The Environmental Impact of AI Image Generation: What You Need to Know

A Look at the Carbon Footprint of AI Models and How They Affect the Planet

Artificial Intelligence (AI) is transforming the way we live, work, and interact with the world. However, as more and more AI models are developed and deployed, concerns about their environmental impact are rising.

Researchers have managed to assess the carbon impact of each AI prompt. According to this assessment, even a single image generated by AI can be costly for the planet.

TLDR;

• Generating images with AI has a carbon footprint equivalent to charging a smartphone while generating AI-based text consumes significantly less energy. 
• Researchers measured carbon emissions of 13 tasks, finding image generating to be the most energy-intensive and text classification to be the least. 
• They call for transparency in the environmental impact of machine learning models. 
• While charging a smartphone per AI image generated may not seem like high energy consumption, the volume of emissions quickly adds up. 
• OpenAI's chatbot served more than 10 million users daily and now has 100 million monthly active users.

In a collaborative study between Carnegie Mellon University and the AI startup Hugging Face, researchers discovered that generating an image with AI — whether for stock images or realistic identity photos — has a carbon footprint equivalent to charging a smartphone. However, the team found that the energy required for text generation tasks, such as initiating a chat with a chatbot or editing an article, is significantly lower. According to the researchers, generating AI-based text consumes an amount of energy equivalent to just 16 percent of a full smartphone charge.

The researchers measured the amount of carbon dioxide produced per 1,000 grams by examining 13 tasks, ranging from summarization to text classification and image and text generation with machine learning programs. They said they conducted the experiments on 88 different models using 30 data sets to ensure fairness and diversity of data sets. In each task, they ran 1,000 commands while collecting the "carbon code" to measure the energy consumed and the carbon emitted during the exchange.

The findings show that the most energy-intensive tasks ask an AI model to create new content, such as making text, summarizing, captioning, or rendering images. Image generating ranked highest in the amount of emissions it produced, and text classification was classified as the least energy-intensive task.

The researchers call on machine learning scientists and practitioners to "show transparency about the nature and impacts of their models to enable a better understanding of their environmental impact." While charging a smartphone per AI image generated may not seem like a very high energy consumption, the volume of emissions quickly adds up, given how popular and publicly available AI models have become.

Looking at ChatGPT, for example, the study's authors point out that at its peak, OpenAI's chatbot served more than 10 million users daily and now has 100 million monthly active users.

In conclusion, the carbon footprint of AI models is a growing concern, and it's important for machine learning scientists and practitioners to be transparent about the environmental impact of their models. As we continue to develop and deploy AI models, we must also work towards more sustainable solutions to ensure a better future for our planet.