Researchers at artificial intelligence initiative Hugging Face, in collaboration with Carnegie Mellon University, have revealed that creating an image using artificial intelligence to create either ready-made images or realistic ID photos has a carbon footprint equivalent to charging a smartphone. However, the researchers say that creating text prompts, such as creating a conversation with a chatbot or editing an article, requires much less energy than creating a photo. Researchers state that the text generated by artificial intelligence consumes enough energy to charge a smartphone to only 16 percent of full charge.
The researchers measured the amount of carbon dioxide produced per 1,000 grams by examining a total of 13 tasks, from summarization to text classification, as well as image and text production with machine learning programs. They said that they carried out the experiments on 88 different models using 30 data sets in order to ensure the fairness of the study and the diversity of the data sets. In each mission, they ran 1,000 commands while collecting the “carbon code” to measure both the energy consumed and the carbon emitted during the exchange.
The findings show that the most energy-intensive tasks are those that ask an AI model to create new content, such as text generation, summarization, image captioning, or rendering. Image rendering ranks highest in terms of the amount of emissions it produces, and text classification was classified as the least energy-intensive task.
The researchers call on machine learning scientists and practitioners to “demonstrate transparency about the nature and impacts of their models to enable a better understanding of their environmental impacts.” While charging a smartphone per AI image created may not seem like a very high energy consumption, the volume of emissions easily increases considering how popular and publicly available AI models have become.
Looking at ChatGPT, for example, the study’s authors point out that OpenAI’s chatbot served more than 10 million daily users at its peak and now has 100 million monthly active users.
32163
