Generative AI has significantly altered the digital world, but its environmental impact is becoming increasingly evident. A recent study by Hugging Face and Carnegie Mellon University highlights this concern, revealing that generating an image using a powerful AI model consumes as much energy as fully charging a smartphone. Interestingly, generating text is less energy-intensive, with 1,000 AI-generated texts using only 16% of the energy required for a full smartphone charge. This study, still awaiting peer review, underscores that the substantial carbon footprint of AI is largely due to its usage, not just its training.
Sasha Luccioni from Hugging Face, who spearheaded this research, aims to raise awareness about the carbon emissions from various AI tasks, which could lead to more eco-conscious AI applications.
The team evaluated emissions from 10 AI tasks, including image generation, which emerged as the most energy-intensive, equating to the CO2 emissions of driving 4.1 miles in a gasoline car. On the other end, text generation was found to have minimal impact. The findings, backed by other experts like Lynn Kaack and Jesse Dodge, suggest that the integration of large generative AI models in everyday tech applications, such as email and search engines, has significantly escalated their carbon footprint. The study also found that for popular models like ChatGPT, usage emissions can rapidly surpass the emissions from their initial training.
Dodge, who is a research scientist at the Allen Institute for AI, pointed out that the energy usage of AI tools has been an overlooked factor in comprehending their overall carbon footprint. Dodge was not involved in the study.
Dodge thinks that comparing the carbon emissions from newer, larger generative models and older AI models is also important, though added [that] “it highlights this idea that the new wave of AI systems are much more carbon intensive than what we had even two or five years ago.”
Vijay Gadepally, a research scientist at the MIT Lincoln lab, who did not participate in the research, believes that studies highlighting the energy consumption and emissions related to AI make the environmental impact of these technologies more tangible. He emphasizes the importance of awareness regarding the carbon footprint associated with AI use. Gadepally expresses his hope that consumers will begin to inquire about this aspect of AI technology.
Dodge further expressed his hope that studies of this nature will aid in holding companies more accountable for their energy usage and emissions.
