Very interesting piece, but I think it finds its limits towards the end. You admit it's legitimate even as a liberal to say "You could be achieving the exact same thing with fewer resources." But doesn't this just push the disagreement back a step to what is "the exact same thing"? If my friend chooses a flight that takes 4 hours 30 minutes door to door over a train that takes 5 hours because he wants to get to his week-long vacation 30 minutes sooner, I guess a truly even-handed liberal would say "I respect that you value that extra 30 minutes at the cabin enough that you are willing to blow your carbon budget on it", but I think even within liberalism there needs to space to say, "You are being irrational, that 30 minutes doesn't matter, you could be achieving *basically* the same thing with fewer resources."
Yeah I think it's legitimate and good to debate how worth the climate cost it is to save some time in general, partly because you can fill your time with anything you personally want so it's more value neutral, but I think in the AI case it's not only saving people time, but preventing them from literally emitting more by drawing more energy on their laptops etc.
I think the AI image case is a great example of salient disagreements on whether it's “the exact same thing”, since a lot of people a) don't like the aesthetics of AI art and/or b) don't like the morals/values of AI art (AI being trained on human art without compensation artists, plagiarism concerns, etc).
The liberal framework allows both you to say what you say in this article, and people who dislike AI images to say that the time and emissions savings is moot because of any or all of:
a) it's not in their view producing the exact same thing and/or
b) it’s immoral (similar to how you, and I, would be anti-factory farming even if it was zero emissions) and/or
c) is also pretty neglible additional emissions in the grand scheme of things — in line with the overall point of your article about full comparisons
If you’ve got a correctly calibrated carbon tax, you don’t need to make these sorts of arguments about whether they have correctly calculated the value of that half hour with respect to its carbon cost: they pay it, so they are forced to do that calculation.
Andy’s framing is even more constraining than that: if they have an actual hard carbon budget that we are all limited to and have to include the trip in that budget then they will be forced to trade it off against all the other things that impact that budget.
This leaves much less necessity for arguing about whether people are valuing their personal choices rightly or wrongly.
It’s all relative though. A one-off half hour might be irrelevant, but a regular half hour could add up to a lot of time. If I’m travelling for work, would I get more work done during train journey or a plane journey, is one more reliable than the other, i.e. what are the chances of the half hour turning into an hour or more? What’s the £ cost difference? I’m not necessarily disagreeing I just think that for most people the surface level difference is only a small part of the comparison.
I don’t think it’s even liberal to tell other people what they should value and expect them to change their behavior to fit your particular opinions. While you’re telling that person to not take the plane because you think the 30 minutes they add to their vacation is wasteful, someone can tell you to just skip the vacation entirely and keep working because that would create more resources for everyone.
It believe that would be an illiberal opinion, as would be the opinion that one should not take a plane or a vacation, but it would not be liberal to prevent someone from saying or publishing such an illiberal opinion , but freedom of speech doesn’t make whatever one says or writes liberal. So, I guess I’ll amend my statement to say that the telling is not the problem, it’s the expectation that others need to respect your illiberal opinions or values that is the issue.
Andy, this was spectacular - both in its particular application and its generalizable modality. I think liberalism is underdefended right now, and I think many attempts to defend liberalism either conflate it with status-quo institutionalism or with moralization about pluralism. What I think you’ve done instead, successfully, is dropped into a gears-level breakdown about why liberalism is the necessary design solution to i) the presence of competing moralities absent ii) an imperative to eradicate them.
One tiny quibble. Everything you say about how the moral is we should have norms that say: "here's your carbon budget, meet it however you like" feel to me like much stronger reasons to think we should advocate for a (global? I know not realistic) carbon tax. Asking people to keep track of their emissions strikes me as not really realistic as a norm. For one, it's not public, so people can't really socially enforce the norm, so it's probably not a social equilibrium, as it's too easy to cheat on. And for another, it's a lot of effort to keep track of one's carbon footprint; not a realistic thing for most people to do.
