The CO2 emissions from generating one AI image are equivalent to charging a smartphone

Jonathan Kemper/Unsplash

Experts from startup Hugging Face and Carnegie Mellon University have calculated the carbon footprint of AI models while performing various tasks. The biggest greenhouse gas emissions are associated with image generation.

Here's What We Know

The study looked at 13 machine learning tasks. To keep the study fair and the datasets diverse, the researchers ran experiments on 88 different models using 30 datasets.

For each task, the researchers performed 1,000 prompts, collecting a "carbon code." They then measured the energy consumed and the carbon released in the process.

According to the results, creating a single picture or photo requires as much energy as charging a smartphone. But generating text, for example, dialogue with a chatbot, is equivalent to charging a phone only 16%.

At the same time, popular models like DALL-E or ChatGPT are used daily by tens of millions of people. Thus, the total energy consumption and carbon dioxide emissions increase manifold.

Experts urge AI developers to increase the transparency of the environmental impact of their models. This will help to better monitor and reduce the negative impact of technology on the environment.

Source: Engadget