AI adds another problem: it generates (and will generate) a lot of garbage

The development of more and more artificial intelligence applications will also lead to exponential growth in the amount of e-waste, waste that typically contains toxic metals and which typically ends up in landfills in countries with lax regulations, warns a study published in Nature Computational Science.

According to experts, unless action is taken to reduce the amount of e-waste produced by artificial intelligence systems, we could reach the 2030s with approximately 2.5 billion tons of waste per year, an increase of almost 1,000% from the previous decade. tons announced in 2023.

The conclusion, reached by researchers from China and Israel, increases pressure on an industry already under question over its climate impact, mainly due to the amount of electricity required by the data centers that power applications such as ChatGPT or Copilot from Microsoft. Statista estimates that by 2026, the electricity consumption of these applications will be equivalent to a quarter of total US consumption.

“I am very optimistic about the opportunity to advance circular economy strategies among the major players involved in generative artificial intelligence,” he says. Country Peng Wang, one of the study’s authors and a researcher at the Institute of Urban Environment of the Chinese Academy of Sciences. “However, I have deep concerns about the competition between the pace of AI adoption and the adoption of the circular economy. Given the unprecedented rise in demand for this technology, shock measures must be taken immediately to win this battle,” he adds.

Circular economy as a palliative

The study authors explore some alternatives to reduce the generation of e-waste generated by the development of artificial intelligence. Strategies relate to the circular economy and range from extending the lifespan of components used to implement AI models to recycling some of their parts.

According to experts, the lifespan of GPUs used in artificial intelligence servers is three to five years. However, their study found that 62% of AI servers are updated every year.

On the other hand, another option would be to recycle some of the GPU components, such as memory modules or conductive metals (such as gold) in the connectors. However, this process is quite expensive and complex. “GPUs cannot be used in a circular economy because recycling their components is very expensive,” he explains to the publication Country Ana Valdivia, professor of artificial intelligence, government and policy at the Internet Institute at the University of Oxford, is an expert who was not involved in the study but was consulted by the Spanish newspaper.

Outside of the study, perhaps another alternative would be to figure out what the ambitious goals are for the development of this technology, which, although promoted by major companies around the world, has known skeptics, including the creator of Linux.