The development of generative artificial intelligence, and in particular large language models like ChatGPT, will produce up to 5 million tonnes of electronic waste (electronic waste) between 2023 and 2030. According to an analysis conducted by Peng Wang and published this Monday in the journal Nature Computational ScienceIn the scenario with the highest growth of AI and if recycling measures are not increased, these technologies will produce up to 1,000 times more e-waste, an amount greater than that generated by India, the one of the most polluting countries in this type of waste.
Wang and his team estimate that generative AI could generate between 1.2 and 5 million tons of accumulated e-waste. The authors calculated the mass of waste generated by physical elements, such as processing units, storage units and electrical systems. By component, the result indicates that e-waste generated could include 1.5 million tonnes of circuit boards and 0.5 million tonnes of batteries, which may contain hazardous materials such as lead and chromium.
The analysis considers four scenarios with different degrees of production and application of generative AI, ranging from an aggressive scenario (with generalized applications) to a conservative scenario (with more specific applications). The authors suggest that implementing a circular economy strategy (in which the useful life of existing infrastructure is extended and/or key modules and materials are reused in the remanufacturing process) could reduce the production of electronic waste up to 86%.
The importance of recycling
The findings highlight the need for responsible use of generative AI and proactive e-waste management strategies to reduce the harmful impacts of pollution. These results add to previous reports which indicated that companies like Google and Microsoft had skyrocketed their polluting emissions thanks to artificial intelligence, which highlighted the massive water consumption of their data centers or that the training of A single artificial intelligence could contaminate up to 2,800 flights. -Barcelona.
This study will invite further discussion on e-waste, a crucial issue that is often overlooked when talking about AI.
Shaolei Ren
— Professor at the University of California at Riverside (USA)
“The conclusions are based on the best data available in the public sphere and on scientifically valid methods,” he says. Shaolei Renassociate professor of electrical and computer engineering at the University of California at Riverside (United States), in statements to SMC. The study uses Nvidia’s DGX H100 server as a reference to estimate the mass of waste coming from new generation servers, explains the expert.
According to him, e-waste is a crucial issue, often overlooked when talking about AI. “This article draws attention to the e-waste generated by generative AI and I believe it will invite further debate,” concludes Ren. “More importantly, it highlights the crucial role of a circular economy in achieving truly sustainable AI. »
