The increasing prevalence of AI image generator tools in our daily lives may be taking a toll on the environment, according to a recent study conducted by researchers of Carnegie Mellon University and Hugging Face, a machine learning community website.
With over 10 million users engaging with machine learning models regularly, the study provides what the researchers claim to be the first systematic comparison of the environmental costs associated with these models, Tech Xplore reported.
The increasing prevalence of AI image generation tools in our daily lives may be taking a toll on the environment, according to a recent study.Alan Frijns from Pixabay
Is AI Good for the Environment?
Contrary to the common belief that AI has minimal environmental impact, the study reveals that using AI models for image generation consumes an amount of energy equivalent to charging a smartphone.
Team leader Alexandra Luccioni emphasized the need to recognize the environmental costs associated with AI usage, challenging the notion that AI exists as an abstract, cloud-based technology.
The research involved testing 30 datasets using 88 models, uncovering significant variations in energy usage across different types of tasks. The team measured carbon dioxide emissions per task to gauge environmental impact.
Notably, Stability AI's Stable Diffusion XL, an image generator, was identified as the most energy-intensive, producing nearly 1,600 grams of carbon dioxide during a session-roughly equivalent to driving four miles in a gas-powered car.
On the lower end of the spectrum, basic text generation tasks were found to be less carbon-intensive, equivalent to a car driving just 3/500 of a mile. The study encompassed various machine learning tasks, including image and text classification, image captioning, summarizations, and question answering.
The study observed that tasks involving generating new content, such as creating images and summarizations, tend to be more demanding regarding energy and carbon footprint than discriminative tasks like ranking movies.
Additionally, the research emphasized that employing multi-purpose models for discriminative tasks consumes more energy than utilizing task-specific models. This observation holds particular significance given the prevailing trend of adopting models designed to handle multiple tasks simultaneously.
Alexandra Luccioni emphasized the need for conscious decision-making regarding AI usage, particularly when smaller, task-specific models could suffice.
While the individual carbon dioxide usage for AI tasks may seem modest, the cumulative impact of millions of users engaging with AI-generated programs daily raises concerns about its significant contribution to environmental waste.
"People think that AI doesn't have any environmental impacts, that it's this abstract technological entity that lives on a 'cloud. But every time we query an AI model, it comes with a cost to the planet, and it's important to calculate that," Luccioni said in a statement.
The study's abstract also reads: "We conclude with a discussion around the current trend of deploying multi-purpose generative ML systems, and caution that their utility should be more intentionally weighed against increased costs in terms of energy and emissions. All the data from our study can be accessed via an interactive demo to carry out further exploration and analysis."
