Why responsible AI is everyone’s priority – Inside track

Governments are scrambling to encourage investment in the AI sector. In the US, President Trump is pushing for a ten year pause on AI regulation. Meanwhile in the UK, the government plans to establish “AI growth zones”, which include the data centres needed to store AI computing infrastructure. But AI is hungry for natural resources. We must carefully understand the full implications of this demand, while also encouraging innovation.

From 2027, at least four billion cubic metres of water will be consumed by AI each year; five times the amount of water consumed annually by the entirety of Denmark. One in four people around the world already lack access to clean water, highlighting the need to manage AI’s growth responsibly. So how can the government align AI innovation with environmental outcomes?

AI is driving innovation, with an environmental toll
AI has the potential to deliver incredible societal benefits. A powerful example of its problem solving abilities is in predicting the 3D structures of over 200 million proteins, accelerating breakthroughs in treating diseases like cancer. Using AI in the public sector, as the government intends, could free up the time 30,000 civil servants spend on administrative tasks and lead to £37 billion in net savings each year.

Applications of AI to sustainability are also on the rise, with tools used to predict the location of deforestation using satellite imagery, improve farmers’ crop yields using weather forecasting and data analytics, and increase energy efficiency for power grids and buildings. But the environmental footprint of AI companies needs better understanding. Data centres run non stop and require constant electricity and water to power and cool computer chips. In the US last year, data centres used enough electricity to power more than seven million homes for the entire year. MIT research demonstrates a worrying lack of transparency by companies about their data centre demands, which makes effective planning impossible for policymakers.

Estimates of data centre emissions vary wildly. Scope 3 emissions, the indirect emissions across the full value chain – including those from e-waste, mining metals for chips and data centre infrastructure – are among the hardest to track, yet account for 40-70 per cent of these companies’ total emissions. While current AI sustainability applications show potential, in some cases they have yet to demonstrate clear, scalable results.

Regulation is starting to catch up
The EU’s Energy Efficiency Directive is a good start, making it compulsory for larger data centres to report information such as water usage and energy consumption back to the EU. In the UK, however, conversations on AI and sustainability are in early stages, with AI infrastructure excluded from national climate and planning policies.

A recent Minderoo Centre for Technology and Democracy report found that AI threatens to place further strain on the UK’s already congested electricity grid. This could undermine the government’s commitment to affordable housing, already delayed for 50,000 families due to grid constraints. Local communities also need to be more included in discussions around AI infrastructure developments.

There are solutions to meet this challenge
Eighty per cent of the UK public think AI regulation is needed. To meet this demand and prepare for the growing environmental footprint of AI, the UK should build upon EU legislation by requiring local data centres to disclose energy and water use, along with Scope 3 emissions, to both the public and the government. This data could be aggregated and anonymised to address concerns around commercial sensitivity.

To further support AI companies and lead on AI sustainability, the government could develop reporting best practices and reward sustainable innovation and research. Suggestions from research done so far include government investments into AI research and R&D, incentivising more efficient computer chip setups and language models, embedding environmental sustainability as a key principle in our AI growth zones and making approval for AI infrastructure dependent on companies sourcing renewable energy. There are also other solutions like data centre heat being used to warm people’s homes, an idea the government is already exploring, and which already happens in Denmark. Joined up regulation would also reduce the uncertainty that currently makes businesses reluctant to invest in AI.

The good news is that we can be cautiously hopeful. That’s coming from a real person, not an agentic AI chatbot. AI could have a remarkable role in accelerating technology innovations and progress towards the climate transition. Recent work by Lord Stern and others suggests that AI applications in areas such as energy forecasting, behavioural change and the development of alternative proteins could deliver a massive 5.4 billion tonnes of global emissions reductions by 2035. This potential is already starting to materialise. Google’s DeepMind recently used AI to identify two million new crystal structures, compared to just 28,000 discovered through traditional research methods in the last decade. These materials could have exciting applications in renewable energy technologies such as solar panels and batteries, and we can expect AI discoveries to grow in impact as models improve.

The UK government is recognising the need for action, with an AI Opportunities Action Plan and a new partnership with OpenAI. These steps could help deliver tangible public benefits, from better healthcare to new jobs. Yet if sustainability remains just a footnote in our AI vision, we risk accelerating into techno-optimistic overdrive and ignoring environmental consequences in the process.