The Everytime generative artificial intelligence The planet is paying for the e-mails or images that are created. Two images may consume the same amount of energy that would be required to charge a smartphone. A single chat with ChatGPT could heat up a computer so much it needs a full bottle of water in order to cool. These costs increase as the scale increases. According to estimates, by 2027 the AI industry could consume the same amount of electricity annually as the Netherlands. one recent estimate .
A new study has been published in Nature Computational Science Another concern is identified The enormous contribution of AI to electronic waste. According to the study, generative AI could contribute 1.2 to 5 million tons of hazardous waste to our planet between 2030 and 2050 depending on the rate at which the industry expands.
This contribution would be added to the millions of tonnes of electronic waste that the world discards each year. Many digital devices, such as cell phones, micro-wave ovens, and computers, contain toxic substances like mercury or lead. If improperly disposed of, these products can pollute air, soil and water. In 2022, the United Nations estimated that 78 percent (or a little more) of all e-waste in the world ended up on landfills and unofficial recycling centers where workers risked their health in order to collect rare metals.
This AI explosion is consuming a lot of data storage units, graphics processors and other components that are high performance. They can process thousands or calculations at once. Hardware can last anywhere between two and five years, but is often upgraded as soon as a newer version becomes available. Asaf Tachor, sustainability researcher and co-author of the study at Israel’s Reichman University says that the findings highlight the importance of monitoring the environmental impact this technology has.
Tzachor, along with his co-workers, examined how long these parts last, as well as the rate at which the sector is growing. Researchers warn that the prediction they made is only a rough estimate and could be affected by a number of additional factors. For example, more people may adopt generative AI that the models predict. In the meantime, hardware design improvements could help reduce the amount of e-waste generated by a particular AI system. Other technological advancements can also make AI systems more affordable and accessible, thereby increasing their use.
Related: A new memory chip that is controlled by magnets and light could make AI less energy-hungry one day
The study is most valuable because it focuses on AI’s environmental impact, according to Shaolei Ren. She studies AI responsible and wasn’t involved with the research. “We might want these [generative AI] companies to slow down a bit,” He says.
Few countries require the correct disposal of electronic waste, and even those who do so often don’t enforce existing laws. 25 U.S. States have policies for e-waste disposal, but no federal laws require electronics recycling. Ed Markey, a Massachusetts Democrat Senator who is a member of the Senate’s Democratic Party, introduced in February a bill that requires federal agencies to develop and study standards regarding AI’s environmental impact. This includes e-waste. The Artificial Intelligence Environmental Impacts Act 2024, which hasn’t passed the Senate yet, would not require AI developers to participate in its voluntary reporting system. Nevertheless, some companies claim that they are taking independent actions. Microsoft and Google both pledged that they would reach zero waste by 2030 and zero emissions, respectively. This likely involves reducing AI-related electronic waste or recycling it.
Receive the most exciting discoveries from around the globe delivered to your email.
AI-based companies have many options for reducing eWaste. You can extend the life of your servers by performing regular maintenance, updating them, or repurposing worn out devices for less intensive applications. Tzachor’s study and that of his co-authors show how refurbishing and reusing outdated hardware components could also reduce waste by 42 per cent. A more efficient algorithm and chip design can reduce the hardware and electrical requirements of generative AI. The study’s authors claim that combining all of these strategies could reduce electronic waste by up to 86 percent.
Kees Blde, a researcher on e-waste at the United Nations Institute for Training and Research who was not involved in the study, points out that AI products are more difficult to recycle because they often contain sensitive data about customers. He points out that big tech firms can afford both to erase the data as well as properly dispose of electronics. “Yes, it costs something,” He says that he is in favor of recycling e-waste more widely. “but the gains for society are much larger.”
