Evaluation of the Energy Impact of Artificial Intelligence
Authored by James Vincent, an experienced journalist specializing in AI and robotics at The Verge for more than eight years.
Assessing the energy usage of AI operations such as those powering email digests, conversational chatbots, and meme-generating algorithms poses a significant challenge. Despite available estimates, the precise magnitude of AI’s energy consumption remains unclear, as major players like Meta, Microsoft, and OpenAI withhold critical data.
The process of training AI models, especially substantial ones like GPT-3, is a notably energy-intensive task, surpassing the electricity consumption of typical daily activities. To illustrate, training GPT-3 consumes approximately 1,300 megawatt hours, equivalent to the annual power usage of 130 US households. In comparison, streaming an hour of content on Netflix requires only a fraction of this energy. The continual growth of AI models in size further complicates predictions of energy expenditure.
The shift from training to deployment, referred to as “inference,” provides insights into the energy dynamics of AI models. Recent research has delved into the energy utilization across various AI functions, highlighting that tasks related to images demand significantly more energy than those centered on text. These conclusions emphasize the necessity for standardized metrics on energy efficiency within the AI sector.
Looking towards the future, experts like Alex de Vries warn of a potential surge in the energy consumption of the AI industry by 2027, potentially rivaling the annual energy requirements of entire nations. As sustainability concerns gain prominence within the sector, the push for energy-frugal AI models and a reassessment of AI’s indispensability in certain domains are gaining momentum.
In the pursuit of a sustainable future for AI, transparency, ingenuity, and mindful energy stewardship are pivotal. By fostering collaborative endeavors to tackle the environmental repercussions of AI technologies, the industry can pave the way for a more sustainable digital environment.
