As Instagram reels and TikTok skyrocket in popularity, many educators have joined these social media platforms as a way to spread information about science, history, and more. I, myself, have followed many climate communicators, which brought my attention to the relationship between the burgeoning use of artificial intelligence and the imminent climate crisis.
At a surface level, the most obvious effect of AI on our environment is its infrastructure. To train these models, large plots of land must be cleared to build the large computers needed to run intelligence software. Northern Virginia, the global hub of these data centers, houses 47.7 million square feet of data center space, however, less than 5% of the energy needed to run these computers comes from renewable sources. Researchers at the University of Massachusetts, Amherst estimated that training a natural language processing AI model, such as Siri or Alexa, would produce 300,000 kilograms of carbon emissions, equivalent to 125 round-trip flights from between New York and Beijing. In addition, the exploitation of human labor to mine precious minerals needed in microchips and processors is further exacerbated by the uptick in AI consumption. The International Energy Agency estimates that should an AI model such as ChatGPT be integrated into the 9 billion searches completed every day, demand for electricity would increase by 10 terawatt-hours per year, the amount used by around 1.5 million European Union residents.
Digging deeper into its effects on the environment, we can see that the environmental effects of AI have taken root in places never expected. For instance, powerful tech companies often market their AI models to oil and gas companies to optimize oil extraction and production. In 2019, Microsoft announced a partnership with ExxonMobil, allowing the oil company to use its cloud software for mining operations. ExxonMobil claimed that by using the AI, it could increase its production to 50,000 oil-equivalent barrels per day by 2025. Moreover, the use of AI to target users with curated ads encourages a consumerist mindset. The fast cycling of trends online leaves no room for outfit repeating, often ending with tons of clothes in landfills.
With all this said, what can be done to mitigate these consequences? An article written by Roel Dobbe and Meredith Whittaker of NYU’s AI Now outlines several solutions to the question. First, government-mandated transparency. Most AI companies are reluctant to share what types of energy are used to power their computers. Compared to the aerospace industry, whose energy efficiency is straightforward due to “standards…and reports about what hardware planes use, how far and how long they fly”, getting energy usage data from big tech companies is difficult. Alexandra Luccioni of Mila-Quebec AI Institute argues that tax incentives for cloud providers to build in places with hydro or solar power would help minimize artificial intelligence’s negative environmental impacts. In addition to making energy efficiency a criterion when evaluating AI’s climate effects, we must also take into account the “full stack supply chain.” By evaluating the entire progression of AI models, from material miners to software developers, researchers can get a better understanding of the environmental and labor resources required to develop, produce, maintain, and dispose of a product. Finally, the integration of climate policy and tech regulation would further help alleviate the damaging effects of AI on the climate. Policy committees are frequently isolated to their specific issue, but global concerns often overlap with each other. Collaborating will bring these conversations together, creating a greater possibility for positive change in our ever-changing world.
Additional References:
- Dhar, P. The carbon impact of artificial intelligence. Nat Mach Intell 2, 423–425 (2020). https://doi.org/10.1038/s42256-020-0219-9