Is AI a climate problem or solution? Environmentally-friendly data and AI practices could drive change for the better

The UN's major climate conference, COP 28, took place last year in a record year of heat and drought, with increasing temperatures and climate crises predicted for the coming years. As the conference ended last December, the UN Climate Change Technology Executive Committee (along with Enterprise Neurosystem, a cross-sector group on AI), launched the AI Innovation Grand Challenge Prize. This prize aims to encourage the development of AI-driven climate action in developing countries. It’s currently open for entries.

Artificial Intelligence is already playing a role in global efforts to solve the climate emergency. For example, AI - and the data that feeds it - is being used to tackle climate-related issues, such as improving crop yields, optimising renewable energy systems and identifying harmful weather patterns. These technologies promise to offer groundbreaking solutions that will help us to manage and mitigate the most harmful impacts of climate change.

However, while new technologies can help to address climate change, they are also adding to the problem, placing additional strain on our resources and planet. According to the Royal Society, in 2020, numerous “studies reported a wide range of estimates of digital technology's contribution to global emissions, varying from 1.4% to 5.9%”. According to one study, training GPT-3 in Microsoft’s state-of-the-art U.S. data centres can directly consume 700,000 litres of clean freshwater. That study suggests the global demand for water for AI could, by 2027, be the equivalent of half the UK’s demand, or that of four to six Denmarks.

In addition to extensive water use, AI’s demands for energy have increased dramatically in recent years. One study points to a 300k-fold increase in AI’s power requirements between 2012 and 2018. Globally, data centres are responsible for consuming over 1% of all electricity - a proportion of which is derived from fossil fuels. Another study projects that within the next decade, AI servers will be consuming more energy than whole countries such as Sweden and Argentina.

As we pointed out in a recent ODI blog, measuring and monitoring the environmental impact of data and technology is vital in supporting that understanding - and is something organisations, governments, and international institutions need to do. Energy consumption, water use and carbon emissions should be reported in a timely manner. This reported data would allow agile policymaking and legislative action. What gets measured gets done!

These facts are alarming. But while AI can certainly be part of the problem, it can also be part of the solution. Innovative cooling methods in data centres, or in tackling some of our most pressing environmental challenges - like managing limited freshwater resources -could help reduce energy use and solve other problems.

A defining factor of AI’s impact on the climate will be the political leadership that shapes and defines the ecosystem within which it exists. Effective incentives could lead us to a new era where AI will be thoughtfully and sustainably developed, and technological advancements can and will be able to help us address the most pressing and harmful effects of climate change. Politicians, innovators and all of us excited by the potential of new technology need to think about the environmental consequences of our decisions around technology.

Our data-centric AI work recognises that generating high-quality data for AI and working to ensure this is used responsibly and equitably is critical to the success of AI. Global efforts aiming to develop AI to tackle climate change should consider which data is available to support these endeavours, who can use it, and how it works towards the principles of responsible AI.

For example, timely and accurate measurement and reporting doesn’t just help to inform governments and organisations about necessary policy and legislative interventions. It can also enable a widespread understanding of the true environmental cost of significant data consumption and AI use. It could help drive a cultural change, leading to better leadership, encourage a greater focus on AI transparency, and incentivise those developing new technologies to incorporate climate considerations into existing efforts to improve sustainability. It could also catalyse a global approach to preventing adverse environmental consequences, and boost innovation in sustainability practices around data collection, sharing, and use.

Internationally, governments must now collaborate to lead in the building and maintenance of international data infrastructure. In the UK, the new Government must seize the opportunity to lead the way in three key areas. First, by sponsoring and promoting open standards for measuring the environmental impact of data and AI throughout the development and deployment of such technology. Second, seeking to implement robust data governance aligned with net zero goals. A useful first step would be to produce the Data Sustainability Charter, promised by the National Data Strategy in 2020, to ‘inform how government works with its suppliers to manage and use data sustainably’. Finally, building on that, there is an opportunity for the Government to demonstrate leadership on the environmental impact of AI through its own practices; putting environmental sustainability front and centre of all its data and AI policy work. This could include tasking the AI Safety Institute with the assessment of environmental, as well as other impacts, of AI systems.

The King’s Speech set out the intention to require water companies to install real-time monitors at every sewage outlet with data independently scrutinised by the water regulators - a positive step in the right direction to use data to tackle environmental harms. However, we believe the government could, and should, take further robust steps to show true leadership as part of its mission on clean power and net zero - steps that can really drive change for the better when it comes to data and the environment.

Let’s ensure we build on the progress thus far by embedding environmentally-friendly data and AI practices - we all have our parts to play in saving our planet.