Photo Credit: Makenna Hughmanic | The Cauldron
Everything we do online, every video we watch, every post we like, every article we read doesn’t just exist in the non-physical world of the internet. All this data is stored in physical data centers and requires energy, lots of it.
The energy used to train and maintain ChatGPT, a generative artificial intelligence model developed by OpenAI, is more than what the average American household would use in 700 years.
To train GPT-3, Microsoft used 700,000 liters (nearly 190,000 gallons) of freshwater, which is enough to last a family of five 95 years. Projections indicate that by 2027, AI will use more energy annually than the entirety of the Netherlands.
As this new technology becomes more and more widespread, a new discussion arises on the implications of the software’s lack of sustainability and the environmental impact that comes with it.
In the same way your laptop heats up after a five-hour gaming session, so do these data centers. They need to be constantly cooled down in order to avoid shutting down. This constant cooling – you guessed it – requires even more energy, typically coming from fossil fuels.
According to the International Energy Agency, data centers account for roughly 1.5% of global energy use, generating a staggering 3.5% of global greenhouse gasses, amounting to more than the annual emissions of the entire aviation industry.
The average data center uses 300,000 gallons of water a day. To put this number into perspective, this is roughly the equivalent of 100,000 homes, a major contribution to climate change and resource depletion.
Programs like ChatGPT use more energy than the average Instagram post or Youtube video. Generative AI utilizes the data and input it receives to generate content such as text and images.
When you ask ChatGPT to write an essay on Martin Luther King for your history class, it needs to process all the training data to generate an accurate output. This requires a vast input. Because of this, ChatGPT consumes 25 times more energy than a Google search.
As companies race to create bigger and better AI models, each one inching to do more than its predecessors to stay ahead of the rest, more and more models are being trained– a process that can involve these systems running for months upon months with no breaks. Even if future models aim to be more energy-friendly. The sheer number of competition will result in increased e-waste.
Contemporary discourse surrounding AI often focuses on academic dishonesty and the replacement of creative jobs. While the average college student using ChatGPT to cheat on the assignments is plagiarism, fewer of them are aware of the environmental impacts that come with it.
