Researchers Conduct Study on the Carbon Footprint of AI Text and Image Generation

Researchers at Carnegie Mellon University and the AI community platform Hugging Face joined forces for a study on the environmental impact of certain types of AI, Engadget reported first.

The beginning of the study, which was published online, notes the widespread rise of AI products and the systems that power them, the goal to implement the technology “comes at a steep cost to the environment, given the amount of energy these systems require and the amount of carbon that they emit.”

Researchers compared the functions of various machine-learning models, considering both single-task models and those with multiple uses, and logged the cost of the energy and the carbon emitted from them when tasked to perform a given task 1,000 times on a select set of data.

For instance, a model may have to respond to a chatbot, edit an essay, generate a photo or make a work of art. They concluded that when it comes to text-based models, completing an action 1,000 times requires the same level of energy used to charge a smartphone to 16%.

AI images were comparably less energy efficient. While image classification – recognizing certain objects in an image – image generating.

“The least efficient image generation model uses as much energy as 950 smartphone charges (11.49 kWh), or nearly 1 charge per image generation,” researchers reported. They did note that there is a “large variation” between the work of generative models, depending on the size of the image they churn out.

While there is more work to be done on investigating the carbon footprint of AI, “these are important data points that can help inform both our fellow AI researchers and practitioners, as well as policy-makers who are working towards estimating and regulating the environmental impacts of AI models,” they said.