AI Tools Use as Much Energy as Charging your Smartphone

­In case you needed another reason to hate AI, apparently, it’s more of an energy hog than most realize. In fact, according to a study from Carnegie Mellon University, AI tools can use as much energy as charging your smartphone or driving 4.1 miles with a gas-powered vehicle.

While the tech world (and social media) has largely focused on AI-generated images, writing, and even videos from an aesthetic perspective, little attention has been paid to AI’s environmental impact.

As Dr. Sasha Luccioni, who led the study, pointed out, “People think that AI doesn’t have any environmental impacts, that it’s this abstract technological entity that lives on a ‘cloud’.”

While we’ve previously looked at the training phase of the machine learning (ML) model life cycle and its carbon impact, as the study noted, “other phases of the ML model life cycle, such as inference, stand to impact the environment just as much, or more, than training due to the computational resources required to deploy modern models at scale.”

The study looked at tools like ChatGPT by way of functions like Text-to-category, Text-to-text, Image-to-category, Image-to-text, and Text-to-image. Ranking the energy consumption per 1,000 inferences, the researchers were able to quantify the environmental effects of AI tasks.

On the lower end of the energy spectrum, image and text classification tasks used from 0.002 to 0.007 kWh for 1,000 inferences, while text generation and summarization jumped up to around 0.05 kWh hours per 1,000 inferences.

And on the highest end is multimodal tasks such as image captioning and image generation (0.06-2.9 kWh hours for 1,000 inferences). Charging the average smartphone uses about 0.012 kWh of energy.

“The most carbon-intensive image generation model (stable-diffusion-xl-base-1.0) generates 1,594 grams of CO2 for 1,000 inferences, which is roughly the equivalent to 4.1 miles driven by an average gasoline-powered vehicles,” claims the study.