I’ve found myself looking for this quote 3 times this month, and LinkedIn is a real pain to use to find it, so I’m posting it here.
This quote from David Rolnick, one of the founders of the non-profit Climate Change AI is a useful one to have handy, because its coming from someone with deep domain knowledge in the field:
I’m so tired of this sales pitch: “AI is using a lot of energy but it’s also helping the climate”. The AI that helps the climate is generally very different from the AI that uses a lot of energy (e.g. LLMs like ChatGPT). We can have the former without the latter.
David is one of the authors of the widely cited paper Aligning artificial intelligence with climate change mitigation, which I still find one of the most useful papers to read to understand the field.
A similar argument, if you prefer video
There’s also a similar argument made Karen Hao, author of the Book Empire of AI. There’s a 45 second long video of her doing so, again on Linkedin (as before, you might need to sign up/ log in to see it).
There she’s talking about the difference between some kinds of AI which are sort of boring and proven (a bit like bicycles), and new showy kinds of AI, like Generative AI which are massively resource intensive (like a rocket ship), and how we lump them together like they’re the same thing.
Using the blanket term “AI” to talk about both kinds of AI is like using the term “transport” to talk about rockets and bicycles together – it’s a very vague and lossy term, and mainly serves people selling the less proven, more resource intensive variant that is widely cited as the cause of so much infrastructure build out.
Anyway, the clip on linkedIn is from the Crafted Podcast by Dan Blumberg. It might be on the corresponding youtube channel for Crafted (which means you may be able to download the video in future once it’s up).
Update: I found this chart via a talk given by Vlad Coraoma recently at the Berlin Greentech Festival, which referenced a report by Schneider Electric published in December 2024, called Artificial Intelligence and Electricity – A System Dynamics Approach. There are other charts in the report, and other scenarios which largely draw similar conclusions in terms of where the demand is coming from, but this one is such a stark chart, that it feels almost like a cartoon:
In this chart, the yellow and orange lines represent Gen AI training and Gen AI inferencing respectively – they’re pretty much all the demand growth. The traditional AI training and inferencing (the blue and dark red lines), are basically flat.
If this is the case, and if it’s the trad AI that’s doing so much of the helpful climate work, but not driving the demand, then it makes the argument about the massive Gen AI build out being worth it for touted climate benefits even more flimsy.
I’ve said it before, but this distinction seems so cartoonish, I immediately find myself asking – are there any really concrete examples of GenAI delivering climate benefits at meaningful scale out there right now?
GenAI is new, but not that new – and I’ve been looking without much success so far, so I’d welcome pointers, because I can find plenty of examples of it being handy in the hands of very experienced software developers like Simon Willison, who essentially uses it like an army of interns to make him almost superhumanly productive, (see this post by him as well), but fewer cases beyond that.