Artificial intelligence (AI) is rapidly honing its abilities to imitate human creators. Today, generative AI can hold conversations, produce art, make films, and even independently learn to replicate computer games. However, a recent study conducted by researchers at the Chinese Academy of Sciences and Reichman University in Israel highlights a growing concern: AI could also reproduce a less noble trait of modernity, namely the propensity to damage the environment.
The boom in generative AI systems, which include chatbots such as ChatGPT and other content creation systems, could generate between 1.2 and 5 million metric tons of additional e-waste by 2030. The study focuses on large language models (LLM), tools that process and produce human-like texts and interpret the complex statistical relationships underlying language. In addition to offering practical and innovative benefits, generative AI raises a number of philosophical and practical questions, such as the threat to jobs or the risk of manipulation. But now, this technology appears to also be contributing to a serious environmental problem.
Language models and the weight of their infrastructure
Large language models require a powerful computing infrastructure, with complex hardware and advanced chips. The continuous improvements necessary to support the growth of technology are at risk of get worse the issue of electronic waste:
LLMs require significant computational resources for training and inference, and this implies a strong impact in terms of energy consumption and carbon footprint.
Previous studies have focused more on energy impact and carbon emissions, but neglected the physical materials needed for the life cycle of these systems and the electronic waste generated.
Peng Wangan expert from the Chinese Academy of Sciences, calculated that by 2030 the production of AI-related e-waste could reach 5 million metric tons, equivalent to each person who throws away a smartphone. In this extreme scenario, the waste would include 1.5 million tons of electronic boards and 500,000 tons of batteries, potentially harmful to the environment.
The team outlined four future scenarios, with different levels of AI diffusion. In the most advanced scenario, e-waste could reach 2.5 million metric tons per year by 2030. This increase in e-waste would add to the general growth in technological waste, expected to increase by 30%up to 82 million tons.
According to the study, however, there are effective ways to reduce this heavy environmental footprint. The circular economy strategysupported by the International Energy Agency and many companies, offers solutions such as extending the useful life of components and reusing materials in production phases, which could reduce the impact of AI on e-waste by up to at 86%.
