Micro-Climate Blog: Digging into Emissions from Artificial Intelligence

By Marcy Rockman, Lifting Rocks Climate and Heritage Consulting, for the SHA Climate Heritage Initiative

One of the many unique capabilities of archaeology is its capacity to make visible things and patterns that otherwise are hard to see. These of course include objects and other remains that are buried; they also include what those objects, remains, and the landscape around them combined can share:  lives and experiences of those not included in written history, events that some parts of society would prefer to forget, and sometimes practices that even in the past were often hidden. 

Which is why I bring an archaeologist’s appreciation to recent reports of the energy and water artificial intelligence (AI) systems are using. If you’re reading this blog, you’re probably seeing many of the same AI prompts that I am – chat boxes showing up unbidden in almost every application, offering to take away the effort of thinking or typing.* To be sure, there are applications of AI and machine learning that are useful. These include analysis of the tens of thousands of climate science articles now being published every year, a rate beyond human capacity to read and synthesize directly. But the everyday tools showing up in so many places, they have the allure of simple magic. What isn’t visible at these interfaces are the quantities of energy and water required for each and every interaction: as the Washington Post (see link below) estimates, a bottle of water and 14 hours of an LED lightbulb for a 100 word email generated by ChatGPT.

 Archaeology is part of exposing social systems that we now understand need our active attention to unwind. As we face the challenges of reducing emissions and improving environmental sustainability, this turning of the spade on what lies beneath AI is also useful. 

*Note: I hereby confirm that none of these blog posts have been written with AI.

Featured Link: https://www.washingtonpost.com/technology/2024/09/18/energy-ai-use-electricity-water-data-centers/

For a listing of all blog posts in this series, visit our Climate Heritage Initiative page.

Photo credit: Illustration of the energy usage of a 100-word email generated by ChatGPT shared by the Washington Post at the link above.