This is an example of something I've seen elsewhere: a focus on a single item of human activity in terms of greenhouse gas emissions, rather than looking at the full sweep. Some will zero in on flying, for example, or driving.
I fear we'll see more of this as climate action becomes more politically difficult.
By our estimates, bitcoin in the USA was around ~80 TWH in 2024, which is in the ballpark of AI specialized servers in the US for the same year
Some context: we think the USA accounts for 75% of AI computation, while we think about 35% of bitcoin mining happens in the US. We also used different methodologies for mainline datcenters vs crypto, with the crypto model having worse data quality and higher uncertainty.
I'm a part of the LBNL team that wrote the 2024 energy use report. I want to thank you for the rigor you are bringing to this discourse. Its a very technical topic that is also pretty tedious and thus theres lots of misunderstanding. Water use isn't my topic area, and I didn't work on that part of the report, but I still appreciate the yeomans work.
One point I would emphasize is that a common way to optimize on-site water use is to use less energy efficient air-cooled chillers. Depending on the water intensity of the power, this will often consume more total water, just not within the fenceline.
It's challenging to communicate that naive water optimization at the facility level (in practice its usually a reason not to retrofit the chiller because "look at how water efficient we are") can lead to more water consuming operations at a system level, to say nothing of energy or carbon. Thanks again for improving the discourse on these issues
What is your opinion on the oxford study that was going around a couple years ago which concluded that a plant-based diet is the single most effective way to reduce your personal emissions? A 75% reduction in emissions from just that change. I've seen the claim worded differently in different articles. But it seems to contradict this graph.
I've been vegan for 10 years so I have a bias in favor, but I honestly don't know enough about how different lifestyles compare. I'd encourage anyone considering making such a big change to see if there are any opportunities to push for changes to their local grid, those usually have way bigger impacts.
Oh yeah for sure. I think I was specifically asking about that study, and how it relates to that graph you posted. But I just realized I'm asking you to do research that I should probably do myself!
Something I'd love to see more about is the relative efficiency of generative AI tool use vs the computer tasks that the tool use replaces. In law, for instance, people sometimes use generative AI to conduct legal research (which they'd otherwise do via Westlaw or Lexis) or to draft briefs (probably in Microsoft Word). It seems plausible that doing legal research via ChatGPT might save energy on net, if it replaces multiple queries/pageloads in Westlaw. But that's just speculation.
He’s not saying they’re literally water powered, he’s saying they use water in the course of generating electricity, like for steam loops, or for cooling, etc.
I think this argument raises some good points but was intent from the start on selling a point. That AI is not harmful but rather beneficial to all around it. Would be interesting to see all the sponsors of the show and then go further.
If you want to sound objective- don’t gaslight. Be objective. You may still have some good points but don’t say you’re objective and then be anything but. Next….
Most AI environmental concerns seem to come from either a place of motivated reasoning, virtue signaling, or climate anxiety.
Motivated reasoners feel uneasy about AI because they fear it might make them obsolete, or lower their pay, or make their skills useless or mundane, or any other reason. This seems to be caused by an odd catch-22 where most AI anxiety is caused by AI's rapidly increasing capabilities. Everyone's afraid of what AI *might* do. A common (and widely accepted) method of self soothing is denying that AI capabilities are meaningfully increasing, but this makes the very rational fear of mass disempowerment or obsolescence suddenly irrational. If you don't think AI is going to get much better, why are you afraid?
The combination of a strong belief that "AI could never do my job" alongside a strong fear of "what if AI could do my job" creates a floating cloud of logically unjustifiable anxiety that latches onto whatever pseudo-logical reasons exist for hating AI. It hurts the environment, it violates copyright, whatever. They don't hate AI for environmental reasons, they're afraid (which, again, we should be!), but they're also not willing to admit they're afraid.
Virtue signalers just follow highly visible trends. They don't care about the environment, they care about other people thinking they care about the environment. They're the kind of person to buy a new phone every year and refuse to use plastic straws. They criticize AI use because it signals environmental awareness. If starting a tire fire signaled environmental awareness they'd do that too.
People with climate anxiety care about the environment, but they do so from a place of neurosis. The virtue signaler needs to convince everyone else they care, but the worrier must convince THEMSELVES that they care about the environment. They're not doing it for social clout, they're genuinely deeply worried. The way they sooth that anxiety is through penance, uncomfortable actions that they view as "helping the environment".
