It’s been my experience that when people think about cool new software advancements, they rarely think about the impact on physical hardware and what it takes to support that hardware. In a recent IEEE article titled “Generative AI Has a Massive E-Waste Problem” Katherine Bourzac does note the increased requirements of water and electrical grid load. The focus of her article is not on this environmental impact, nor the need for more raw materials to build new servers. Rather, her focus is on the need for evermore capable hardware platforms to run the evermore sophisticated algorithms that search growing databases. Predictive AI ‘learns’ patterns from all sorts of information sources. This approach is known as a large language model or LLM. As we collectively speed up our upgrades, Bourzac notes that it could result in 2.5 million tons of waste increase of old machines we get rid of in the process.
The author reminds us that, “electronic waste contains toxic metals and other chemicals that can leach out into the environment and cause health problems.” She shares that a staggering 62 million tons of e-waste was produced worldwide in 2022 alone. That means our e-waste is growing five times as fast as our ability to recycle it. To make it worse, Bourzac was only considering studies associated with LLM models of AI. There are others as well.
Despite the fact that our machines and the chips that run them get more efficient over time, the volume of systems is still growing. Bourzac recommends ‘downcycling’, meaning repurposing servers for more simple tasks such as web hosting. That prolongs usage so long as the equipment continues to function. At some point though, it will still become waste. Large tech firms have announced ‘sustainability goals’, but these generally relate to carbon footprint and not so much about e-waste.
Here's the link to the article:
https://spectrum.ieee.org/e-waste
The author reminds us that, “electronic waste contains toxic metals and other chemicals that can leach out into the environment and cause health problems.” She shares that a staggering 62 million tons of e-waste was produced worldwide in 2022 alone. That means our e-waste is growing five times as fast as our ability to recycle it. To make it worse, Bourzac was only considering studies associated with LLM models of AI. There are others as well.
Despite the fact that our machines and the chips that run them get more efficient over time, the volume of systems is still growing. Bourzac recommends ‘downcycling’, meaning repurposing servers for more simple tasks such as web hosting. That prolongs usage so long as the equipment continues to function. At some point though, it will still become waste. Large tech firms have announced ‘sustainability goals’, but these generally relate to carbon footprint and not so much about e-waste.
Here's the link to the article:
https://spectrum.ieee.org/e-waste
RSS Feed