Rather, asking people to treat using carbon as a reason not to perform an activity, but one that can be outweighed if the activity is high enough in (subjective) value is exactly the kind of thing that the price system is meant to do best. Nice thing about prices is that when some good becomes more scarce, prices can get everybody to consume less of it, without keeping track of how much of that good their various activities use. E.g., if eggs become scarce, and thereby become more expensive, then I automatically get a reason to consume fewer eggs. It would be a lot harder if everybody had to keep track of how scarce eggs were at all times, and attempt to consume less of them when they were scarcer (but without anything like price signals to convey that info compactly). So I feel like you've offered an (excellent) extended argument that a Pigouvian tax is the right social response to the fact that carbon emissions are costly, while treating it as an argument for a complex social/moral norm instead.
Thanks so much! Yeah this is kind of laying a lot of groundwork to say “Hey look at that carbon taxes are so great” but wanted to keep the post short enough. I think if people are making any normative claims about cutting emissions sans carbon tax, they mostly need to be with some kind of general goal for generally reducing emissions. It makes less sense to go industry by industry, so just asking people to pick a number and aim for it makes more sense, but less sense than if we just priced it correctly
Love this: If my friend drove a few miles to film a video, I wouldn’t say “You know, for the energy you used to drive over to film that video, you could have filtered 1.5 million spam emails.”
Enjoyed this and fully aligns with my views. In the UK at least, the Green parties have now become totally focused on an authoritarian view of changing the political system to the point where they barely mention the climate or environment any more. They could have built a large and sympathetic base but instead they opt for being the extremist, populist left in England and propping up the nationalist delusion in Scotland.
“Harm = CO2 emitted / (how much the thing lets people pursue their own values, or how much someone personally values it).”
I read this piece as an extended argument for why we should price emissions and water and let the market sort it out (with the right design features to address distributional and other issues that arise).
"At the extreme end, some environmentalists argue explicitly that political liberalism is itself responsible for the climate crisis, a claim that usually relies on heterodox economic theories. I don’t want to get into that right now."
Do you plan to get into that at a later date? I'm interested in your take.
As always Andy, I find your work on this topic rigorous, your perspective worthy, and your writing compelling. I too found the UNU document puzzling, and was reminded of another example of the kind of numerate but very narratively questionable interpretation pattern in the academic literature I came across earlier this year. https://www.nature.com/articles/s43247-026-03537-5
I want to note that I I think the core experimental work is admirable and as far as I can tell thoughtfully executed, so this isn't my attempt to disparage these authors. I just felt their careful experimental diligence for analyzing composition of GPUs was then paired with a questionable classification of what materials were hazardous. Then in the discussion, they specifically avoided putting their estimates in comparative contexts with other material flows because doing so would "risk obscuring rather than contextualizing the scale of AI’s material demand".
This is a fascinating and perspective enlarging article. Thank you. It does interestingly focus almost entirely on personal choice and less on the role of values in determining policy in a democracy. How does this argument apply to communal resources that are managed through policy and structural choices beyond our personal ones? For example, if I think too much of our federal budget is spent on the military and not enough on social services, and I vote for people who agree with that or lobby or protest to change policy, is that illiberal? Under your argument, it would seem like it might be, as I am assigning a moral value to one kind of government action, and then using spending as a cudgel to inhibit that, whereas many other people might think the military is the best use of federal dollars. But it’s not enough to just say well be more peaceful in your day-to-day life so that there is less chance we will go to war, but don’t expect others to accommodate your personal values as between military spending and social services. It seems to me that an exercise of values at the communal level (the environmental resources are finite resource = I think we should be spending less of them on things that don’t advantage us as a society) is exactly the kind of conversations we should be having at the community and national level in a functional democracy.
Great post, Andy. Ultimately, everything comes down to values but there are more or less liberal ways to bracket facts from values, and it's really important to be*aware* of how one is indeed bracketing. That's been a big problem with the environmental movement.
This is so good. I wonder how it might be generalized beyond environmental issues to other issues where people use an unrelated topic as a cudgel to try to impose their value preferences. Other manifestations of nimbyism feel like an obvious extension. Social conservatives often engage in this as well, claiming to want to help kids or whatever as a veneer for imposing their sexual morality or aesthetic preferences, in a way that was a pretty central part of political discourse before Trump. But because tribal, illiberal urges are so widespread, there's probably a version for every political faction (including liberals).
Excellent article as usual, and I also have to add as a sidenote that I mentally credit you for changing my views when I make choices to cut down on my meat consumption.
