AI and the Environment

In my last article, I mentioned a potential follow-up about the way that AI training can impact the environment and vice versa. However, while researching this I found that I was wholly unqualified to be writing about that, so instead this week I have gathered some interesting related articles and webpages. I am sharing these not to imply that I think AI is bad and we should stop, but because I think it is important to consider the potential hidden costs of new technology so we can work to integrate them into our lives in a way that mitigates unintended negative consequences.

Carbon Footprint

The most obvious connection between AI and the environment is the carbon footprint of storing the data needed for training, actually training the models, and then performing inference (using the models after they are trained). Researchers at Columbia provide estimates for three of those steps, and list some actions that can be taken to reduce them. Forbes and the MIT Technology Review have also published on this topic.

Sand Shortage

Sand, particularly a subset of sand called silica sand, is a necessary ingredient to making the CPUs and GPUs needed for training AI models. However, it is also a necessary ingredient in concrete, and as nations such as China industrialize the demand far outstrips supply. Gangs plunder the natural resources of sand-rich areas in Africa and India, leading to a shortage. While the AI boom itself does not appear to be driving this phenomenon, it will be affected by it. Scientific American covers this in depth, but it is paywalled. Le Monde and something called “GlobeNewsWire” have free articles that are less detailed but still capture the gist of the problem.

Cobalt Mining

Cobalt is a precious metal used in computers and rechargable batteries. American Geosciences reports on the disastrous environmental effects of the mining, while NPR covers the human-rights abuses. In addition, the infrastructure and technology needed to recycle cobalt does not yet exist. Again, AI training is not the sole or even the main cause of this issue, but they are related.