Artificial intelligence risks becoming an environmental disaster

Artificial intelligence (AI) is being heralded as the potential saviour of financially constrained health systems worldwide. Its implementation, however, could be disastrous for the climate. Although the ambitious claims of AI and its benefits are compelling, it is no panacea. Health systems are some of the world’s greatest carbon polluters and the expansion of AI risks derailing the progress that has been made to meet net zero targets.1 Critical appraisal, rather than unchecked enthusiasm, is required to ensure that AI is not used to solve certain problems facing health systems only to exacerbate a much greater problem.

The UK government is positioning itself to be a global leader in AI, alongside the US and China. Its recently published AI Opportunities Action Plan refers multiple times to the use of AI within the NHS.2 Promises and new realities include transcription technology to reduce the administrative burden and a £21m AI Diagnostics fund to deploy technologies in high demand areas such as lung cancer imaging diagnostics.3

Yet the energy and infrastructure requirements, carbon intensity, and water usage required to run AI technologies far outweigh those of current IT systems. ChatGPT prompts, for example, require 10 times the energy of a simple search engine.4 To meet the tech industry’s burgeoning energy demand, the US is set to build as many as 80 new fossil fuel plants by 2030.5 Similar energy demand challenges are faced by the UK and globally as demand continues to rise.67 The high water usage of large data centres for cooling, often drawing from local sources, threatens the local availability of drinking water.8 AI is also projected to increase electronic waste by up to 12% per year globally, the equivalent of discarding 13 billion iPhones annually.9

The NHS is currently partway through an ambitious roadmap to net zero by 2040.10 When the Delivering Net Zero strategy document was embedded into legislation and republished as statutory guidance in 2022, it contained a baseline and projections for the carbon footprint of NHS England. Since 2022, the potential of AI to disrupt the progress of healthcare systems has grown exponentially. It is unclear whether the climate impact of increased AI use within the NHS has been considered as part of these projections. Consider also that the reduction in NHS emissions has slowed since 2020, and the net zero goal for the NHS carbon footprint by 2040 was already acknowledged as “ambitious” by Nick Watts, the previous NHS England chief sustainability officer.11 There cannot be a sudden, widespread introduction of an energy and resource intensive solution such as AI, with its associated carbon emissions, without full consideration of how it will limit the NHS’s ability to reach its crucial net zero target within preset timescales.

AI’s integration into health systems must therefore be done critically and with full awareness of the health and environmental trade-offs of these technologies. Health systems, including the NHS, must refute the technological solutionism fallacy—the assumption that technology always equals improvement.1213 Indeed, a recent UK study has shown that newer (and more expensive) technology in the UK doesn’t translate into better health outcomes when resources are diverted away from similar but more effective pre-existing interventions or services.14 Increased rates of imaging, cardiovascular risk prediction, or even transcription of consultations facilitated by AI might not produce better health outcomes or improve quality across health systems. Even if they do, we must determine the environmental cost and consider whether resources would be better spent on low carbon and low cost interventions.

Several further pitfalls warrant careful navigation. Firstly, assessments of risk-benefit must consider whether AI will increase the activity of health systems without a demonstrable improvement in associated health benefits, as the supply of technological solutions drives increasing demand. Secondly, policymakers and commissioners must remain wary of creative carbon accounting practices and carbon offsetting used by tech companies that are used to mask true emissions.47 It is essential that the environmental harms of these technologies remain transparent and appropriately measured. Thirdly, claims by the tech industry that AI will help solve climate change challenges must be approached with healthy scepticism. The promise of solving a problem shouldn’t negate the role a technology or company has in creating or exacerbating the problem.

We must be confident that the health or system gains of AI technology clearly outweigh the environmental and health harms inflicted by the system’s energy and infrastructure requirements. These assessments are complex and demand sufficient resources to ensure that the right decisions are being made. Health systems should also use their purchasing power to ensure that new AI technologies are purchased only from companies that uphold the highest industry environmental standards and commit to transparent reporting. Finally, health systems must work hard to minimise use of AI technologies wherever possible. After all, the first step of environmental sustainability is to reduce. Otherwise, we will be left taking one step forward and two steps back.