Near the end of the article, you mention that "the average AI image" takes 0.3-1Wh. The source for this is a study from May 2025 that uses a 1024x1024 image from Stable Diffusion 3 Medium (a 2B parameter model) as its benchmark. This is not representative of the average AI image generation in July 2026. For your book mill image, I'm guessing you used the image generation function in either Gemini or ChatGPT. Both of these are much larger and more sophisticated than SD3, which can comfortably run fully offline on a laptop but can't produce accurate text the way your image shows. Most AI images I see now in 2026 in "real life" are posters or ad images with complex text.
Neither company publishes their parameter counts or architecture for the image models, so it's extremely hard to estimate energy use, particularly when they're hosted in data centers that may be more efficient than what was available in May 2025. I spent a while doing research about what the energy consumption might be if the model was 20B parameters, which is a reasonable guess based on the capabilities of the largest open-source image models. The range I got was "maybe as little as 3x, maybe as much as 30x" depending on the way the service decides to structure the prompts to the image models. Sometimes, for example, Gemini runs a generation loop where it generates draft images to refine the final version.
This is not that helpful of a critique, I'm aware! I just felt this particular number jumped out at me as surprisingly low, especially in context of the state of the art of image generation today.
This is very good. Thank you for writing it. Liberal values are really under attack from all sides, including from people who dont even intend to as you lay out! Great example with Catholicism and overall great piece.
Hi Andy, just read your piece - thank you so much for writing, definitely got me thinking! A few questions that came up for me that I'd love your help thinking about if you can find a moment:
- What would be the liberal way to discuss whether or not climate change is something that's even worth addressing by curbing carbon emissions in the first place? What is the liberal way to discuss this topic with people who don't see climate change as an important / existential issue that merits prioritization (as is the case for the current US political administration and many of their supporters).
- Is it illiberal try to convince people that something is bad - even if it's only bad from a subjective values perspective? In a liberal democracy, if the vast majority of society ends up agreeing that emissions spent on industrial animal agriculture are not worth it because they believe industrial animal agriculture is bad - as I understand it, you on a personal level would perceive this to be a positive development. In line with this, would it not be useful for you to try to persuade people that industrial animal agriculture is not only carbon emitting, but also bad for other values-based reasons? Is that persuasion effort illiberal? Is a similar persuasion for those who believe AI is bad also illiberal? Or is your argument that if these two conversations happen separately - 1. the carbon emission limit that we should be setting per person & 2. the ethical issues with industrial animal agriculture (not including carbon emissions) - then this is in line with liberalism?
Yup my position’s that climate change is very bad, this is something that people with lots of different values can all agree on, so it’s pretty easy to just say so directly and communicate about why. I don’t think it’s illiberal to go out and try to convince people for your subjective values, it’s only bad to try to force them into those values using physical force or something that amounts to basically lying
To return to the Catholic environmentalists, for them to behave liberally, rather than argue in their report that other churches should be shutdown in order to curb emissions, they should've outlined a plan to indoctrinate the supporters of other churches such that those supporters willingly leave those churches and join the Catholic church. Then with those other churches shutdown, there will be greater carbon efficiency with all supporters using the single Catholic church infrastructure. Am I understanding that correctly?
What is the role of regulation in all of this? Do you think industrial animal agriculture (IAA) should never be banned in alignment with liberalism? Even if you successfully persuaded the vast majority of the public that IAA is bad, and legislators voted democratically agreed that it is bad, do you believe that the market should be allowed to continue existing so long as some consumers are willing to participate in the market for IAA? And if eventually there is a threshold by which it is ok, in your opinion, to ban IAA, how does this then apply back to churches? Can smaller churches eventually be banned to curb carbon emissions if enough people think they are bad - or should there be a special carve out protecting religious freedoms no matter how unpopular? What about a ban on AI if enough people think it's bad?
I asked ChatGPT to compare your take here with my views. It came up with this, which really does capture my reaction:
Masley grounds the argument in liberal neutrality and people’s ability to pursue different values. I would ground it more fundamentally in suffering. Environmental harms deserve attention according to the suffering they cause; environmental benefits deserve attention according to the suffering they prevent. Nothing additional is gained by invoking “the environment” as an independent value.