Because the outcomes they're seeking are ultimately emotional (rather than trying to maximize some quantity of CO2 saved in a spreadsheet or something) they fall prey to emotional traps. We as humans are terrible at reasoning with very big and very small numbers, and we rely heavily on the social signals other people send us. When people say ChatGPT is a problem, you feel anxious about your ChatGPT use, so you reduce it. People with climate anxiety tend to act similarly to virtue signalers, just more honestly, more consistently, and more privately.
You wrote: "I can’t find any other places where people have gotten worried about things they do that use such tiny amounts of energy," and "We should mostly focus on systematic change over personal lifestyle changes, but if we do want to do personal lifestyle changes, we should prioritize cutting things that are actually significant parts of our personal emissions." I agree-- I get very frustrated when the given solution to environmental problems is to turn off the water while brushing your teeth, turn off the lights when you're not in the room, don't litter, all while the companies dumping loads of chemicals into the river are relatively unexamined.
I agree with your statements, but also, the actual information, the statistics on actual impact, have been very difficult (for me) to find in order to speak to any point (land use, raw materials, water use, energy use...) in an informed way. There are definitely articles that speak to the harm. And the lack of transparency is true of other technologies as well, like electric scooters, Zoom, polyester... But without knowing the actual impact, it's hard to know if it's "such tiny amounts of energy" and difficult to "prioritize cutting things that are actually significant parts of our personal emissions."
The time displacement is a very important factor, that the time used with AI displaces some other use. I'm appreciating your research and articles on this new-ish and somewhat opaque topic.
This is an example of something I've seen elsewhere: a focus on a single item of human activity in terms of greenhouse gas emissions, rather than looking at the full sweep. Some will zero in on flying, for example, or driving.
I fear we'll see more of this as climate action becomes more politically difficult.
Despite the campaigning of the moral panic industry complex, these mass communications do provide benefit to the average outsider, Thanks.
I wonder how it compares to bitcoin? I know some artists that were boycotting that (like for nfts)
By our estimates, bitcoin in the USA was around ~80 TWH in 2024, which is in the ballpark of AI specialized servers in the US for the same year
Some context: we think the USA accounts for 75% of AI computation, while we think about 35% of bitcoin mining happens in the US. We also used different methodologies for mainline datcenters vs crypto, with the crypto model having worse data quality and higher uncertainty.
See the report (https://escholarship.org/uc/item/32d6m0d1) for more detail about the underlying methods
I'm a part of the LBNL team that wrote the 2024 energy use report. I want to thank you for the rigor you are bringing to this discourse. Its a very technical topic that is also pretty tedious and thus theres lots of misunderstanding. Water use isn't my topic area, and I didn't work on that part of the report, but I still appreciate the yeomans work.
One point I would emphasize is that a common way to optimize on-site water use is to use less energy efficient air-cooled chillers. Depending on the water intensity of the power, this will often consume more total water, just not within the fenceline.
It's challenging to communicate that naive water optimization at the facility level (in practice its usually a reason not to retrofit the chiller because "look at how water efficient we are") can lead to more water consuming operations at a system level, to say nothing of energy or carbon. Thanks again for improving the discourse on these issues
Just wanna add that I’ve become a massive fanboy of the LBNL in my time looking into this so thank you for your work!
I hope you’re right. I hope that forwarding this post to a friend of mine will be useful in our discussion about AI.
What is your opinion on the oxford study that was going around a couple years ago which concluded that a plant-based diet is the single most effective way to reduce your personal emissions? A 75% reduction in emissions from just that change. I've seen the claim worded differently in different articles. But it seems to contradict this graph.
Here is an example. Love to know your thoughts.
https://www.globalcitizen.org/en/content/going-vegan-is-best-way-to-help-planet/
I've been vegan for 10 years so I have a bias in favor, but I honestly don't know enough about how different lifestyles compare. I'd encourage anyone considering making such a big change to see if there are any opportunities to push for changes to their local grid, those usually have way bigger impacts.
Oh yeah for sure. I think I was specifically asking about that study, and how it relates to that graph you posted. But I just realized I'm asking you to do research that I should probably do myself!
Appreciate your work.
Excellent article, Andy!
Something I'd love to see more about is the relative efficiency of generative AI tool use vs the computer tasks that the tool use replaces. In law, for instance, people sometimes use generative AI to conduct legal research (which they'd otherwise do via Westlaw or Lexis) or to draft briefs (probably in Microsoft Word). It seems plausible that doing legal research via ChatGPT might save energy on net, if it replaces multiple queries/pageloads in Westlaw. But that's just speculation.
it sucks when i try to explain that AI does not really consume as much as well, driving your car to the grocery store lol
"Most electricity is generated using water" I haven't seen evidence supporting this assumption. I think it'd good if you added the source.
If I take i.e. this (https://www.eia.gov/energyexplained/electricity/electricity-in-the-us.php#tab2), renewable sources account for around 25% of the energy, so hydropower should be even less than that.
He’s not saying they’re literally water powered, he’s saying they use water in the course of generating electricity, like for steam loops, or for cooling, etc.
I think this argument raises some good points but was intent from the start on selling a point. That AI is not harmful but rather beneficial to all around it. Would be interesting to see all the sponsors of the show and then go further.
If you want to sound objective- don’t gaslight. Be objective. You may still have some good points but don’t say you’re objective and then be anything but. Next….
I'm not at all saying that AI is beneficial for everyone?
It sure seems that way. The episode I just listened to was not unbiased. I'm a big AI fan but there are as many caution lights as green lights...
This is a problem with your reading comprehension not the article.
Good to see another one afflicted with entitlement…good try SVF.
Most AI environmental concerns seem to come from either a place of motivated reasoning, virtue signaling, or climate anxiety.
Motivated reasoners feel uneasy about AI because they fear it might make them obsolete, or lower their pay, or make their skills useless or mundane, or any other reason. This seems to be caused by an odd catch-22 where most AI anxiety is caused by AI's rapidly increasing capabilities. Everyone's afraid of what AI *might* do. A common (and widely accepted) method of self soothing is denying that AI capabilities are meaningfully increasing, but this makes the very rational fear of mass disempowerment or obsolescence suddenly irrational. If you don't think AI is going to get much better, why are you afraid?
The combination of a strong belief that "AI could never do my job" alongside a strong fear of "what if AI could do my job" creates a floating cloud of logically unjustifiable anxiety that latches onto whatever pseudo-logical reasons exist for hating AI. It hurts the environment, it violates copyright, whatever. They don't hate AI for environmental reasons, they're afraid (which, again, we should be!), but they're also not willing to admit they're afraid.
Virtue signalers just follow highly visible trends. They don't care about the environment, they care about other people thinking they care about the environment. They're the kind of person to buy a new phone every year and refuse to use plastic straws. They criticize AI use because it signals environmental awareness. If starting a tire fire signaled environmental awareness they'd do that too.
People with climate anxiety care about the environment, but they do so from a place of neurosis. The virtue signaler needs to convince everyone else they care, but the worrier must convince THEMSELVES that they care about the environment. They're not doing it for social clout, they're genuinely deeply worried. The way they sooth that anxiety is through penance, uncomfortable actions that they view as "helping the environment".
Because the outcomes they're seeking are ultimately emotional (rather than trying to maximize some quantity of CO2 saved in a spreadsheet or something) they fall prey to emotional traps. We as humans are terrible at reasoning with very big and very small numbers, and we rely heavily on the social signals other people send us. When people say ChatGPT is a problem, you feel anxious about your ChatGPT use, so you reduce it. People with climate anxiety tend to act similarly to virtue signalers, just more honestly, more consistently, and more privately.
You wrote: "I can’t find any other places where people have gotten worried about things they do that use such tiny amounts of energy," and "We should mostly focus on systematic change over personal lifestyle changes, but if we do want to do personal lifestyle changes, we should prioritize cutting things that are actually significant parts of our personal emissions." I agree-- I get very frustrated when the given solution to environmental problems is to turn off the water while brushing your teeth, turn off the lights when you're not in the room, don't litter, all while the companies dumping loads of chemicals into the river are relatively unexamined.
I agree with your statements, but also, the actual information, the statistics on actual impact, have been very difficult (for me) to find in order to speak to any point (land use, raw materials, water use, energy use...) in an informed way. There are definitely articles that speak to the harm. And the lack of transparency is true of other technologies as well, like electric scooters, Zoom, polyester... But without knowing the actual impact, it's hard to know if it's "such tiny amounts of energy" and difficult to "prioritize cutting things that are actually significant parts of our personal emissions."
The time displacement is a very important factor, that the time used with AI displaces some other use. I'm appreciating your research and articles on this new-ish and somewhat opaque topic.