This just doesn't make sense in light of last week's addition of chip export controls.
The first 1/3 argues blocking construction wouldn't actually slow AI development that much. But with chip export controls, the Bernie+AOC proposal is actually the closest thing to a pause anyone in office is offering.
And the arguments about chip controls don't make sense. US leverage is eroding regardless of what the US does. China is already trying as hard as it can to catch up the frontier on AI chips, but experts think this is still years away.
As someone who frantically updated their book in response on exactly this topic last week, I'm sympathetic to the difficulty of covering a rapidly evolving issue, but as it's written, this piece doesn't engage with the strongest version of the proposal.
yes, the bill text does lead with x-risk quotes in the findings, but i cited the public statement intentionally! i thought it was pretty revealing that the public-facing messaging Sanders chose leads with electricity costs / the environment, jobs, distinctly non-ai safety issues. if the bill itself were deeply ai-safety oriented, that gap might not matter much; but then the operative provisions lean just as heavily on labor and resource consumption as the statement, even if ai safety is in there.
re: export controls — we agree that US chip leverage is eroding, and i certainly agree that this proposal is “the closest thing to a pause anyone in office is offering” in the sense that if it was politically plausible, it would at least marginally slow ai development.
so, i’m not sure why you’d spend our lead and narrowing window on pushing us and other countries to adopt a bundle of American labor and environmental priorities, rather than deploying export controls strategically and independently like actual AI safety tools.
Moreover, the export controls are conditioned on "laws and regulations in place to protect humanity from AI safety concerns and existential risks, protect workers, and protect the environment to the effect of Section 3."
AI safety and x-risk are the first things named! And honestly, based on Bernie's reorientation on this, those policies seem like the bigger non-negotiables to him than the labor and env stuff.
I think there's a case for a messaging bill like this from an AI safety perspective. Essentially it may be useful as an exercise in coalition building. It's introducing people who are worried about environmental or job market impacts to the concept of societal AI risks in a way that they might actually listen, since it's also taking their concerns seriously. It's seeding the idea that worrying about AI risk is a natural companion to worrying about environmental or job market impacts of AI.
A lot of politics works this way, where you have to stitch together groups that are coming from different starting points into a semi-coherent coalition. We see this in YIMBYism, for example, which bundles together free marketers, developers, urbanists, young renters in expensive cities, homeowners who want greater property rights (e.g. allowing ADU construction), and even socialists concerned about cost of living like Mamdani. Some of these folks detest one another: I doubt libertarians concerned about property rights have much love of Mamdani, and visa versa. YIMBYs have been good about avoiding alienating potential allies, and I think AI risk advocates should do the same.
In this case, hopefully this bill can get AI safety folks and anti-datacenter folks together in the same room, and hopefully the next iteration of the bill will include more of the kinds of ideas Nat lays out here.
Thank you for this comprehensive take on what Bernie Sanders is doing. I wasn't aware he's been fighting the uphill battle against AI data centers, but I'm glad he's doing it. To answer your question, "what would you do with the time bought", I would hope Congress and local leaders put into place proper protections - not only to benefit them, but also our environment and people. I love the initiatives you wrote about - I would hope they take action on those!
This just doesn't make sense in light of last week's addition of chip export controls.
The first 1/3 argues blocking construction wouldn't actually slow AI development that much. But with chip export controls, the Bernie+AOC proposal is actually the closest thing to a pause anyone in office is offering.
And the arguments about chip controls don't make sense. US leverage is eroding regardless of what the US does. China is already trying as hard as it can to catch up the frontier on AI chips, but experts think this is still years away.
It's hard not to read this as a piece that was written about just the data center moratorium and then was insufficiently updated in light of the change to the proposal. For instance, it cites Bernie's February proposal when observing that he doesn't lead with x-risk. But his proposal with AOC last week opens with a bunch of quotes about job loss and x-risk: https://www.sanders.senate.gov/wp-content/uploads/Artificial-Intelligence-Data-Center-Moratorium-Act-Section-by-Section.pdf
As someone who frantically updated their book in response on exactly this topic last week, I'm sympathetic to the difficulty of covering a rapidly evolving issue, but as it's written, this piece doesn't engage with the strongest version of the proposal.
copying this reply here from twitter:
thanks for the comments! a couple notes:
yes, the bill text does lead with x-risk quotes in the findings, but i cited the public statement intentionally! i thought it was pretty revealing that the public-facing messaging Sanders chose leads with electricity costs / the environment, jobs, distinctly non-ai safety issues. if the bill itself were deeply ai-safety oriented, that gap might not matter much; but then the operative provisions lean just as heavily on labor and resource consumption as the statement, even if ai safety is in there.
re: export controls — we agree that US chip leverage is eroding, and i certainly agree that this proposal is “the closest thing to a pause anyone in office is offering” in the sense that if it was politically plausible, it would at least marginally slow ai development.
so, i’m not sure why you’d spend our lead and narrowing window on pushing us and other countries to adopt a bundle of American labor and environmental priorities, rather than deploying export controls strategically and independently like actual AI safety tools.
doing the same!
But you cited an old press release from Feb when there was this one from last week, with this as its first line "legislation that would enact a reasonable pause to the development of AI to ensure the safety of humanity": https://sanders.senate.gov/press-releases/news-sanders-ocasio-cortez-announce-ai-data-center-moratorium-act/
Moreover, the export controls are conditioned on "laws and regulations in place to protect humanity from AI safety concerns and existential risks, protect workers, and protect the environment to the effect of Section 3."
AI safety and x-risk are the first things named! And honestly, based on Bernie's reorientation on this, those policies seem like the bigger non-negotiables to him than the labor and env stuff.
I think there's a case for a messaging bill like this from an AI safety perspective. Essentially it may be useful as an exercise in coalition building. It's introducing people who are worried about environmental or job market impacts to the concept of societal AI risks in a way that they might actually listen, since it's also taking their concerns seriously. It's seeding the idea that worrying about AI risk is a natural companion to worrying about environmental or job market impacts of AI.
A lot of politics works this way, where you have to stitch together groups that are coming from different starting points into a semi-coherent coalition. We see this in YIMBYism, for example, which bundles together free marketers, developers, urbanists, young renters in expensive cities, homeowners who want greater property rights (e.g. allowing ADU construction), and even socialists concerned about cost of living like Mamdani. Some of these folks detest one another: I doubt libertarians concerned about property rights have much love of Mamdani, and visa versa. YIMBYs have been good about avoiding alienating potential allies, and I think AI risk advocates should do the same.
In this case, hopefully this bill can get AI safety folks and anti-datacenter folks together in the same room, and hopefully the next iteration of the bill will include more of the kinds of ideas Nat lays out here.
Thank you for this comprehensive take on what Bernie Sanders is doing. I wasn't aware he's been fighting the uphill battle against AI data centers, but I'm glad he's doing it. To answer your question, "what would you do with the time bought", I would hope Congress and local leaders put into place proper protections - not only to benefit them, but also our environment and people. I love the initiatives you wrote about - I would hope they take action on those!