Very interesting piece, but I think it finds its limits towards the end. You admit it's legitimate even as a liberal to say "You could be achieving the exact same thing with fewer resources." But doesn't this just push the disagreement back a step to what is "the exact same thing"? If my friend chooses a flight that takes 4 hours 30 minutes door to door over a train that takes 5 hours because he wants to get to his week-long vacation 30 minutes sooner, I guess a truly even-handed liberal would say "I respect that you value that extra 30 minutes at the cabin enough that you are willing to blow your carbon budget on it", but I think even within liberalism there needs to space to say, "You are being irrational, that 30 minutes doesn't matter, you could be achieving *basically* the same thing with fewer resources."
Yeah I think it's legitimate and good to debate how worth the climate cost it is to save some time in general, partly because you can fill your time with anything you personally want so it's more value neutral, but I think in the AI case it's not only saving people time, but preventing them from literally emitting more by drawing more energy on their laptops etc.
I think the AI image case is a great example of salient disagreements on whether it's “the exact same thing”, since a lot of people a) don't like the aesthetics of AI art and/or b) don't like the morals/values of AI art (AI being trained on human art without compensation artists, plagiarism concerns, etc).
The liberal framework allows both you to say what you say in this article, and people who dislike AI images to say that the time and emissions savings is moot because of any or all of:
a) it's not in their view producing the exact same thing and/or
b) it’s immoral (similar to how you, and I, would be anti-factory farming even if it was zero emissions) and/or
c) is also pretty neglible additional emissions in the grand scheme of things — in line with the overall point of your article about full comparisons
If you’ve got a correctly calibrated carbon tax, you don’t need to make these sorts of arguments about whether they have correctly calculated the value of that half hour with respect to its carbon cost: they pay it, so they are forced to do that calculation.
Andy’s framing is even more constraining than that: if they have an actual hard carbon budget that we are all limited to and have to include the trip in that budget then they will be forced to trade it off against all the other things that impact that budget.
This leaves much less necessity for arguing about whether people are valuing their personal choices rightly or wrongly.
It’s all relative though. A one-off half hour might be irrelevant, but a regular half hour could add up to a lot of time. If I’m travelling for work, would I get more work done during train journey or a plane journey, is one more reliable than the other, i.e. what are the chances of the half hour turning into an hour or more? What’s the £ cost difference? I’m not necessarily disagreeing I just think that for most people the surface level difference is only a small part of the comparison.
I don’t think it’s even liberal to tell other people what they should value and expect them to change their behavior to fit your particular opinions. While you’re telling that person to not take the plane because you think the 30 minutes they add to their vacation is wasteful, someone can tell you to just skip the vacation entirely and keep working because that would create more resources for everyone.
I mean do you think it's illiberal for a Catholic to write a book about how more people should be Catholic?
It believe that would be an illiberal opinion, as would be the opinion that one should not take a plane or a vacation, but it would not be liberal to prevent someone from saying or publishing such an illiberal opinion , but freedom of speech doesn’t make whatever one says or writes liberal. So, I guess I’ll amend my statement to say that the telling is not the problem, it’s the expectation that others need to respect your illiberal opinions or values that is the issue.
Andy, this was spectacular - both in its particular application and its generalizable modality. I think liberalism is underdefended right now, and I think many attempts to defend liberalism either conflate it with status-quo institutionalism or with moralization about pluralism. What I think you’ve done instead, successfully, is dropped into a gears-level breakdown about why liberalism is the necessary design solution to i) the presence of competing moralities absent ii) an imperative to eradicate them.
This was a *wonderful* post.
One tiny quibble. Everything you say about how the moral is we should have norms that say: "here's your carbon budget, meet it however you like" feel to me like much stronger reasons to think we should advocate for a (global? I know not realistic) carbon tax. Asking people to keep track of their emissions strikes me as not really realistic as a norm. For one, it's not public, so people can't really socially enforce the norm, so it's probably not a social equilibrium, as it's too easy to cheat on. And for another, it's a lot of effort to keep track of one's carbon footprint; not a realistic thing for most people to do.
