatGPT and Claude, is behind many emerging innovative medical AI tools like ambient scribes that can support the medical workforce. It is driving much of the recent growth in demand for data centres that house and service AI providers. Australia has become the second-largest destination, behind the US, for data centre investments worldwide, according to a recent Climate Council report<\/a>.\u00a0<\/p>\n Although the healthcare sector \u2018has the most potential to benefit from AI adoption\u2019 according to a 2024 Productivity Commission healthcare research report<\/a>, these data centres come with environmental risks. As the World Resources Institute<\/a> reports, data centres need a constant and reliable power supply. Many facilities rely on gas-fired generation for routine operations alongside diesel generators for emergency backup, which emit toxic pollution.\u00a0<\/p>\n Associate Professor Zerina Lokmic-Tomkins<\/a>, a climate and health researcher at Monash University, says generative AI is far more computationally intensive than the predictive AI that has been used in health care for years for narrow tasks such as reading mammograms and flagging arrhythmias on ECGs.<\/p>\n Meanwhile, the health impacts of this are receiving less attention. Can AI be used to advance health without creating environmental pressures that are also detrimental to health, and how?<\/p>\n The environmental costs of AI\u00a0<\/p>\n Professor Tomkins distinguishes between direct and indirect environmental impacts.\u00a0<\/p>\n \u201cDirect impacts occur at or near the facility and along its supply chain, for example, local air pollution, local water stress, and exposure to toxic materials when hardware is manufactured or discarded. Indirect impacts flow through larger systems, such as the greenhouse gas emissions from electricity generation that drive climate change. The two have different victims, different timeframes, and different solutions.\u201d<\/p>\n Greenhouse gas emissions and global warming<\/p>\n According to the Australian Energy Market Operator, data centres already draw about 2% of the electricity<\/a> on Australia’s main grid, and energy-market forecasts have that tripling by 2030<\/a>.<\/p>\n The increase in energy demand from these data centres is likely to put a setback to Australia\u2019s transition to renewables. As the Climate Council<\/a> explains, Australia may need to rely on coal for longer, increase gas generation and if data centres are built without additional renewable capacity, Australia\u2019s National Energy Market climate pollution could be 14% higher than it would otherwise be in 2035.<\/p>\n Professor Tomkins says, \u201cThere is also the question of what happens during heatwaves and natural disasters, when the electricity grid is under strain and priority must go to essential and emergency services and health care. At those times, data centres may switch to diesel backup generators. In Australia, we are yet to see publicly available strategies for how this will be managed over our summers.\u201d<\/p>\n E-waste<\/p>\n The issue that concerns Professor Tomkins most is e-waste.\u00a0<\/p>\n \u201cData centres also depend on hardware made from plastics and rare minerals such as lead, mercury and chromium. If extracted or discarded (as e-waste) in environmentally unsustainable ways, these toxic elements pollute waterways, land and air, can enter food chains, and have serious effects on child development. Exposure often falls on informal recycling workers overseas.\u201d<\/p>\n According to Greenpeace\u2019s report, Environmental Impacts of Artificial Intelligence<\/a>, large quantities of up to 5 million tonnes of additional electronic waste will be generated by the expansion of data centres and AI capacities until 2030.<\/p>\n Air pollution<\/p>\n Michael Cork<\/a> is an environmental health scientist and co-founder of EmPower Analytics Group, a consulting company that provides independent, science-based health and economic analyses of AI infrastructure.<\/p>\n He says data centres are also a public health issue because of the dangers of fine particulate matter they emit, estimated to drive nearly 90% of the health impacts from air pollution.\u00a0<\/p>\n Water<\/p>\n Cooling data centres requires large quantities of water. As the University of NSW<\/a> reports, this can place strain on local water supplies, particularly in areas already facing water scarcity and in hotter climates.\u00a0<\/p>\n Researchers<\/a> report that data centres increase the threat of water insecurity, and thus dehydration and poor hygiene.<\/p>\n Considering the environmental footprint of using digital tools, including AI<\/p>\n When using medical AI, Professor Tomkins says the more useful question is a clinical one that balances care with climate. \u201cDoes this tool improve care or reduce waste, or does it just add activity? If an AI tool prevents one unnecessary CT scan, it has likely more than justified its footprint. A tool that generates longer notes nobody reads probably has not. The same logic applies to all our digital habits. Duplicate storage, unnecessary images and endless email chains also run on data centres.\u201d<\/p>\n A Radiology Journal article<\/a> discusses an example of this low-impact AI use, which reduces resource-intensive care instead of increasing computation. It finds that AI can shorten MRI scans, which should reduce scanner energy use, and improve scanner scheduling efficiency, helping imaging systems run with less idle time and waste.\u00a0<\/p>\n Can AI contribute to health gains while mitigating environmental health risks<\/p>\n Dr Liam Liu<\/a> is a Postdoctoral Research Fellow in the School of Public Health and Preventive Medicine at Monash University. He says there are ways AI could assist in mitigating environmental challenges. \u201cResearchers have used machine learning for years to estimate pollution and study exposure to environmental hazards. These methods remain useful. Forecasting healthcare demand during heatwaves is one application. Better forecasts could help hospitals plan staffing and capacity and help community services contact vulnerable patients earlier.\u201d<\/p>\n Professor Tomkins talks about the choices that can be made. \u201cUse the smallest model that does the job safely. We can still transition to renewable energy and run generative AI on renewably powered infrastructure. Hardware designed to last longer means less obsolete equipment and therefore less e-waste. And we can deploy AI only where it demonstrably improves outcomes or reduces resource use, rather than using it indiscriminately. Ultimately, though, sustainable AI only makes sense within sustainable digital health more broadly. We are hyper-focused on AI but our whole digital footprint needs the same scrutiny, not just the newest part of it.\u201d<\/p>\n