Electric Dreams, Carbon Nightmares: The Hidden Environmental Cost of Generative AI
Image credits: Carolyn Geason-Beissel/MIT SMR | Getty Images
AI chatbots are freaky. To someone like me with little knowledge about technology, programming or any other computer science jargon, I find them incomprehensible.
A strange thing inside my computer, which knows how to simplify my readings, how to solve maths equations, how to quell the torrent of spiteful emails I send to my landlord about the leaky fridge. As it obeys my commands to ‘make this email sound more polite’, I often find myself soothed by its mechanical rearranging of my fiery words into ones which are clinical and assertive. It is probably due to this thing that I have not yet been evicted. Or that I am able to think of my article titles.
Despite all of its pros, chatbots are very scary. Their not-quite-human responses, as well as the unsettling videos and images they produce, provoke a deep fear of the uncanny for many. It is thought that their reinforcement of instant gratification is making people lazier.
As well as this, they are also trained on stolen information from real people, who receive no compensation for their work. Indeed, in 2023 the New York Times sued Microsoft and OpenAI, claiming that their chatbots were trained using articles from journalists at the company without consent, thus violating their copyrights.
As unethical as this all is, the biggest threat we as a species are facing from generative AI is their huge environmental impact. To explain it in the simplest of terms (for people like me with little knowledge of computers) the most complex Ais require the most power. The more power needed, the more carbon emissions are created in the process.
Generative AI refers to artificial intelligence which can answer questions, create images and videos, solve problems and provide ideas. In other words, the chatbots we are so familiar with today. These are the said Ais which require the most power, due to their complex and advanced nature.
These chatbots have not only increased in accessibility, but they are also thrust onto consumers in an attempt for constant innovation. In this case, to ensure search engines keep up to date with the newest technologies and quickest ways to retrieve information.
This becomes an annoyance for two reasons. The first is that often searchers come across inaccurate and often laughable information, as is seen in the case of the viral ‘first person to backflip’ search. When google users went to search the much-asked question, AI overview confidently stated that it was curated by the medieval trickster ‘John Backflip’.
As entertaining as it is to play around with AI and its inaccuracies, every futile search takes up 4 to 5 times the amount of energy as typing your query into a regular search engine. With every search engine now being equipped with an AI chatbot, this poses a huge environmental challenge.
The size of generative AI is measured by parameters, with the larger models being the most advanced, thus taking up the most energy. According to the Scientific American, GPT-3 has a whopping 175 billion parameters. They state that the model went through ‘1287 megawatt hours of electricity and generated 552 tons of carbon dioxide equivalent’. To put this in perspective, this is comparable to the emissions of 123 standard petrol vehicles for a whole year of driving. And with the site reporting approximately 3.1 billion websites in September of 2024, its mind-boggling popularity suggests it will only continue its path of destruction.
Likewise in 2019 it was found that the generative AI model BERT (which was 110 million parameters) depleted the energy of a ‘round trip transcontinental flight for one person’. The lack of tactility in chatbots often means people are unaware that they have a real physical impact. By comparing them to practices that we have known for years cause huge environmental damage demonstrates just how sinister they are.
So how exactly do we avoid this? Chatbots are now seemingly ubiquitous and feel impossible to avoid. But, a recent study by google suggested that size matters less than some think when it comes to sustainable AI.
The research suggested that for the same or similar size, using a ‘more efficient model architecture, processor and greener data centre can reduce the carbon.’ It is clear, then, that change needs to come from those at the top. Companies must invest in more sustainable processes to create generative AI. Public pressure- such as petitions, emails and spreading awareness- can help to achieve this, as well as boycotting sites such as Chat-GPT.
There are also eco-conscious alternatives out there for use, such as the non-profit company ‘Ecosia’, which is equipped with a ‘green filtered’ AI. Promoted by green energy such as solar power, the chatbot also offers sustainable advice and suggestions which are mindful of the planet. Thus, by promoting practices which are environmentally conscious, users are encouraged to incorporate these into their daily lives. As a bonus, Ecosia’s profits are distributed worldwide to support tree-planting initiatives. It is certainly worth switching your browser knowing that each search is not actively aiding the destruction of the planet.
It is easy to feel despondent about the ways in which small parts of our life have a big environmental impact. It can feel particularly overwhelming when software which is so damaging has seen an exponential rise in popularity, which is only getting larger.
However, education is crucial. Knowing what happens as a consequence of your small search allows you to make the first small changes and move forwards making environmentally conscious decisions. So, next time you need a twelve-fingered picture of Jesus in a theme park to send to your uncle on Facebook, do your research before turning to Chat-GPT.
Words by Daisy Morrow