Rather, asking people to treat using carbon as a reason not to perform an activity, but one that can be outweighed if the activity is high enough in (subjective) value is exactly the kind of thing that the price system is meant to do best. Nice thing about prices is that when some good becomes more scarce, prices can get everybody to consume less of it, without keeping track of how much of that good their various activities use. E.g., if eggs become scarce, and thereby become more expensive, then I automatically get a reason to consume fewer eggs. It would be a lot harder if everybody had to keep track of how scarce eggs were at all times, and attempt to consume less of them when they were scarcer (but without anything like price signals to convey that info compactly). So I feel like you've offered an (excellent) extended argument that a Pigouvian tax is the right social response to the fact that carbon emissions are costly, while treating it as an argument for a complex social/moral norm instead.
Thanks so much! Yeah this is kind of laying a lot of groundwork to say “Hey look at that carbon taxes are so great” but wanted to keep the post short enough. I think if people are making any normative claims about cutting emissions sans carbon tax, they mostly need to be with some kind of general goal for generally reducing emissions. It makes less sense to go industry by industry, so just asking people to pick a number and aim for it makes more sense, but less sense than if we just priced it correctly
Added a bit about carbon taxes because I really should've originally
Love this: If my friend drove a few miles to film a video, I wouldn’t say “You know, for the energy you used to drive over to film that video, you could have filtered 1.5 million spam emails.”
Enjoyed this and fully aligns with my views. In the UK at least, the Green parties have now become totally focused on an authoritarian view of changing the political system to the point where they barely mention the climate or environment any more. They could have built a large and sympathetic base but instead they opt for being the extremist, populist left in England and propping up the nationalist delusion in Scotland.
“Harm = CO2 emitted / (how much the thing lets people pursue their own values, or how much someone personally values it).”
I read this piece as an extended argument for why we should price emissions and water and let the market sort it out (with the right design features to address distributional and other issues that arise).
"At the extreme end, some environmentalists argue explicitly that political liberalism is itself responsible for the climate crisis, a claim that usually relies on heterodox economic theories. I don’t want to get into that right now."
Do you plan to get into that at a later date? I'm interested in your take.
Yup!
As always Andy, I find your work on this topic rigorous, your perspective worthy, and your writing compelling. I too found the UNU document puzzling, and was reminded of another example of the kind of numerate but very narratively questionable interpretation pattern in the academic literature I came across earlier this year. https://www.nature.com/articles/s43247-026-03537-5
I want to note that I I think the core experimental work is admirable and as far as I can tell thoughtfully executed, so this isn't my attempt to disparage these authors. I just felt their careful experimental diligence for analyzing composition of GPUs was then paired with a questionable classification of what materials were hazardous. Then in the discussion, they specifically avoided putting their estimates in comparative contexts with other material flows because doing so would "risk obscuring rather than contextualizing the scale of AI’s material demand".
This is a fascinating and perspective enlarging article. Thank you. It does interestingly focus almost entirely on personal choice and less on the role of values in determining policy in a democracy. How does this argument apply to communal resources that are managed through policy and structural choices beyond our personal ones? For example, if I think too much of our federal budget is spent on the military and not enough on social services, and I vote for people who agree with that or lobby or protest to change policy, is that illiberal? Under your argument, it would seem like it might be, as I am assigning a moral value to one kind of government action, and then using spending as a cudgel to inhibit that, whereas many other people might think the military is the best use of federal dollars. But it’s not enough to just say well be more peaceful in your day-to-day life so that there is less chance we will go to war, but don’t expect others to accommodate your personal values as between military spending and social services. It seems to me that an exercise of values at the communal level (the environmental resources are finite resource = I think we should be spending less of them on things that don’t advantage us as a society) is exactly the kind of conversations we should be having at the community and national level in a functional democracy.
Great post, Andy. Ultimately, everything comes down to values but there are more or less liberal ways to bracket facts from values, and it's really important to be*aware* of how one is indeed bracketing. That's been a big problem with the environmental movement.
This is so good. I wonder how it might be generalized beyond environmental issues to other issues where people use an unrelated topic as a cudgel to try to impose their value preferences. Other manifestations of nimbyism feel like an obvious extension. Social conservatives often engage in this as well, claiming to want to help kids or whatever as a veneer for imposing their sexual morality or aesthetic preferences, in a way that was a pretty central part of political discourse before Trump. But because tribal, illiberal urges are so widespread, there's probably a version for every political faction (including liberals).
Excellent article as usual, and I also have to add as a sidenote that I mentally credit you for changing my views when I make choices to cut down on my meat consumption.
