Artificial intelligence is becoming increasingly adept at copying its human creators. Generative AI can now conduct persuasive conversations, create artwork, produce films and even learn independently how to reproduce computer games.
Yet a new study from researchers at the Chinese Academy of Sciences and Israel's Reichman University suggests that artificial intelligence may be unintentionally copying a less admirable feature of modern life too: damaging the environment.
Driven by the fast-growing popularity of generative AI tools, including chatbots such as ChatGPT and other content-generation systems, the technology could create between 1.2 million and 5 million metric tonnes of extra electronic waste by the close of this decade.
Generative AI and the growing e-waste challenge
The research concentrates on large language models (LLMs), a category of AI software able to understand and generate human language, as well as carry out connected tasks.
LLMs are trained using enormous text datasets. They detect the statistical links that underpin linguistic rules and patterns, then use those relationships to create comparable material. This gives them striking abilities, including responding to questions, generating images and composing text.
Despite its many advantages, generative AI has prompted a wide range of social, philosophical and practical concerns. These range from worries that AI could replace workers to fears that people could misuse it, that it may deceive us, or even that it could become self-aware and defiant.
As the new research makes clear, generative AI is now also raising concern over the potentially vast volume of additional e-waste it may indirectly produce.
Generative AI depends on rapid advances in technology, both in hardware infrastructure and chips. The upgrades required to match the technology's expansion could worsen existing e-waste problems, the researchers say, unless waste-reduction measures are put in place.
"LLMs demand considerable computational resources for training and inference, which require extensive computing hardware and infrastructure," the study's authors write. "This necessity raises critical sustainability issues, including the energy consumption and carbon footprint associated with these operations."
The researchers point out that earlier studies have mainly examined AI models' energy demand and the resulting carbon emissions. Comparatively little attention has been paid to the physical materials used throughout models' lifecycles, or to the stream of discarded electronic equipment they leave behind.
Forecasting generative AI electronic waste to 2030
The authors, led by Peng Wang, a resource-management specialist at the Chinese Academy of Sciences' Key Lab of Urban Environment and Health, estimated the potential quantity of e-waste generated by generative AI from 2020 to 2030.
They modelled four scenarios, each representing a different level of generative AI production and use. These ranged from an aggressive scenario of widespread adoption to a conservative scenario with tighter limits.
In the more aggressive scenario, e-waste attributable to generative AI could reach 5 million metric tonnes in total between 2023 and 2030. Annual e-waste could rise to 2.5 million metric tonnes by the end of the decade. That is roughly equivalent to every person worldwide throwing away a smartphone.
The high-use scenario further projected that AI-related additional e-waste would contain 1.5 million metric tonnes of printed circuit boards and 500,000 metric tonnes of batteries. These may contain hazardous substances including lead, mercury and chromium.
Only last year, just 2.6 thousand tons of electronic equipment from AI-dedicated technology was thrown away. With overall e-waste from technology expected to increase by roughly a third to an enormous 82 million tonnes by 2030, AI is clearly adding to an already grave issue.
Circular economy strategies for AI hardware
By assessing these scenarios, Wang and his colleagues emphasise a crucial point: generative AI need not inevitably create such a heavy e-waste load.
The researchers note that the International Energy Agency, alongside many technology companies, supports circular economy approaches for tackling e-waste.
The new study identifies lifespan extension and model reuse as the most effective measures. These involve prolonging the service life of existing infrastructure and reusing vital materials and modules during remanufacturing.
The authors report that applying circular economy measures of this kind could cut the generative AI e-waste burden by as much as 86 percent.
The study appeared in Nature Computational Science.
Comments
No comments yet. Be the first to comment!
Leave a Comment