In our previous article, we discussed why the cloud computing used to train AI isn’t all silver lining, but let’s compare that with a form of data management we’re more familiar with – the internet. While AI’s emissions are concentrated into the hands of (relatively) few researchers and companies who train then repurpose them for general use, the internet has a distributed environmental load – it’s used by everyone with less individual onus.
In 2010, the creator of the world wide web calculated that the average business user’s annual emails have the same emissions as 200 miles in a family car . More recently OVO Energy predicted “If every Brit sent one less thank you email a day, we would save 16,433 tonnes of carbon a year – the same as 81,152 flights to Madrid.” 
As of writing this, ClimaTalk has a website hosted by WordPress.com, a Facebook page, and an Instagram page. For the website – every visit is a cost of almost 6 g of carbon dioxide, or 0.015 miles in a passenger car , putting it in the top 5% of website emissions. The Facebook and Instagram pages are hosted more sustainably than WordPress however, with 0.77 g and 0.87 g per visit .
On another note, Blockchain cryptocurrency like Bitcoin took a wild up and down in recent years , and in that time many that took on the trend started ‘mining’ their own coins . The computing resources required for this has been compared to AI mining – with a 2018 emissions equivalent to 1 million transatlantic flights .
So what we’re seeing is an industry wide trend, with ICT and communications emissions getting too high to ignore . While this may deter many to absolutely reduce our internet and cloud presence, others have continued to use AI for the good it can deliver, on the promise of cleaner data services . In the next article we’ll look specifically at the applications of AI in climate modelling, and how it can influence policymaking. https://www.theguardian.com/environment/green-living-blog/2010/oct/21/carbon-footprint-email