Artificial Intelligence, the NHS, and Navigating the Path to Net Zero
By Duncan Reynolds
Duncan Reynolds questions whether the environmental impact of AI on the NHS has been sufficiently accounted for.
The challenges facing the NHS today are so great, that it is unlikely that minor policy tinkering will be enough to ensure its survival. One area which is being turned to help revolutionise the NHS is artificial intelligence (AI). Health and Social Care Secretary, Wes Streeting aims to make “the UK a life sciences and medical technology superpower” and he has emphasised the importance of leveraging technology and data to improve healthcare. Currently, AI is used in the NHS in the reading of mammograms, supporting people in virtual wards, GP triage, writing notes, and more. Proponents of these systems argue that they can lead to improved patient outcomes and save time and money in a resource scares environment. The potential impacts, both positive and negative of the increased use of AI in healthcare have been written about elsewhere. However, an area which I believe remains under explored amongst the hype and doom, is the environmental impact of AI on the NHS. If AI is to become an integral part of UK healthcare, it is essential to consider the environmental implications of this technology, particularly in light of the NHS’s commitment to achieving net zero emissions by 2040.
The negative environmental impacts of AI are reasonably well known (although generative AI is still very new so the full impact has not yet been revealed). AI’s environmental impact primarily stems from its energy-intensive processes. Training large AI models, for example, can require several days or even weeks of computing time, depending on the complexity of the model. This process not only consumes large amounts of electricity but also relies on sophisticated hardware, including data centres, Processing Units, and specialised AI processors, all of which have environmental costs associated with production, operation, and disposal. For example, a traditional Google search emits around 0.2g of CO2, where as ChatGPT query emits up to 4.32g of CO2.
With an ever greater need for environmental reporting, much of it enshrined in law, there is increasing attention being paid to AI’s carbon footprint (for example, Deloitte has released an AI carbon footprint calculator). However, this attention appears to have been bypassed in UK healthcare. Currently, the NHS accounts for around 4% of all of the UK’s carbon emissions, and therefore the organisation has a crucial role to play in the UK’s goals of Net Zero by 2050. In order to help achieve this, the NHS became the first healthcare system in the world to commit to carbon reduction targets, with the dual aims of:
- For emissions which can be directly controlled by the NHS, net zero by 2040 (of which 80% will be reduced by 2028-2032).
- For emissions which can be influenced by the NHS, net zero from 2045 (of which 80% will be reduced by 2036-39).
Whilst, there have been previous studies on the environmental impact of different healthcare interventions, notably from the Apollo team’s Stephen Hibbs who showed the carbon footprint per unit of red blood cell transfusion as well as an environmental impact of haematology care , how implementing AI systems into the NHS is compatible with the organisations net-zero strategy is seemingly under-explored. In fact, the NHS Net zero policy document itself only mentions AI in a positive sense, arguing that using AI will contribute up to 2.3% of the total required reduction in carbon emissions. Whilst there may be potential environmental benefits to using AI in the NHS, a failure to recognise the negative impact of these systems could lead to the NHS drastically missing its environmental goals, or the true carbon footprint of the NHS may not be being accurately measured. Both of these would have a substantially negative impact in the fight against climate change.
Achieving net zero goals will require stringent measures across all aspects of healthcare delivery, from supply chain optimisation to greener buildings, and transportation. However, the increased energy consumption linked to AI adoption poses a potential hurdle to these targets. As AI-driven solutions become more embedded in healthcare, their cumulative energy footprint could counteract other emission-reducing efforts within the NHS. Additionally, the indirect environmental impacts of AI deployment, such as hardware disposal, add to the challenge.
To reconcile the adoption of AI with the NHS’s net zero ambitions, urgent action is needed. This will in part be practical, such as greater work in developing greener AI models, promoting sustainable data centres, and exploring responsibly hardware disposable. Other will be focused on reporting and disclosure, such as implementing carbon accounting for AI projects. A lot of further research is needed on how AI emissions will be calculated towards the NHS’s net zero targets, as well as work needing to be done on developing and implementing the technology in as minimally environmentally disruptive manner as possible.