AI’S Environmental Challenges

Enjoying the summer? On balance we’ve had our share of sweltering heat, counterpointed by consistent downpours and bratty algae blooms. Last month saw the hottest day on earth ever recorded. Global warming due to man-made climate change is a wicked problem. Yet scientists are using a critical tool to better understand and combat climate risks, one that’s quickly become ubiquitous in our lives: Artificial intelligence (AI). AI applications are among the most promising tools in the fight. However, critics have warned against AI’s increasing environmental impacts, setting up a kind of “green dilemma” for this highly-touted technology. 

 

AI’s use in sustainability projects has grown and evolved exponentially. Researchers and businesses have developed an array of AI applications, from interpretation of large datasets gathered from satellite imagery to track global methane emissions to shifting through spectroscopic data to improve detection of unharvested rare earth minerals and metals. These metals have wide-ranging industrial applications for sustainable technologies, including manufacture of magnets for wind turbines, inexpensive solar panels, and lithium batteries for electric vehicles. 

AI has helped improve waste recycling, reforestation, and a myriad of agricultural needs such as soil/crop monitoring and pathogen detection, among others. Environmental scientists use AI to enhance climate modeling, improve methods and data gathering in research, and speed interpretation of large datasets to improve policy quality.

 

AI makes for a phenomenal data sifter, but it’s an energy hog! The technology’s environmental footprint is huge. Researchers at the University of Massachusetts atAmherst reported that training a couple large AI models can emit over 626,000 pounds of carbon dioxide – which equates to five times the emissions of an average American car over its lifetime. 

AI technology is also hoggish on resource consumption like fresh water. Data centers house large numbers of computers driven by AI models that sift through data and provide a designed response according to the model. This complex of machinery consumes vast amounts of water for cooling. 

This may be particularly acute where data centers are located in water scarce regions. A researcher at the University of California, Riverside found that training a software model using GPT-3 in a data center evaporated 185,000 gallons of water. GPT-3 is an AI technology that processes text input to generate natural language. You know, things like stories, news reporting, even poetry. Explosive growth in AI use will only exacerbate these problems without creative new ways to make AI use more sustainable. 

 

One novel idea is to turn AI modeling on itself. Aware of the sustainability problem, some data scientists are using a program called Carbontracker to monitor AI training sessions’ environmental impact to some degree. The program is reported to be able to predict energy consumption in these sessions, the carbon footprint, and length of training time. This at least can begin to create baselines for implementing energy reduction measures.

Another ongoing initiative is to design more efficient AI models and algorithms to speed up training to reduce resource use. Better management strategies to balance computing and non-computing resources can further sustainability efforts, especially in terms of taking into account resource disparities geographically. It wouldn’t make sense to increase water footprints of data centers in states in the southwest that are already grappling with chronic overuse of water resources and drought conditions. 

 

We can also take measures to help make our communities more sustainable and climate risk resilient. Chief among them is planting trees. Trees are a natural system for pulling carbon dioxide and other pollution particulates out of the atmosphere efficiently. Unlike AI, which is at least a step away from fighting the effects of climate change, trees directly contribute. Simply, trees store captured CO2 and release oxygen into the atmosphere. Additionally, trees can help prevent flooding and provide wildlife with food and nesting sources. 

Frogtown homeowners can now choose a tree variety for free fall planting by Frogtown Green volunteers. Learn more at:

Rennie Gaither is a Frogtown Green volunteer. Frogtown Green is a resident-led and volunteer-powered environmental initiative in St Paul’s most diverse neighborhood. If you’d like to know more, browse frogtowngreen.com or call 651-757-5970.