I sat down with Henry Shevlin and Andy Masley to discuss AI’s environmental impact and why Andy thinks the panic is largely misplaced. Andy’s core argument: a single ChatGPT prompt uses a tiny fraction of your daily emissions, so even heavy usage barely moves the needle.
The real issue, he argues, isn’t that data centers are wasteful—they’re actually highly efficient—but that they make visible what’s normally invisible by aggregating hundreds of thousands of individually tiny tasks in one location. And he argues that the water concerns are even more overblown, with data centers using a small fraction compared to many other industries. We also explored why “every little bit counts” is harmful climate advice that distracts from interventions differing by orders of magnitude in impact.
In the second half of the conversation, we moved on to other interesting issues concerning the philosophy and politics of AI. For example, we discussed the “stochastic parrot” critique of chatbots and why there’s a huge middle ground between “useless autocomplete” and “human-level intelligence.” We also discussed Marx, “technological determinism”, and how AI can benefit authoritarian regimes.
Finally, we touched on effective altruism, the problem of “arguments as soldiers” in AI discourse, and why even high-brow information environments contain significant misinformation.
I enjoyed this conversation and feel like I learned a lot. Let me know in the comments if you think we got anything wrong!
Links
* Andy’s Weird Turn Pro Substack
* Using ChatGPT Is Not Bad for the Environment - A Cheat Sheet
* All the Ways I Want the AI Debate to Be Better
* “Sustainable Energy Without the Hot Air” by David MacKay:
* George Orwell - “You and the Atomic Bomb”:
Transcript
(Note: this transcript is AI-generated and so might be mistaken in parts).
Dan Williams: I’m here with my good friend Henry Shevlin and today we’re joined by our first ever guest, the great Andy Masley. Andy is one of my all-time favorite bloggers. He writes at the Weird Turn Pro Substack and he is the director of Effective Altruism DC and he’s published a ton of incredibly interesting articles about the philosophy of AI, the politics of AI, why so much AI discourse is so bad. And he’s also written about the main thing that we’re going to start talking about today to kick off the conversation, which is AI and the environment.
So I think many people have come across some version of the following view that says there’s a climate crisis. We have to drastically and rapidly reduce greenhouse gas emissions at the same time as AI companies are using a vast and growing amount of energy. So if we care about the environment, we should feel guilty about using systems like ChatGPT. And maybe if we’re very environmentally conscious, we should boycott these technologies altogether. So Andy, what’s your kind of high level take on that perspective?
Andy Masley: A lot to say. Just going down the list here. So basically for your personal environmental footprint, using chatbots is basically never going to make a dent. I think a lot of people have a lot of really wildly off takes about how big or small a part of their environmental footprint chatbots are.
There are a few specific issues that the media has definitely hyped up a lot, especially around water, which I talk about a lot. So living around data centers, I think is not as bad as the media is currently portraying. But in the long run, I’m kind of unsure. There are a lot of wild different directions AI could go. So I don’t want to speak too confidently about that. And I also just want to flag that I’m kind of a hobbyist on this. I feel like I know a lot of basic stuff, but I don’t have any kind of strong expertise in this stuff. So I’m very open to being wrong about a lot of the specific takes.
Dan Williams: But I think one of the things that—sorry Andy, just to cut you off—but I think one of the things that you point out in your very, very long, very, very detailed blog posts is that you don’t claim to be an expert, but you just cite the expert consensus on all of the very specific things that you’re talking about.
Andy Masley: Yeah, I do want to be clear that every factual statement I make, I think I can back up. How to interpret the facts is on me. I’m using some basic arguments. I was a philosophy major in undergrad, so I like to think I can deploy a few at least convincing or thought out arguments about this stuff. I’ve also been thinking about climate change stuff since I was a teenager, basically. So I have a lot of pent up thoughts about just basic climate ethics and have been pretty interested in this for a while.
Yeah, not claiming to know more than experts on this. What I am claiming is that I think if you interpret the facts and just look at how the facts are presented in a lot of media, I think a lot of journalism on this is getting really basic interpretations kind of wrong. I remember my first article that I read about this years ago—I think this was 2023—when an article came out that framed ChatGPT as a whole as using a lot of energy because it was using at the time like two times as much energy as a whole person. And at the time I was like, man, that’s not very much. A lot of people are using this app. And if you just add up the number of people using it, it shouldn’t surprise us that this app is using two times as much as an individual person’s lifestyle.
So there are a lot of small things like that over time that seem to have built up. It seems like there’s kind of a media consensus on like, this thing is pretty bad for the environment in general, so we should all report this way. And so any facts that are presented are kind of framed as “this is so much energy” or “this is so much water.” And if you just step back and contextualize it, it’s usually pretty shockingly small actually.
So I have a ton of other things to say about this. I think a part of the reason this is happening is that a lot of people just see AI as being very weird and new. And I agree there are valid reasons to freak out about AI. I want to flag that I’m not saying don’t freak out about AI, but I think the general energy and water use has been really overblown. We just need to compare the numbers to other things that we do.
The Problem with AI Environmental Reporting
Henry Shevlin: Yeah, this seems to me such a problem with the debate. And Andy, you say you don’t claim to be an expert in this, but I regularly interact with academics working in AI ethics and policy who make grand claims about the environment, but just don’t seem to have a good grasp on the actual figures. And I often ask people—they’ll say ChatGPT uses X amount of water or X amount of electricity, and this was before I started reading your blog post and I knew these figures themselves. I’d ask basic questions like, okay, what is that as a percentage of overall electricity use? Or is that just for the training runs or is that inference costs? And half the time they looked at me like I was a Martian. Or I was really saying, “Hang on, don’t you—you’re not supposed to ask questions like that. I’ve just told you, isn’t 80 million liters or whatever, isn’t that a really big number? Isn’t that enough? Why do you need to know what percentage that is of...?”
So I think honestly, you’re one of the very few people in this space who’s actually sitting down and doing the patient, boring work of quantifying these things and putting them in contrast with other forms of other ways in which humans use electricity and water.
Andy Masley: Yeah, and I will flag for anybody else who wants to look into this—I’m very proud of the work I’ve done, but I have to say it’s actually not boring for me anyway. It’s quite exciting to dig into and be like, “Wow, this like, you know, almost everything we do uses really crazy amounts of water” or “Here’s how energy is distributed around the world.” And so I think one hobby horse I’d like to push a litt…