A research team from Carnegie Mellon University tested 30 datasets using 88 models, looking at energy emissions linked to various AI activities. 

“Every time we query an AI model, it comes with a cost to the planet, and it’s important to calculate that,” said Alexandra Luccioni, leader of the research team, with Tech Xplore. 

According to the researchers, the research was done to dispel the notion that AI is some abstract entity that lives on a “cloud.”

The findings showed that the most energy-intensive AI model was Stability AI’s Stable Diffusion XL, producing nearly 1,600 grams of carbon dioxide per session, which is about the same environmental impact of driving four miles in a gas-powered car. 

Basic text generation tasks have a much smaller carbon footprint, with the amount of carbon dioxide used being comparable to driving 3/500 of a mile. 

One of the revelations revealed in the study was that generative tasks that involve creating new content, such as images and summarizations, are more energy and carbon-intensive than discriminative tasks, like ranking movies.

Additionally, the study noted that using multi-purpose models for discriminative tasks consumes more energy than employing task-specific models for the same activities.

Recommended reading

• What Powers Artificial Intelligence? A Guide for Business
• What is IoT and What Are the Benefits for Local Authorities?
• EU Creates ChatGPT Task Force to Keep Up with AI Developments

“We find this last point to be the most compelling takeaway of our study, given the current paradigm shift away from smaller models fine-tuned for a specific task towards models that are meant to carry out a multitude of tasks at once, deployed to respond to a barrage of user queries in real-time,” read the report.

While the carbon dioxide usage for individual AI tasks may seem small, the cumulative impact, considering millions of users and multiple requests daily, could pose a significant environmental challenge. 

“I think that for generative AI overall, we should be conscious of where and how we use it, comparing its cost and its benefits,” Luccioni said.