Near the end of the article, you mention that "the average AI image" takes 0.3-1Wh. The source for this is a study from May 2025 that uses a 1024x1024 image from Stable Diffusion 3 Medium (a 2B parameter model) as its benchmark. This is not representative of the average AI image generation in July 2026. For your book mill image, I'm guessing you used the image generation function in either Gemini or ChatGPT. Both of these are much larger and more sophisticated than SD3, which can comfortably run fully offline on a laptop but can't produce accurate text the way your image shows. Most AI images I see now in 2026 in "real life" are posters or ad images with complex text.
Neither company publishes their parameter counts or architecture for the image models, so it's extremely hard to estimate energy use, particularly when they're hosted in data centers that may be more efficient than what was available in May 2025. I spent a while doing research about what the energy consumption might be if the model was 20B parameters, which is a reasonable guess based on the capabilities of the largest open-source image models. The range I got was "maybe as little as 3x, maybe as much as 30x" depending on the way the service decides to structure the prompts to the image models. Sometimes, for example, Gemini runs a generation loop where it generates draft images to refine the final version.
This is not that helpful of a critique, I'm aware! I just felt this particular number jumped out at me as surprisingly low, especially in context of the state of the art of image generation today.
This is fantastic Andy. I’ll be reading this with my students.
This is very good. Thank you for writing it. Liberal values are really under attack from all sides, including from people who dont even intend to as you lay out! Great example with Catholicism and overall great piece.
Hi Andy, just read your piece - thank you so much for writing, definitely got me thinking! A few questions that came up for me that I'd love your help thinking about if you can find a moment:
- What would be the liberal way to discuss whether or not climate change is something that's even worth addressing by curbing carbon emissions in the first place? What is the liberal way to discuss this topic with people who don't see climate change as an important / existential issue that merits prioritization (as is the case for the current US political administration and many of their supporters).
- Is it illiberal try to convince people that something is bad - even if it's only bad from a subjective values perspective? In a liberal democracy, if the vast majority of society ends up agreeing that emissions spent on industrial animal agriculture are not worth it because they believe industrial animal agriculture is bad - as I understand it, you on a personal level would perceive this to be a positive development. In line with this, would it not be useful for you to try to persuade people that industrial animal agriculture is not only carbon emitting, but also bad for other values-based reasons? Is that persuasion effort illiberal? Is a similar persuasion for those who believe AI is bad also illiberal? Or is your argument that if these two conversations happen separately - 1. the carbon emission limit that we should be setting per person & 2. the ethical issues with industrial animal agriculture (not including carbon emissions) - then this is in line with liberalism?
Yup my position’s that climate change is very bad, this is something that people with lots of different values can all agree on, so it’s pretty easy to just say so directly and communicate about why. I don’t think it’s illiberal to go out and try to convince people for your subjective values, it’s only bad to try to force them into those values using physical force or something that amounts to basically lying
To return to the Catholic environmentalists, for them to behave liberally, rather than argue in their report that other churches should be shutdown in order to curb emissions, they should've outlined a plan to indoctrinate the supporters of other churches such that those supporters willingly leave those churches and join the Catholic church. Then with those other churches shutdown, there will be greater carbon efficiency with all supporters using the single Catholic church infrastructure. Am I understanding that correctly?
What is the role of regulation in all of this? Do you think industrial animal agriculture (IAA) should never be banned in alignment with liberalism? Even if you successfully persuaded the vast majority of the public that IAA is bad, and legislators voted democratically agreed that it is bad, do you believe that the market should be allowed to continue existing so long as some consumers are willing to participate in the market for IAA? And if eventually there is a threshold by which it is ok, in your opinion, to ban IAA, how does this then apply back to churches? Can smaller churches eventually be banned to curb carbon emissions if enough people think they are bad - or should there be a special carve out protecting religious freedoms no matter how unpopular? What about a ban on AI if enough people think it's bad?
I asked ChatGPT to compare your take here with my views. It came up with this, which really does capture my reaction:
Masley grounds the argument in liberal neutrality and people’s ability to pursue different values. I would ground it more fundamentally in suffering. Environmental harms deserve attention according to the suffering they cause; environmental benefits deserve attention according to the suffering they prevent. Nothing additional is gained by invoking “the environment” as an independent value.