sustAInability (is AI sustainable?)

on

 




Hello friendly readers and welcome back to another article on the environment and living a sustainable lifestyle, this one about social media and AI.

 

I've been curious for a while about the true impact of technology on our planet, and if you're interested in the sustainable potential of other technological advances, do check out some of my other related works such as:

 

 

This time, with the popularity of social media and advancements in AI, I'm researching how sustainable AI is, and what impacts it does/does not have. So first, for a bit of context, I started off by researching the environmental impact of social media..

 

In an experiment conducted by Greenspector, the carbon impact of 12 popular social media apps was measured through use, and then modelled to predict the impact of use over one year. Now this particular study was conducted in France, so there will be a bit of difference depending on the country, but the study does also specify the server location, device use, server type, even network, brightness of device, and repeats carried out.

 

Over an hour, the most impactful app was TikTok at 57gCO2eq, followed by Reddit at 55gCO2eq, YouTube and Instagram at 52gCO2eq. The least impactful was LinkedIn at 28gCO2eq. So, now we have context on social media usage, where does AI fit into this?

 

So first, let's get into how AI functions within society and infrastructure, and then we can begin to understand its areas of impact.

 

Artificial Intelligence, better known as AI, has been used in all sorts of mind blowing and thought provoking ways since its popularity increased, but in order to work, it has to be powered. In a guide written by Forbes, it is explained that traditional data centres run with 5-10kW per rack but with AI, these now need 60+ kW per rack and that AI produces more data than other work, requiring a higher capacity too.

 

As well as this, the increase in intense tasks mean that the machines give off more heat, so it's not just the number and capacity that need to be upgraded, but keeping the data centres cool also has to be a focus- which can be water intensive or use other chemical solutions to achieve temperature control.

 

This is already being noticed, take Google's 2024 Environmental Report for example: In this report, it was appreciated that data centre electricity consumption over 2023 was 17% higher and GHG emissions had risen 13% (compared I assume to 2022). Furthermore, specific note was made on AI increasing the pressure on data centres, reporting that 17% more water had been used to cool data centres in 2023 than 2022- to bring the water consumption to 6.1 billion gallons of water.

 

This is not just specific to one platform. In Meta's Sustainability Report of 2024, it was also reported that the total water consumption for data centres was 2,938 megalitres or roughly 0.8 billion US Gallons by my calculations (okay…google conversion). The report also highlights emission from data centres, which were 12,283 metric tons of CO2 eq in 2023. This may be a random number out of context but the total operational GHG emissions reported in the same year by Meta was 50,610metric tons of CO2 eq, meaning that the data centre emissions were 24% of the total emissions.

 

So we know that AI can put extra demand on data centres, but can they handle it? An article published in October last year by the International Environment Agency tracking the construction of new data centres "driven by growing digitalisation and the uptake of artificial intelligence" highlighted that companies such as Google, Microsoft and Amazon are leaders in using AI and data centres. According to the article, In 2023, the capital investment in AI by these three companies was higher than the entire oil and gas industry of the US.

 

However the big question remains: is expanding AI simply unsustainable?

 

Well according to the International Environment Agency, the current electricity consumption from household appliances is still double that of data centres, and increased efficiency in AI moving forward will mitigate some of the expected expansion from data centres. Furthermore, a study in 2024 published in Nature,  found that AI systems generate less CO2e when used to generate text or images compared to humans doing the same.

 

However, the same article did also note that whilst the AI technology can carry out certain tasks, factors that weren't considered in calculations include social effects, displacement, and legality to name a few. Furthermore, although it is concluded that text generation for example does exude fewer CO2e than humans, it is also appreciated early on in the article that training a system in the first place requires an amount of emissions and power too.

 

In the study published in Nature, it references a study that states the CO2e released training GPT-3 is the same released through the lifetime of five cars, and when it comes to ChatGPT, are emissions  a whole different equation to generating text and images? In an article published by the World Economic Forum (WEF) in July 2024, asking a question to ChatGPT uses ten times the electricity as a Google search. Whilst the Nature study claimed a certain CO2e for training GPT-3, the WEF  article references studies that put a number on it of just under 1,300mWh, and also states that training GPT-4 used 50 times more power according to estimates.

 

So, AI…is it sustainable? It's not an easy question to answer, not least due to the numerous considerations involved. Though basic data centre structure may be in place, it is clear from the studies referenced above that this still needs to expand and adapt to cope with increased AI usage, and this could have repercussions on energy usage and water demand. I have also referenced studies in this article that make commentary on the cost of simply using AI or training it to do the tasks that need to be performed, and yet I don't think I've started to even scratch the surface. For now, I guess I'll keep researching into usage stats, see how the research pans out after new studies and reports come to light, and I will probably bring out an updated study- but please tell me what you think- can AI be sustainable?

 

 

By for now!

Your friendly Mischief Weaver

 

 


Comments

Post a Comment