As the University of Ottawa Library’s Digital Literacy Librarian, my role increasingly involves developing students’ AI literacy, their knowledge about AI and skills using it. Within AI literacy, one area of knowledge I believe is particularly important is the environmental impact of AI.
How often do you think about AI’s impact on the environment when answering a question or creating an image? Unless you already pay close attention to environmental issues, probably not that often. It’s tricky because the impact is seemingly invisible. When we use AI, we see no smoke, no waste, no tally of watt-hours. The AI’s answers appear on our screens as if by magic.
But AI isn’t magic. The technology behind it requires real energy which affects our world. A picture is emerging that the energy requirements to train and run AI are massive, not to mention the water needed to cool data centres and minerals to manufacture the technology in those data centres.
To provide some mental footholds, researcher Alex de Vries estimates that AI-enhanced Google searches use ten times the energy of traditional Google searches. And the MIT Technology Review reveals that generating an image with AI uses as much energy as charging your phone. With hundreds of millions of people using AI and billions of Google searches conducted each day, these new energy draws scale fast.
Already, the technology sector’s ability to meet greenhouse gas reduction targets has been impacted. Google’s 2024 environmental report reveals that the company’s emissions have increased nearly 50% since 2019. Microsoft’s 2024 sustainability report similarly reveals that its Scope 1–3 emissions are up almost 30% since 2020.
OpenAI CEO, Sam Altman, says that more energy is needed to power AI. He is looking into nuclear fusion, a not-yet-viable form of energy, to do this. In the meantime, Bloomberg Law reports that dozens of coal plants in the US are being delayed from closing to keep up with increased demand.
Bill Gates claims that AI will provide powerful climate solutions. Perhaps, but how soon can we expect this to be ready? Already we are deep into a climate crisis. How much further in climate debt can we go, waiting for AI to save us?
So, what to do? Ideally, we would prioritize and protect the environment from runaway energy demand. Strong regulation is needed. Also, we would act faster and better. While waiting for large solutions, I will be addressing this topic in my instructions on AI. I will speak on the above to make the impacts known. I will suggest students slow down when using AI. If they plan to generate text or images, I will suggest they take time to imagine the desired output and write a detailed prompt. With luck, this will get them what they need with fewer iterations. I will also suggest students exercise restraint. For a given task, consider whether AI is even needed. Is it the most appropriate tool for the job? And if it’s just about saving time, is the time saved worth the time it may be taking from our future? The impact of these messages may be negligible, I know. But until we are well out of this climate crisis—which won’t be during any of our lifetimes—how can I do otherwise?