The Environmental Cost of AI

With the UK Government's recent announcement about investing heavily in AI, it's appropriate to ask: what is the environmental footprint of all the computing infrastructure that this AI requires?

The answer is simple: it's massive. While there are huge uncertainties, some of the projections have the AI boom causing data centres to generate more carbon emissions than Africa and South America combined. 

The recently announced plans for AI expansion lead to huge increases in resource requirements, and on relatively short timescales. The power needed by AI might exceed that used in all the world's current data centres within 3 years.

That's going to need a lot more electricity, which is likely to increase carbon emissions in the short term, before renewable generation ramps up – and that’s not to mention the carbon footprint of putting up such massive buildings to house the servers and manufacturing the servers that go into them. 

For the longer term, some of the tech giants are starting to look at small modular reactors (SMR, i.e. small nuclear reactors) to power their data centres. Microsoft is trying to bring one of the reactors at Three Mile Island back online. Even if you can generate the electricity, generally there isn't enough grid capacity to get it to the right place, which is one reason why there's such interest in putting the generator, such as an SMR, right next to a data centre.

It's not just drawing electrical power that's a problem. Gargantuan amounts of water, already a scarce resource, are needed for cooling, as all this electricity gets converted into heat.

All this environmental damage leaves open another question: will the benefits from AI outweigh the environmental damage it causes?

For a good overview of the issues about AI and sustainability, this report from the Green Web Foundation is a good read, and this article from UNEP closes with five recommendations to “rein in the environmental fallout of AI”. A great book on the topic is the “Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence” by Kate Crawford.

To find out how to turn off AI in your search engine or on your device, just search the phrase “how to turn off AI on…” and instructions often come up.

Blog post by Peter Tribble. Peter also has a personal website: https://ptribble.blogspot.com/

Improving your web experience - and saving carbon emissions

We spend a lot of our time browsing the web, but have we thought of the carbon emissions involved in getting a web page to be shown on our screen?

There's a long chain involved. The web page has to be stored on a server, potentially processed for consumption, sent out across the network, being passed through multiple devices on its path to you, before finally being processed by your own device so as to look just right for you to read it.

Fortunately, much of the work involved in calculating the environmental impact of the web has been done for you. Sites such as Website Carbon and Ecograder can estimate the carbon emissions due to viewing a web page. This takes a range of factors into account, from storage and processing of the page, transmitting it across the network, but also whether the server is powered by electricity from renewables or fossil fuels.