Our thinking

Intensive intelligence

3 March, 2023

Last week, we highlighted some of the social risks inherent in generative artificial intelligence (AI). But what about the climate impact?  

Cloud computing needs electricity to run the servers, water to cool servers and then there are the embodied emissions from all the equipment. And AI is particularly energy hungry. Large language models (LLMs) require vast amounts of energy to train and run. And in the rush to experiment with and utilise services such as ChatGPT, it’s easy to forget that our online, dematerialised activities have real world consequences.  

But it’s possible that AI could be part of the solution, as well as the problem. One significant challenge to a smooth energy transition is the unpredictability of renewable energy sources. When will the sun shine, or the wind blow? Without that, planning and adapting renewable energy capacity and supply is extremely difficult.  

So there is encouraging news on this front from an ongoing trial for an AI-based solution to optimise wind energy management. The AI solution enabled energy company Engie to predict how much electricity generated by wind turbines should be sold on which power markets, at what time, and at what price.  

Perhaps better data and insights could save the UK taxpayer too, who paid £215 million to turn off wind turbines last year, and £717 million to buy gas-powered electricity to make up the shortfall in generating capacity. Sounds crazy, but apparently on the windiest of days, the grid is not able to deal with the power that wind turbines generate, meaning they need to be turned off.  

There are enormous opportunities, and risks, from any new technology, and AI is no exception. Thinking about how we use it – for good – needs to go hand in hand with thinking about how to manage and mitigate its impact.  

By Bertie Bateman