Generative AI could generate millions more tonnes of e-waste by 2030

Generative AI poised to worsen e-waste crisis

Written by Saima S. Iqbal

Every time a generative artificial intelligence composes an email or envisions an image, the planet pays for it. Creating two images uses as much energy as charging a smartphone. A single exchange with ChatGPT can cause the server to heat up so much that it may require an entire bottle of water to cool down. As scale increases, these costs increase. Recent estimates suggest that by 2027, the global AI sector could consume as much electricity per year as the Netherlands. And in new research natural computational science reveals another concern. That’s AI’s huge contribution to the growing mountain of e-waste around the world. The study found that depending on the rate of industry growth, generative AI applications alone could add between 1.2 million and 5 million tons of this harmful trash to the planet by 2030.

These contributions add to the tens of millions of tons of electronic products that are discarded on the planet each year. Cell phones, microwaves, computers, and other ubiquitous digital products often contain mercury, lead, and other toxins. Improper disposal can contaminate air, water, and soil. The United Nations found that in 2022, around 78 percent of the world’s e-waste will end up in landfills or informal recycling facilities, where workers risk their health to recover rare metals.

The global AI boom is rapidly moving through physical data storage devices, as well as graphics processing units and other high-performance components needed to handle thousands of simultaneous calculations. This hardware has a lifespan of 2 to 5 years, but is often replaced as soon as a new version becomes available. Sustainability researcher Asaf Tzakor of Israel’s Reichmann University, a co-author of the new study, said the findings highlight the need to monitor and reduce the environmental impact of this technology. .

About supporting science journalism

If you enjoyed this article, please consider supporting our award-winning journalism. Currently subscribing. By subscribing, you help ensure future generations of influential stories about the discoveries and ideas that shape the world today.

To calculate how much generative AI is contributing to this problem, Tzachor and his colleagues looked at the type and amount of hardware used to run large-scale language models, and the persistence of these components. We looked at the time and growth rate of the generative AI sector. The researchers cautioned that their predictions are rough estimates and are subject to change based on several additional factors. For example, more people may adopt generative AI than the authors’ models predict. On the other hand, innovations in hardware design may reduce e-waste for certain AI systems, while other technological advances may make systems cheaper, more accessible to the public, and less likely to be used. The number may increase.

The study’s greatest value lies in its focus on the broader environmental impacts of AI, said Xiaolei Ren, a researcher at the University of California, Riverside who studies responsible AI. He was not involved in this new study. “We might want these (generative AI) companies to slow down a little bit,” he says.

Few countries mandate the proper disposal of e-waste, and many fail to enforce current laws regarding e-waste. Although 25 states in the United States have electronic waste management policies, there is no federal law mandating the recycling of electronics. In February, Massachusetts Democratic Sen. Ed Markey introduced a bill that would require federal agencies to research and develop standards for the environmental impacts of AI, including electronic waste. However, the bill, the Artificial Intelligence Environmental Impact Act of 2024 (which has not yet been passed by the Senate), would not force AI developers to cooperate with the voluntary reporting system. But some companies say they are acting independently. Microsoft and Google have pledged to achieve net-zero waste and net-zero emissions, respectively, by 2030. This could include reducing or recycling AI-related e-waste.

Companies using AI have many options to limit e-waste. For example, it is possible to extend the lifespan of servers through regular maintenance and updates, or by migrating aging devices to less demanding applications. Refurbishing and reusing old hardware components can also reduce waste by 42%, Tzachor and his co-authors say in a new study. And more efficient chip and algorithm designs could reduce the hardware and power demands of generative AI. The study authors estimate that all these strategies combined will reduce e-waste by 86%.

There is also another problem. AI products tend to be more difficult to recycle than standard electronics. Because the former often contains a lot of sensitive customer data. said Keith Balde, an e-waste researcher at the United Nations Institute for Training and Research. Not involved in new research. But big tech companies can afford to erase that data and properly dispose of electronic devices, he points out. “Sure, it costs some money,” he says of widespread e-waste recycling. “But the benefits to society are far greater.”