Many AI critics have pointed out the environmental cost of LLMs. The rapid expansion of data centres, running energy-intensive operations such as training, have led to Google’s emissions jumping by almost 50% in the last 5 years, with similarly high figures for other AI companies.

Defenders of these companies point out that renewables and nuclear will be deployed to meet this energy demand, as well as making the (frankly implausible) claim that LLMs will contribute to new ways to reduce emissions.

Critics have several responses. The first is to point out that here in the present, AI is being powered by fossil fuels — and in at least one case, a coal plant has had its decommissioning postponed due to AI energy demands. Second, renewables aren’t free — they require land, labour, time, and resources to construct. If renewables in some area are simply going to power rapidly expanding AI data centres, then they’re not meeting the existing energy needs of that area — so everyone else will be left continuing to use fossil fuels.

Moreover, efficiency improvements in AI technology can’t solve the problem. Jevons' paradox points out that efficiency gains often lead to higher total consumption of a resource, because greater efficiency = more bang for your buck = higher demand.

AI is not currently providing many benefits for many people — sure it’s helping some people write email more efficiently, and it is assisting some programmers. But most people don’t use AI, and don’t have a use for AI. All AI is doing for them is putting more spam and fake news into their feeds, and making search results worse.

So the obvious way to bring down the emissions from AI is less AI. We don’t need it. Many (most?) of us don’t want it. And if the market doesn’t bring about this change, then we can use policy.

But why limit this critique to AI? There are plenty of industries providing very limited social benefit while being environmentally destructive. These sectors are competing with the more socially necessary ones for the same energy and materials, using up renewables and preventing the decommissioning of fossil power. A big part of the solution is less of the bad stuff, rather than trying to outpace its growth with renewables.