Can AI and sustainability co-exist?

Professor John visits the Indonesian Bureau of Meteorology, Climate and Geophysics (BMKG)’s centre for Earthquakes and Tsunami early warnings to discuss research related to incorporating AI in modeling. This is an example of an application where AI could make a difference between life, and death. Photo credit: Ali Azimi, BMKG

You might think that my work is self-contradictory. As Professor of Data Science for Sustainability and the Environment at Queen Mary University of London, I use artificial intelligence (AI) to address environment-related challenges. Yet a picture is emerging of the negative environmental impacts of AI, as people all over the world incorporate it into their daily lives. So how do I justify using technology that is harming the environment to undo environmental harm? 

On the one hand, AI presents serious environmental risks due to the high energy and water consumption of data centres. On the other, it brings unprecedented opportunities to tackle many of the world's most urgent sustainability issues.  

So, given AI is becoming more and more embedded in how we work and live, we need to ask ourselves: how we can use AI to drive sustainability, while minimising the harm it causes to the environment?  

Why worry?  

It’s becoming ever clearer that the negative impact of AI on the environment is largely to do with its immense energy requirements. Most of this energy consumption occurs during two phases: training the AI models and deploying them for inference (that is, the process of generating responses or predictions, like what happens when you’re waiting for a response to a question).  

Training large models – like those behind ChatGPT or other large language systems – requires vast computational resources. In fact, it can take as long as weeks or months, using supercomputers that require significant amounts of power. Once trained, the models also require substantial energy for inference because they are accessed repeatedly by millions of users. 

This energy consumption wouldn’t be such a problem if our energy grids were fully powered by renewable energy. But fossil fuels still dominate global energy production. Currently, nearly 70% of the world’s electricity comes from non-renewable sources. With experts predicting that AI could cause Europe’s power demand to grow by 40-50% in the next ten years, integrating this technology’s growing energy demand is adding unprecedented strain to an already challenging transition to clean energy.  

Can AI help? 

Despite these challenges and the harm that AI poses to the environment, it also holds immense promise in advancing sustainability, especially in the Earth Sciences.  

For example, AI-driven models are transforming how we predict weather and model climate scenarios. Last year saw a series of truly exciting breakthroughs in the use of AI for weather and air pollution prediction, including Google's GenCast program outperforming the world-leader, ENS, from the European Centre for Medium-Range Weather Forecasts.  

Traditional physics-based models, like ENS and those used at other meteorological offices, are computationally expensive, slow, and energy intensive, meaning forecasts are typically run every six hours.  

Enter AI. As well as predicting both-day-to-day and extreme weather more accurately than current methods, AI can reduce the computational demands of weather forecasting while improving accuracy by providing forecasts hourly – or even at a higher frequency. This means AI could provide faster, more detailed weather predictions at a fraction of the energy cost. It’s a win-win situation that could help identify potential natural disasters before they strike, minimising their impact on the environment and on human life. 

In the Earth Sciences, AI also enables us to analyse decades-worth of previously little used satellite imagery, uncovering patterns that would be impossible to detect manually. This capability to automate tasks supports critical activities like tracking deforestation, monitoring ocean health, and assessing the impact of natural disasters. For example, my research group is using AI to improve our understanding of the earth's climate, develop techniques to speed up the energy transition, improve ocean conservation by tracking coral reefs health, planetary exploration and more. 

In doing this – by more closely studying the environment and climate change using AI technology – we can develop a better understanding of what has happened to date, if / how it can be undone and what can be done to stop – or at least slow down and adapt to – climate change moving forwards.  

Yes, AI can help 

Reducing AI’s negative environmental impact requires a multi-faceted approach. For example, at Queen Mary, we recently refurbished our physics datacentre to improve energy efficiency. We are now using waste heat generated by the computer servers to heat campus buildings, reducing both costs and emissions. We believe this is a model for datacentres across the world, including those used for AI, which shows how technology and sustainability can advance hand-in-hand. 

Innovations in both computing hardware and software are also essential. Advances in hardware, such as quantum transistors that reduce energy leakage, could dramatically cut AI’s energy demands. Optimising software to use less computational power is also critical. These are areas where universities, including Queen Mary, are conducting cutting-edge research. Similarly, the water consumption of AI datacentres could be reduced by improving cooling infrastructures and applying AI to manage water use more efficiently.  

Ultimately, the sustainability of AI depends on greening the energy grid itself. Transitioning to renewable energy sources – such as solar, wind, and nuclear – is crucial and needs to be done at scale. To support work like this, we have recently opened a multidisciplinary Green Energy hub at Queen Mary, which aims to accelerate new ideas and solutions in green energy technologies. 

Equity is vital  

Like any new development in technology, AI’s positive sustainability applications stand to disproportionately serve the Global North, where infrastructure, funding, and expertise are more readily available. This imbalance risks exacerbating existing global inequalities, leaving communities in the Global South without access to tools that could significantly improve their resilience to climate change and minimise its negative impact.  

To address this, we need to insist on equitable access to AI technology. At Queen Mary, we’re working to ensure this happens by collaborating with partners in the Global South. For example, we are working with Sierra Leone to enhance local weather prediction capabilities, which are critical for agriculture and disaster preparedness. Similarly, in Indonesia, we’re training scientists to apply AI in climate research, equipping them with the tools they need to tackle local environmental challenges. 

We also need policies that incentivise energy efficiency and investment in the energy transition. Governments, universities and the private sector must collaborate to ensure that AI is developed sustainably, something we at Queen Mary will continue to lead the way in, by continuing to drive innovative practices, equitable partnerships, and sharing of knowledge.   

Because when it comes to AI and the environment – the challenge is huge, but so is the opportunity.  

Professor John is Head of Data Science for the Environment and Sustainability at Queen Mary's Digital Environment Research Institute (DERI). 

Read more:  

Join DERI 

Collaborate with DERI 

Watch Cédric John’s inaugural lecture, 'Data-Driven Earth Sciences: My Journey from Fieldwork to Artificial Intelligence'.