DeepMind on building responsible AI for a sustainable future

A Climate Compass conversation with DeepMind’s Ben Gaiarin on ethics, energy, and impact

As the climate crisis intensifies, Artificial Intelligence is becoming an indispensable tool — not just for emissions reductions, but for understanding and managing climate risk. At Sweep’s Climate Compass thought leadership event, Sweep Co-founder and CPO Raphael Güller sat down with Ben Gaiarin, Technical Program Manager at Google DeepMind, to explore how AI can support climate adaptation responsibly.

Gaiarin is part of the team leading DeepMind’s sustainability research programme, including its flagship initiative, WeatherNext. This project doesn’t just showcase cutting-edge AI innovation, it’s also a powerful case study in how machine learning can help us anticipate and respond to extreme weather, one of the most visible and dangerous consequences of climate change.

AI as early warning system

At the heart of WeatherNext is GraphCast, an AI model that replaces traditional physics-based simulations — models that require supercomputers and enormous processing power — with a machine learning system trained on historical weather data.

The initial results are impressive. Forecasts that once took hours across tens of thousands of processors can now run in eight minutes on a single TPU (Tensor Processing Unit – Google’s custom-developed, application-specific integrated circuits used to accelerate machine learning workloads). That means faster warnings, more frequent updates, and far lower energy consumption.

“We replaced the entire simulation stack with a learned model—and we can predict 15 days ahead in minutes. That speed and efficiency could be game-changing in disaster preparation.”

DeepMind’s new Weather Lab site brings this to life. Launched just days before the Climate Compass event, it lets users explore how different AI models compare on cyclone path prediction, a valuable asset for businesses, government agencies and private individuals when anticipating hurricanes and other high-impact events. Google DeepMind’s cutting-edge tool is now being evaluated in partnership with the U.S. National Hurricane Center to assess its usability in real-world forecasting.

This kind of advancement could save both lives and livelihoods, particularly in vulnerable regions. And it’s a prime example of “responsible AI” — technology developed not because it’s possible, but because it’s urgently needed.

Climate mitigation through smarter prediction

Better forecasts also help cut emissions. Gaiarin highlighted how high-precision weather models can improve power forecasting, allowing grid operators to rely more confidently on wind and solar energy.

Several studies show that improved weather models directly correlate with increased renewable energy uptake. For grid operators balancing volatility, a narrower “cone of uncertainty” allows more aggressive use of clean energy — meaning lower greenhouse gas emissions and more system stability.

For Gaiarin, this is the kind of “carbon return on investment” that justifies the energy used to train and run AI models. And it’s central to the case for building AI responsibly: not just as a technology, but as a driver of smarter systems that actively reduce environmental impact.

Mindful innovation: AI at the right scale

Beyond the specific use case of energy grid efficiency gains, Ben Gaiarin acknowledged that the trade-off between the large energy and water use of large models, and their potential climate benefits, can’t be ignored. 

Google DeepMind, and the broader AI field, is seeing significant gains in efficiency, particularly in how models are served. But it’s the impact per watt that ultimately defines responsible innovation. In fields like genomics, weather, energy, and others, AI has the potential to radically accelerate research and deployment of climate solutions.

That acceleration, Gaiarin argues, is worth the investment, but only on the condition at the AI is applied with intention, transparency, and meaningful evaluation.

What responsible AI looks like in practice

When asked what advice he’d give to sustainability leaders adopting AI tools, Gaiarin emphasized the importance of thorough evaluation and establishing clarity of purpose.

He also encouraged leaders to think beyond today’s common use cases. “It’s not just about chatbots,” he said. “Imagine what happens when AI can generate entirely new scientific algorithms or simulate energy systems more accurately than any human modeler could.”

In other words, the future of AI lies not just in automation, but in discovery.