I like these articles. Have to say though that it's possible to waste power with computing. Crypto mining does that well. When you create an algorithm which becomes less efficient the more computing power is put into it, it doesn't matter how efficient computing is. But crypto is human programming, not computer programming, and I can't think of another example of blatant waste of computing power.
Yup crypto's the one exception to this bc the way crypto's mined incentives users to use as much compute as possible for lower and lower reward over time. Kind of a race to the bottom on energy that most other computer applications don't resemble.
I find it fascinating that the short quote from the WP under the McKinsey chart says that data centers "sapped" so much of "the country's electricity".
This framing is the issue in a nutshell. Electricity is depicted as a strained or fixed resource, while data centers using any of it is frivolous and wasteful, leaving less for its proper and valuable uses.
I do wonder how much of this is downstream of the fact that climate activism is often proposed in personal moral terms. Without any context for how much power usage is 'a lot', anything you do is problematic if you pay enough attention to it.
Really enjoyed this piece; quite eye-opening. I recently wrote a blog—which is not exactly along the same lines—where I argued that piracy was a morally justified stance in progressive circles. However, since the advent of LLMs, the progressive stance has changed, demanding stricter regulations and copyright laws—which will only hurt the small guy.
Wow, another great example showing how bad our intuitions about energy consumption are sometimes.
I think dematerialization is something people are aware of, but we're easily fooled by how powerful computers are, which makes us overestimate their energy use.
What I am kind of missing from your article is a graph about energy consumption in the US over time to support your point. To me it seems to do so, but it's not quite obvious - better efficiency does not necessarily mean less consumption.
Thanks for your coverage of this! Karen Hao's Empire of AI book also promotes the "generative AI is bad because of the climate impact" idea. Why do you think this line of argument is so compelling to people when it is so easily refuted?
I think a few things: 1) AI's so relatively new and fast-growing that a lot of people haven't had time to make sense of it. I think for a lot of people this is their first time hearing about how data centers work. 2) There's an understandable fear of AI either making things dumb or displacing workers, and people have bad vibes around it and are more willing to believe other stories about it being bad. 3) A lot of people just kind of struggle with climate numeracy in general. There have been a ton of instances of people just completely misunderstanding where the real problems are and this might just be a continuation of that. 4) A lot of people are assuming AI is useless after some bad initial experiences with it. IMO everyone should try poking around the models a bit.
I like these articles. Have to say though that it's possible to waste power with computing. Crypto mining does that well. When you create an algorithm which becomes less efficient the more computing power is put into it, it doesn't matter how efficient computing is. But crypto is human programming, not computer programming, and I can't think of another example of blatant waste of computing power.
Yup crypto's the one exception to this bc the way crypto's mined incentives users to use as much compute as possible for lower and lower reward over time. Kind of a race to the bottom on energy that most other computer applications don't resemble.
I find it fascinating that the short quote from the WP under the McKinsey chart says that data centers "sapped" so much of "the country's electricity".
This framing is the issue in a nutshell. Electricity is depicted as a strained or fixed resource, while data centers using any of it is frivolous and wasteful, leaving less for its proper and valuable uses.
I do wonder how much of this is downstream of the fact that climate activism is often proposed in personal moral terms. Without any context for how much power usage is 'a lot', anything you do is problematic if you pay enough attention to it.
I thought in 2008 that “Focus on complex system stuff instead of just trying to make yourself personally morally pure” would catch on more than it has
Really enjoyed this piece; quite eye-opening. I recently wrote a blog—which is not exactly along the same lines—where I argued that piracy was a morally justified stance in progressive circles. However, since the advent of LLMs, the progressive stance has changed, demanding stricter regulations and copyright laws—which will only hurt the small guy.
Wow, another great example showing how bad our intuitions about energy consumption are sometimes.
I think dematerialization is something people are aware of, but we're easily fooled by how powerful computers are, which makes us overestimate their energy use.
What I am kind of missing from your article is a graph about energy consumption in the US over time to support your point. To me it seems to do so, but it's not quite obvious - better efficiency does not necessarily mean less consumption.
Happy to add that!
Just added
Thanks for your coverage of this! Karen Hao's Empire of AI book also promotes the "generative AI is bad because of the climate impact" idea. Why do you think this line of argument is so compelling to people when it is so easily refuted?
I think a few things: 1) AI's so relatively new and fast-growing that a lot of people haven't had time to make sense of it. I think for a lot of people this is their first time hearing about how data centers work. 2) There's an understandable fear of AI either making things dumb or displacing workers, and people have bad vibes around it and are more willing to believe other stories about it being bad. 3) A lot of people just kind of struggle with climate numeracy in general. There have been a ton of instances of people just completely misunderstanding where the real problems are and this might just be a continuation of that. 4) A lot of people are assuming AI is useless after some bad initial experiences with it. IMO everyone should try poking around the models a bit.