Generative AI burns more energy on non-English text
AI models can use up to 15 times more energy to process and generate text in widely spoken languages, often with lower-quality output.
Generative AI models consume more energy when processing non-English text, with accuracy sometimes declining as a result, potentially exacerbating global inequalities, according to researchers. The disparity is most pronounced in languages without Latin scripts or extensive digital content for training. Energy usage could range from 1.4 to 180 times greater for non-English text, with Hindi and Bengali requiring notably more energy than English.
Naihao Deng, a doctoral student at the University of Michigan AI Lab, notes that accuracy suffers for languages consuming more energy, putting certain users at a disadvantage as AI becomes more ubiquitous. Around 5.2 billion people face double the energy requirements for generating text in their first language compared to English speakers.
Recommendations include optimizing services for additional languages and potentially translating prompts to English, but more resources are needed for languages facing energy and cost disparities. The research was partially funded by the National Science Foundation and DARPA and involved measuring energy usage across three open-source AI models.
Written by urgent.news from Futurity's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
