The environmental impact of generative AI

I’ve had Chomsky’s line that generative AI is “basically just a way of wasting a lot of the energy in California” reverberating in my mind for the last few days. In his excellent Resisting AI Dan McQuillan observes that the refinements involved in training “are for the sake of the very marginal improvements in overall accuracy, more related to cut-throat competition between industry research labs than practical utility”. Given the “innate centralisation and the barriers to entry it creates” when “AI research is largely privatized, or at least wholly dependent on the cloud computing resources” of the big four (pg 23), the ecological wastefulness of technological competition is now the battlefield on which big tech firms will fight for dominance in the least congenial economic climate they’ve enjoyed since the early 90s. He cites two papers in relation to this which I’m making a note to come back to:

• https://www.ourwebofinconvenienttruths.com/wp-content/uploads/2015/02/ICT-Global-Emissions-Footprint-Online-version.pdf
• https://arxiv.org/pdf/1906.02243.pdf%22%3EWachstum