ChatGTP & Rubik’s Cubes: AI & the Carbon Puzzle

Words by Clarissa Bell

Ah yes, it’s the obligatory ChatGTP article – even in your beloved environmental university publication!  

With the artificial intelligence (AI) company OpenAI recently developing the language model ChatGTP, it has made countless headlines and sustained media interest through debates about potential mishaps in universities. However, since it blew up online, it seems more and more companies and brands alike are rushing to respond. On one end of the spectrum, Google rushed to release is own rival language, Bard. On the other hand, companies – like Amazon and JP Morgan – banned staff from implementing or using it at all. It’s become so engrained in technological consciousness that Dubai’s Electricity and Water Authority, working with none other than Microsoft, will attempt to use the ChatGTP software to better the experiences of staff and customers. 

All this recent explosion and technological debate from a programme that started out as an opensource chatbot – with undoubtedly huge backings from major companies. Surely this is just incredibly interesting discussion about the potential development of technology and not a topic for an environmental magazine…right? 

Well, with the rise in media opinion about artificial intelligence comes a rise in discussion about the energy it takes to produce and run everyday technology. Not only that, but it’s resurfacing huge ongoing talks about the role of technology in combatting climate change. For example, in March 2021 25 digital technology companies signed the United Nation Environment Programme’s European Green Digital Coalition pact to lower greenhouse gas emissions. Most notably though, was the launch of the United Nations Framework Convention on Climate Change’s (UNFCC) Technology Mechanism at COP27 just a few months ago, seeking to use smart technology in climate resilience and mitigation. 

As the rise in use of ChatGTP coincided with much of the recent discussion about the role of smart technology, especially with its recognition in COP27, there seems to be no better time to ask the question: what is the cost of training technology for climate change mitigation?  

I am by no means the first to ask the question and will certainly not be the last. Even this concern about ChatGTP specifically is found in this article outlining the costs and benefits of using this single software. And from the looks of it, and as usual, it’s complicated. Everything ranging from the hardware used to train it, the time it takes, carbon intensity, how much electricity is produced – a lot of things we don’t even think about. 

However, the iffy variable calculations in other AI training historically have been opaque at best. In January 2020, journalists attempted to investigate another one of OpenAI’s algorithms – not a chatbot, but a Rubik’s cube solver. Pretty amazing right? Surely a computer can be trained like a human brain fairly simply to solve a puzzle with minimal carbon output in the 2020s? Well, it would, if we can figure out how to make the 1,000 computers and 12 industrial machines producing a carbon output equivalent to that of 3 nuclear power plants per hour more efficient. 

This is not to say that greener technology does not have a place in the strive for climate mitigation. We could have an optimum opportunity to talk about and propose new green designs and renewable energy for a lot of commonly used technology, as well as those to train monolithic algorithms. OpenAI stated in 2018 that since 2012 AI training emissions grow exponentially and there is a need to prepare for these implications. But what about making the training process greener? Do we even need vast amounts of algorithms to tell us how best to use them? 

There is so much potential in the technological arena for greener development practices in large companies. For example, Facebook announced a 94% emission reduction rate in April of 2021. Another example: in Colorado, USA, researchers managed to manufacture carbon-neutral cement without the need for quarrying – a huge development for the future of architecture and infrastructure. So, green innovation is not only possibly but underway. 

We’ve got goals to meet and big steps to take if we’re going to continue pushing for the role of technological development in the fight against climate change. OpenAI’s previous statements about the carbon costs of computing already reveal where we need to do the work. 

What better time than now to talk about if technology really knows what’s best for the future of the planet? 

%d bloggers like this: