For what it's worth, I think you've made a solid case and I'm pretty much convinced. I don't have any special knowledge about data centers, but I'm an AI safety researcher and former software engineer, with some background in climate research (I developed a widely used algorithm for measuring precipitation).
Prior to your series on this, I assumed that the claims about data centers having unusual bad effects were mostly correct, so I was initially skeptical of your counterclaims. I'm not sure I've read literally everything you've written on the issue, but I've certainly read most of them, and have dug into various of your sources and worked through some of the math, and your arguments have consistently held up modulo small quibbles here and there.
I think you've successfully made the case that this is essentially a moral panic, and the energy going into fighting data centers ought to be redirected to more important issues. I remain open to the possibility that data centers are actually causing disproportionate harm, but as you say here, at this point I would have to see some solid concrete evidence to be convinced of that.
I live not far from the Saline Twp Michigan datacenter and here are some factors I have seen:
- The datacenter did a great job of casting itself as a sinister villain. The company made a deal with the Governor, a D, and then strong-armed a volunteer township board. The board first pushed back and basically wanted more scrutiny including around the tax situation, and then were beset by some of the highest paid corporate attorneys in the world. Oh and these are mostly R people.
- The construction is running dozens of to-legal-limit double gravel haulers right through the center of town each day. The subcontractor truck drivers also tried out every rural and residential shortcut to speed their route, further alienating virtually everyone.
- The site of the project is right in the middle of multigenerational small and medium agriculture businesses that are already struggling- the names of the owners are also the names on all the roads, schools and plaques around here.
- This on top of the usual “this shopping center used to be a beautiful farm” and sprawling subdivisions.
They did everything but poison the firehouse Dalmatian.
Everyone already has great unease with the pace of change, growing uncertainty, a sense of being out of control, and then here comes Oracle, of all the companies, to do a Scooby-Doo-level assault on the community. A bunch of people, standing around with pitchforks and torches but nothing to march against, and from the sky drops a slicked up Silicon Valley caricature in an ogre costume.
And just last night I saw another chapter. I don’t know how to get this across without writing a lake woebegone type essay.
The datacenter company just pledged $10m to the local Rec Center, which is an indoor pool, gym etc. Wonderful, right? Not so fast.
First there’s the whole, how do I put this tactfully, sense that you’re buying me breakfast after the worst night of my life.
Then, locally, the rec center is actually a symbol of the city-versus-rural divide. It was built with a levy against the City and several townships based on the school district boundaries. The City (small town, but legally a city) people are enthusiastic, live near it, use it. The Township people would have to drive 20+ minutes to get there, rarely use it, think the whole thing is a boondoggle, and are politically against government services.
Did I mention a lot of the “city folk” work in the People’s Republic of Ann Arbor?
So not only is the datacenter company issuing an assault against the conservative multigenerational farming way of life, at a time when it is already under great pressure, but they are delivering the spoils to the symbol of liberal government spending culture.
“Oh so sorry I poisoned your dog, here let me donate to the local we-hate-dogs chapter.”
This reminds me of the Shale Oil boom that captured my hometown, and many others, in the 2010s. Overnight the town was swarmed by construction and oil workers, and the water trucks and haulers that passed through caused massive traffic jams, and building and road damage throughout the county. There was a mad rush on 'useless' land all around in the hope that you'd get royalties.
> - The construction is running dozens of to-legal-limit double gravel haulers right through the center of town each day. The subcontractor truck drivers also tried out every rural and residential shortcut to speed their route, further alienating virtually everyone.
Water trucks would pass through town by the hundreds daily and their weight ruined many roads and caused foundational damage to a number of buildings in the downtown corridor. The traffic my parents faced on their two mile commute to their store was incredible.
> - The site of the project is right in the middle of multigenerational small and medium agriculture businesses that are already struggling- the names of the owners are also the names on all the roads, schools and plaques around here.
Speaking of my parent's store, there used to be a small forest behind it, but that is all gone now, cleared by a gas company putting up facilities there. They were not alone in having their surroundings altered like that.
What is the result today? In my hometown there are a few hotels that were hastily built that now are largely unoccupied. Many of the gas wells are capped for future use. A few folks made big money, some made a trickle, and most made none. There's a new natural-gas power plant.
What makes this interesting is there's a big data center that is likely going up soon, and there is a lot of local pushback. There was plenty of pushback to the oil and gas workers too. Just interesting how this cycles with resource booms, and the new norm settles in over time.
Everything about this - including specifically using the phrase "sinister villain" - reminds me of how NIMBYs talk about people trying to build housing. I'm from the San Francisco bay area, I can't afford a house in the city I grew up in, and I am not inclined to look favorably on NIMBYs.
Agree for sure. I spent some time in the Bay Area in the early 00s and it was unaffordable -then-, I can’t imagine now. At the time everyone was all “build density near transit (not here on my hill)” until they wanted to put some apartments near Millbrae BART and then it was all, think of the children!
And it is NIMBY, data centers or other infra like power plants, or even malls and subdivisions, need to go somewhere. Maybe don’t put the data centers next to the elementary school, but this thing is five miles out of town on a state highway.
It’s more the folly and hubris of the planners, to not understand what they were getting into. Especially since they are at least culturally from the Bay Area and have seen the premiere leagues of nimbyism. Maybe they just thought they were dropping into a pile of midwestern rubes and didn’t stop to think that modern farmers are actually very savvy and well connected (otherwise they would have gone out of business two generations ago), and uniting against Oracle and Big Bad AI was the one thing that would pull the university town and its rural surroundings together.
Well, its also that AI leaders have sold AI to investors on the basis that it can eliminate a massive swath of the workforce. There's a bunch of people for whom data centers are only negatives.
I really don't think so; AI can help people with lots of every day things like negotiating for a reduction of your phone bill or figuring out what to do with the kids on weekends.
If AI makes everyone 3x more productive and the amount of productivity stays the same it's not really related to AI whether, as a society, we choose to reduce the amount everyone has to work by 3x or make it so that only 1/3 of able bodied people have jobs. but it's true that the average person doesn't really have power over that.
I’m persuaded by your arguments but any time I share it with an environmental critic of AI, they dismiss the arguments because I’m citing “some guy with a Substack”. I feel like there’s a need for some credibility laundering by having articles by you or about you and/or your arguments in major media outlets like NYT. It was great to hear you on Hard Fork a while back by the way! I don’t know how much influence you have on that. If I could tell someone in conversation “I heard on NPR that the environmental concerns about data centers are overblown” that would make a difference.
I also think it's more of a vibes thing than a I'm-against-the-datacenter-buildout-for-these-rational-reasons thing. I'm sympathetic to the vibes camp though, and in the spirit of steelmanning them I think it results from a lossy compression of a lot of real (and valid) anti-AI/tech sentiment. This is my best understanding of their worldview:
(note that I personally am very excited about what I can do with AI, have sent your water debunking article to ~10 people, and don't really 'agree' with this view).
---
Someone may have heard about 'AI!' a lot but doesn't use it, or tried ChatGPT once in instant mode and it had an ad and got 3 things wrong, but it was able to re-create a picture in the style of their artist friend who did mildly successful commissions for e.g. businesses logos and now can't find any work. Also their kid is using it to cheat through high school so he can play video games. At the same time this datacenter is proposed and they've seen pictures that are unbelievably ugly and their cousin tells them he lives near one and it's loud and awful and then they go to the town hall meeting and the guy proposing it is an asshole and reminds you of that dick from shark tank and it's just bad vibes all around and you don't want it and you don't see any use for it and apparently it uses all the water too and idk that becomes the 'reason,' but you just don't want it. Big tech is trying to shove it down on you and they send some asshat to force/bribe the town committee into it because they don't care enough to show up themselves.
There's also absolute distrust of everyone involved. Someone shares with them that story about the couple who had their well ruined by a datacenter and it sounds horrible and the couple seems lovely. I don't think their mind is going to change because someone else shared a 10k+ word article from a guy on substack saying 'well actually that was because the construction went wrong not because the AI uses water' (think "bro idgaf the datacenter ruined them. They're just going to lie and convince me it's okay to build here then something bad will happen and they'll say 'well actually it didn't break *this* thing it broke *that* thing and we never said it wouldn't break *that* thing so really this one's on you.'")
---
I think the 'solution,' if one exists, is for the senior managers of datacenter projects to spend a lot of time in the prospective host town and just get to know the local community. They should come armed with your research, but also be friendly and decent and align themselves with locals in good faith.
The correction then becomes: "this center is legitimately going to use less water than the local golf course; that bad event you heard about happened because those people were careless with the construction. It won't happen here because you know me and I promise we aren't cutting corners, and I'm spending most of my time here and need good water for myself." THEN, when you tell them 'also there are more jobs and the tax revenue is going to come right back to all of you' it stops being a bribe and starts winning hearts and minds.
Good arguments can correct bad arguments for people who argue, but in my experience only good vibes can counter bad vibes.
This is correct. I think vibes concerns about ai are completely reasonable even when i think a lot of the theoretically fact based complaints are pretty shaky.
I entirely understand that you have to be almost-preternaturally charitable to the media, anti-AI campaigners, the (mostly AI-hating) public, etc.: when you're in a minority of essentially one, in the public discourse, if you're not sufficiently scrupulous in treating all claims as faithfully as possible - far more faithfully than your opponents treat your own claims - then you would immediately be accused of unrigourousness, bias, ignorance, dismissiveness, deception, etc.
(Meanwhile, of course, because nearly everybody is on their side, your opponents can freely operate with considerably more relaxed standards of rigour - to put it mildly! - without much fear of criticism..)
Despite this, it seems only fair and reasonable to point out that the media, the public, etc. aren't starting from a question of whether AI is environmentally-harmful or not and then trying to decide whether it is good or bad, but rather starting from a position of "AI is bad" and then working-backwards to environmental harms. It's abundantly clear that explains essentially every subsequently-falsified finding of env. harms, every misleading claim, every publicly-held belief of env. harms, etc.
Compare the public, media, etc. position on (to take one example!) the meat and dairy industries. These industries produce catastrophic environmental harms, both in an absolute sense and relative to any benefits we might get from meat, milk, etc. But the public and media consider claims of env. harms fromthese industries to be essentially ludicrous scaremongering from a few crackpot hippies, not a comprehensively-studied, rigourously-documented issue. Because, of course, the media, public, etc. are starting from a position of "meat and dairy are great" and working-backwards.
*Even assuming* vast AI env. harms, an internally-consistent public and media would need to also be outraged at the scale of the meat and dairy industries, the establishment of cattle farms, the widespread sale of meat and dairy in shops, etc. The fact that they are not only *not* outraged but are also overwhelmingly scornful of the idea of meat and dairy env. harms (or the env. harms of any other similar industry) essentially betrays the self-inconsistency and prejudice wherefrom the claims of AI env. harms seem to originate.
That being said - what I would really like to see from you, but which I haven't seen (apologies if I've simply missed it!) is an analysis of the env. harms that may arise from *centralisation*.
A single coal-fired power plant is considerably less harmful than a million individual coal fires in a million living-rooms, both because of efficiencies-of-scale and because the pollution is moved into a big building far away and out of *literally the same room in which your family is sitting*.
However - even though overall the coal plant is vastly environmentally beneficial compared to individual domestic coal fires, if you happened to live right next door to it, your standard of air quality, risk of lung disease, etc. would be considerably worsened. It's not the overall harm that's the issue, it's the localised harm from concentrating a million domestic-coal-fire-equivalents in one place.
As you point out, this doesn't seem to be the case for most AI harms, presumably owing to some combination of datacentres' distance from populations, low overall harms, and env receptors' having sufficient capacity for absorbing the (generally-minor) additional impact of the datcentre.
But - there are undoubtedly cases where the distance to a population doesn't seem too relevant (eg. the water table works on a very large scale, as evidenced by eg. irrigation systems in one region causing droughts many hundreds of miles away); the low overall harms don't seem too relevant (just as with the coal fires, an incredibly low env. cost per AI query wouldn't necessarily help us locally if we're concentrating trillions of AI queries in one place); and the env. receptors don't seem to have sufficient capacity to absorb additional impacts (eg. much of the UK's water supply has very little headroom such that in dry weather the additional usage from people using hosepipes when gardening or washing their car is by itself enough to push us over 100% capacity and cause droughts).
Perhaps these sorts of "centralisation problems" don't occur much with AI infrastructure, I don't know - but I haven't seen much rigourous investigation of centralisation-related env. impacts (as distinct from investigation of overall env. impacts), and I don't think there's anybody really capable of and interested in investigating them rigourously other than you!
> This is a weird position for me to be in! I’ve always been a good predictable lib and don’t find myself on the opposite side of almost all media coverage often. It’s driven me a little crazy and does make me worry I’m missing something everyone else seems to see. But after a year of following this with no really satisfying answers, it’s become hard not to start to think I’m the one with my feet on the ground here.
You should probably sit down and think to yourself about what other issues "almost all" media coverage that a "good predictable lib" might be exposed to takes a position on, where that position is actually badly wrong; and isn't getting systematically corrected because people keep insisting that the people who bring up arguments against it are shills for some power or conspiracy theorists. It's certainly not just the issue of AI data centers.
Thanks, Andy. I think it is just vibes, as Ruth below says. It is just another piece of misguided virtue signaling, like being anti-nuclear power, being entirely incapable of rational thought about climate change, etc. I greatly appreciate your work in this area.
Even if AI significantly increases electricity demand, it does not automatically follow that this is environmentally negative although the concern is very legitimate. Historically, large new sources of demand have often triggered investment in new supply. The question is not whether AI will require more energy. It almost certainly will. The question is whether that demand accelerates deployment of cleaner energy sources or locks in more fossil fuel generation.
An excellent read. As a journalist, I have to agree that a lot of the writing leads to dead ends. But perhaps it’s the old newspaper adage: “Never let the facts stand in the way of a good story”
This is the kind of environmental argument I wish we saw more often: specific, numerate, and willing to separate actual harm from vibes in a panic hat.
Fear travels faster than numbers, and numbers require everyone to calm down long enough to read them.
For what it's worth, I think you've made a solid case and I'm pretty much convinced. I don't have any special knowledge about data centers, but I'm an AI safety researcher and former software engineer, with some background in climate research (I developed a widely used algorithm for measuring precipitation).
Prior to your series on this, I assumed that the claims about data centers having unusual bad effects were mostly correct, so I was initially skeptical of your counterclaims. I'm not sure I've read literally everything you've written on the issue, but I've certainly read most of them, and have dug into various of your sources and worked through some of the math, and your arguments have consistently held up modulo small quibbles here and there.
I think you've successfully made the case that this is essentially a moral panic, and the energy going into fighting data centers ought to be redirected to more important issues. I remain open to the possibility that data centers are actually causing disproportionate harm, but as you say here, at this point I would have to see some solid concrete evidence to be convinced of that.
I live not far from the Saline Twp Michigan datacenter and here are some factors I have seen:
- The datacenter did a great job of casting itself as a sinister villain. The company made a deal with the Governor, a D, and then strong-armed a volunteer township board. The board first pushed back and basically wanted more scrutiny including around the tax situation, and then were beset by some of the highest paid corporate attorneys in the world. Oh and these are mostly R people.
- The construction is running dozens of to-legal-limit double gravel haulers right through the center of town each day. The subcontractor truck drivers also tried out every rural and residential shortcut to speed their route, further alienating virtually everyone.
- The site of the project is right in the middle of multigenerational small and medium agriculture businesses that are already struggling- the names of the owners are also the names on all the roads, schools and plaques around here.
- This on top of the usual “this shopping center used to be a beautiful farm” and sprawling subdivisions.
They did everything but poison the firehouse Dalmatian.
Everyone already has great unease with the pace of change, growing uncertainty, a sense of being out of control, and then here comes Oracle, of all the companies, to do a Scooby-Doo-level assault on the community. A bunch of people, standing around with pitchforks and torches but nothing to march against, and from the sky drops a slicked up Silicon Valley caricature in an ogre costume.
And just last night I saw another chapter. I don’t know how to get this across without writing a lake woebegone type essay.
The datacenter company just pledged $10m to the local Rec Center, which is an indoor pool, gym etc. Wonderful, right? Not so fast.
First there’s the whole, how do I put this tactfully, sense that you’re buying me breakfast after the worst night of my life.
Then, locally, the rec center is actually a symbol of the city-versus-rural divide. It was built with a levy against the City and several townships based on the school district boundaries. The City (small town, but legally a city) people are enthusiastic, live near it, use it. The Township people would have to drive 20+ minutes to get there, rarely use it, think the whole thing is a boondoggle, and are politically against government services.
Did I mention a lot of the “city folk” work in the People’s Republic of Ann Arbor?
So not only is the datacenter company issuing an assault against the conservative multigenerational farming way of life, at a time when it is already under great pressure, but they are delivering the spoils to the symbol of liberal government spending culture.
“Oh so sorry I poisoned your dog, here let me donate to the local we-hate-dogs chapter.”
This reminds me of the Shale Oil boom that captured my hometown, and many others, in the 2010s. Overnight the town was swarmed by construction and oil workers, and the water trucks and haulers that passed through caused massive traffic jams, and building and road damage throughout the county. There was a mad rush on 'useless' land all around in the hope that you'd get royalties.
> - The construction is running dozens of to-legal-limit double gravel haulers right through the center of town each day. The subcontractor truck drivers also tried out every rural and residential shortcut to speed their route, further alienating virtually everyone.
Water trucks would pass through town by the hundreds daily and their weight ruined many roads and caused foundational damage to a number of buildings in the downtown corridor. The traffic my parents faced on their two mile commute to their store was incredible.
> - The site of the project is right in the middle of multigenerational small and medium agriculture businesses that are already struggling- the names of the owners are also the names on all the roads, schools and plaques around here.
Speaking of my parent's store, there used to be a small forest behind it, but that is all gone now, cleared by a gas company putting up facilities there. They were not alone in having their surroundings altered like that.
What is the result today? In my hometown there are a few hotels that were hastily built that now are largely unoccupied. Many of the gas wells are capped for future use. A few folks made big money, some made a trickle, and most made none. There's a new natural-gas power plant.
What makes this interesting is there's a big data center that is likely going up soon, and there is a lot of local pushback. There was plenty of pushback to the oil and gas workers too. Just interesting how this cycles with resource booms, and the new norm settles in over time.
Everything about this - including specifically using the phrase "sinister villain" - reminds me of how NIMBYs talk about people trying to build housing. I'm from the San Francisco bay area, I can't afford a house in the city I grew up in, and I am not inclined to look favorably on NIMBYs.
Agree for sure. I spent some time in the Bay Area in the early 00s and it was unaffordable -then-, I can’t imagine now. At the time everyone was all “build density near transit (not here on my hill)” until they wanted to put some apartments near Millbrae BART and then it was all, think of the children!
And it is NIMBY, data centers or other infra like power plants, or even malls and subdivisions, need to go somewhere. Maybe don’t put the data centers next to the elementary school, but this thing is five miles out of town on a state highway.
It’s more the folly and hubris of the planners, to not understand what they were getting into. Especially since they are at least culturally from the Bay Area and have seen the premiere leagues of nimbyism. Maybe they just thought they were dropping into a pile of midwestern rubes and didn’t stop to think that modern farmers are actually very savvy and well connected (otherwise they would have gone out of business two generations ago), and uniting against Oracle and Big Bad AI was the one thing that would pull the university town and its rural surroundings together.
its the vibes... AI has bad vibes for people that don't use it (not counting google search). I think that's still probably most people.
Well, its also that AI leaders have sold AI to investors on the basis that it can eliminate a massive swath of the workforce. There's a bunch of people for whom data centers are only negatives.
I really don't think so; AI can help people with lots of every day things like negotiating for a reduction of your phone bill or figuring out what to do with the kids on weekends.
If AI makes everyone 3x more productive and the amount of productivity stays the same it's not really related to AI whether, as a society, we choose to reduce the amount everyone has to work by 3x or make it so that only 1/3 of able bodied people have jobs. but it's true that the average person doesn't really have power over that.
I’m persuaded by your arguments but any time I share it with an environmental critic of AI, they dismiss the arguments because I’m citing “some guy with a Substack”. I feel like there’s a need for some credibility laundering by having articles by you or about you and/or your arguments in major media outlets like NYT. It was great to hear you on Hard Fork a while back by the way! I don’t know how much influence you have on that. If I could tell someone in conversation “I heard on NPR that the environmental concerns about data centers are overblown” that would make a difference.
Actually have some stuff coming up that should help a lot here
I also think it's more of a vibes thing than a I'm-against-the-datacenter-buildout-for-these-rational-reasons thing. I'm sympathetic to the vibes camp though, and in the spirit of steelmanning them I think it results from a lossy compression of a lot of real (and valid) anti-AI/tech sentiment. This is my best understanding of their worldview:
(note that I personally am very excited about what I can do with AI, have sent your water debunking article to ~10 people, and don't really 'agree' with this view).
---
Someone may have heard about 'AI!' a lot but doesn't use it, or tried ChatGPT once in instant mode and it had an ad and got 3 things wrong, but it was able to re-create a picture in the style of their artist friend who did mildly successful commissions for e.g. businesses logos and now can't find any work. Also their kid is using it to cheat through high school so he can play video games. At the same time this datacenter is proposed and they've seen pictures that are unbelievably ugly and their cousin tells them he lives near one and it's loud and awful and then they go to the town hall meeting and the guy proposing it is an asshole and reminds you of that dick from shark tank and it's just bad vibes all around and you don't want it and you don't see any use for it and apparently it uses all the water too and idk that becomes the 'reason,' but you just don't want it. Big tech is trying to shove it down on you and they send some asshat to force/bribe the town committee into it because they don't care enough to show up themselves.
There's also absolute distrust of everyone involved. Someone shares with them that story about the couple who had their well ruined by a datacenter and it sounds horrible and the couple seems lovely. I don't think their mind is going to change because someone else shared a 10k+ word article from a guy on substack saying 'well actually that was because the construction went wrong not because the AI uses water' (think "bro idgaf the datacenter ruined them. They're just going to lie and convince me it's okay to build here then something bad will happen and they'll say 'well actually it didn't break *this* thing it broke *that* thing and we never said it wouldn't break *that* thing so really this one's on you.'")
---
I think the 'solution,' if one exists, is for the senior managers of datacenter projects to spend a lot of time in the prospective host town and just get to know the local community. They should come armed with your research, but also be friendly and decent and align themselves with locals in good faith.
The correction then becomes: "this center is legitimately going to use less water than the local golf course; that bad event you heard about happened because those people were careless with the construction. It won't happen here because you know me and I promise we aren't cutting corners, and I'm spending most of my time here and need good water for myself." THEN, when you tell them 'also there are more jobs and the tax revenue is going to come right back to all of you' it stops being a bribe and starts winning hearts and minds.
Good arguments can correct bad arguments for people who argue, but in my experience only good vibes can counter bad vibes.
This is correct. I think vibes concerns about ai are completely reasonable even when i think a lot of the theoretically fact based complaints are pretty shaky.
well steel-manned!
Your work is very important and very appreciated. Please continue
I entirely understand that you have to be almost-preternaturally charitable to the media, anti-AI campaigners, the (mostly AI-hating) public, etc.: when you're in a minority of essentially one, in the public discourse, if you're not sufficiently scrupulous in treating all claims as faithfully as possible - far more faithfully than your opponents treat your own claims - then you would immediately be accused of unrigourousness, bias, ignorance, dismissiveness, deception, etc.
(Meanwhile, of course, because nearly everybody is on their side, your opponents can freely operate with considerably more relaxed standards of rigour - to put it mildly! - without much fear of criticism..)
Despite this, it seems only fair and reasonable to point out that the media, the public, etc. aren't starting from a question of whether AI is environmentally-harmful or not and then trying to decide whether it is good or bad, but rather starting from a position of "AI is bad" and then working-backwards to environmental harms. It's abundantly clear that explains essentially every subsequently-falsified finding of env. harms, every misleading claim, every publicly-held belief of env. harms, etc.
Compare the public, media, etc. position on (to take one example!) the meat and dairy industries. These industries produce catastrophic environmental harms, both in an absolute sense and relative to any benefits we might get from meat, milk, etc. But the public and media consider claims of env. harms fromthese industries to be essentially ludicrous scaremongering from a few crackpot hippies, not a comprehensively-studied, rigourously-documented issue. Because, of course, the media, public, etc. are starting from a position of "meat and dairy are great" and working-backwards.
*Even assuming* vast AI env. harms, an internally-consistent public and media would need to also be outraged at the scale of the meat and dairy industries, the establishment of cattle farms, the widespread sale of meat and dairy in shops, etc. The fact that they are not only *not* outraged but are also overwhelmingly scornful of the idea of meat and dairy env. harms (or the env. harms of any other similar industry) essentially betrays the self-inconsistency and prejudice wherefrom the claims of AI env. harms seem to originate.
That being said - what I would really like to see from you, but which I haven't seen (apologies if I've simply missed it!) is an analysis of the env. harms that may arise from *centralisation*.
A single coal-fired power plant is considerably less harmful than a million individual coal fires in a million living-rooms, both because of efficiencies-of-scale and because the pollution is moved into a big building far away and out of *literally the same room in which your family is sitting*.
However - even though overall the coal plant is vastly environmentally beneficial compared to individual domestic coal fires, if you happened to live right next door to it, your standard of air quality, risk of lung disease, etc. would be considerably worsened. It's not the overall harm that's the issue, it's the localised harm from concentrating a million domestic-coal-fire-equivalents in one place.
As you point out, this doesn't seem to be the case for most AI harms, presumably owing to some combination of datacentres' distance from populations, low overall harms, and env receptors' having sufficient capacity for absorbing the (generally-minor) additional impact of the datcentre.
But - there are undoubtedly cases where the distance to a population doesn't seem too relevant (eg. the water table works on a very large scale, as evidenced by eg. irrigation systems in one region causing droughts many hundreds of miles away); the low overall harms don't seem too relevant (just as with the coal fires, an incredibly low env. cost per AI query wouldn't necessarily help us locally if we're concentrating trillions of AI queries in one place); and the env. receptors don't seem to have sufficient capacity to absorb additional impacts (eg. much of the UK's water supply has very little headroom such that in dry weather the additional usage from people using hosepipes when gardening or washing their car is by itself enough to push us over 100% capacity and cause droughts).
Perhaps these sorts of "centralisation problems" don't occur much with AI infrastructure, I don't know - but I haven't seen much rigourous investigation of centralisation-related env. impacts (as distinct from investigation of overall env. impacts), and I don't think there's anybody really capable of and interested in investigating them rigourously other than you!
> This is a weird position for me to be in! I’ve always been a good predictable lib and don’t find myself on the opposite side of almost all media coverage often. It’s driven me a little crazy and does make me worry I’m missing something everyone else seems to see. But after a year of following this with no really satisfying answers, it’s become hard not to start to think I’m the one with my feet on the ground here.
You should probably sit down and think to yourself about what other issues "almost all" media coverage that a "good predictable lib" might be exposed to takes a position on, where that position is actually badly wrong; and isn't getting systematically corrected because people keep insisting that the people who bring up arguments against it are shills for some power or conspiracy theorists. It's certainly not just the issue of AI data centers.
You know Andy, some of us have been waiting for your *non* data center writing too
Thanks, Andy. I think it is just vibes, as Ruth below says. It is just another piece of misguided virtue signaling, like being anti-nuclear power, being entirely incapable of rational thought about climate change, etc. I greatly appreciate your work in this area.
I like your writing and it has also made me more obnoxious because people will randomly bring up data centers and water.
I dunno, I don't follow mainstream news really, so I don't really get rage baited by it but you don't have the same option
Even if AI significantly increases electricity demand, it does not automatically follow that this is environmentally negative although the concern is very legitimate. Historically, large new sources of demand have often triggered investment in new supply. The question is not whether AI will require more energy. It almost certainly will. The question is whether that demand accelerates deployment of cleaner energy sources or locks in more fossil fuel generation.
An excellent read. As a journalist, I have to agree that a lot of the writing leads to dead ends. But perhaps it’s the old newspaper adage: “Never let the facts stand in the way of a good story”
This is the kind of environmental argument I wish we saw more often: specific, numerate, and willing to separate actual harm from vibes in a panic hat.
Fear travels faster than numbers, and numbers require everyone to calm down long enough to read them.
Thanks for your efforts to demystify the controversy surrounding AI data centres. Keep it up!
Thank you for the thorough and patient analysis. Do you think data centers need to do a massively better PR job? Would that move the debate?