We’re getting a better idea of AI’s true carbon footprint
Hugging Face’s comprehensive assessment of its BLOOM model reveals that the environmental impact of large language models extends far beyond mere training energy. By incorporating emissions from hardware manufacturing, infrastructure maintenance, and post-deployment usage, the study demonstrates that the total carbon footprint is significantly higher than initial calculations suggest. This holistic approach highlights the necessity of considering the entire lifecycle of AI systems to accurately understand their ecological cost. The research underscores the critical importance of energy sources in determining AI’s environmental footprint. BLOOM’s relatively low emissions are largely attributed to being trained on a nuclear-powered French supercomputer, whereas models trained in regions reliant on fossil fuels incur substantially higher pollution levels. This disparity implies that the geographical location of data centers and the local energy grid composition are decisive factors in the sustainability of artificial intelligence development. This study is highly relevant to the open data community as it establishes a rigorous, transparent standard for measuring carbon footprints in machine learning. By providing a detailed and honest analysis of BLOOM’s impact, Hugging Face offers a replicable framework that encourages other organizations to adopt similar thoroughness. This shift promotes accountability and guides the open-source AI community toward more sustainable practices, ensuring that the drive for technological advancement does not come at an unchecked environmental expense.
Source: technologyreview.comPublished on 2023-04-11