An example I'd like to borrow is the perchlorate contamination of groundwater issue, common during the late cold war era. On one hand the nation justifies these occurrences as national defense necessity; on the other hand whatever justifications that be, there is no justification for poisoning people's food and water and subjecting them to diseases.
Thus I say geopolitical and community concerns are two aspects of national AI mobilization, not two levels. The word "level" would invite the slippery thinking that higher level concerns ought to overrule lower level concerns. It is not necessarily true in every case, and not true in the case of today's AI construction spree. That slippery thinking was the norm, at the height of the cold war. This is why these lessons become the mandatory knowledge of university students, so that we know not to repeat these mistakes.
On the first point, it make sense, and in general I get the problem with thinking AI is magical, but I don't see what opposition one can have generally to the proposed non-proliferation framework? What is your general position about that?
"But making them find other places to build is the whole reason for the grassroots opposition to data center construction." But where are those other places? Climate apocalyse is a real possibility, even a reality now, even if superintelligence apocalypse is not, and data centers are going to accelerate it, or make things worse with respect to the environment.
So to base everything on what 'the people want' and what their reasons for opposing AI centres are, doesn't seem like a great idea to me. My guess is that not many in Wisconsin will be protesting against data centres being built in other parts of the world, where the people have lesser political agency, and those are going to mess the whole world up anyway, it seems like..
Non-proliferation of data centres that essentially no people in the world are asking for, or need, is a good idea maybe? Or are the climate concerns around data centres also unfounded?
Hi Ayush! If it was a *data center* non-proliferation treaty that would be one thing. But what he's talking about is essentially throttling the use of chips and data centers to limit the power of the models that can be trained. That doesn't get rid of the data centers. In the best case for the communities involved, they would have a *longer* useful life extracting resources from their communities, etc. Even in the best case scenario, you'd end up with comparable water consumption, emissions, etc, just stretched out over a longer period of time.
More likely, in my view, the AI bubble will burst and leave many data centers and chips as stranded assets, and in that case, the slow burn of throttling the number of chips that can be used could have an even worse effect--until the bubble bursts, companies would rush to build new data centers with the most efficient chips (bear in mind that data center construction and chip purchase agreements are made at least a year and often 2 or more before they come online, so the data centers with the most efficient chips are at the earliest stage in that pipeline) so that they can squeeze the most utility out of the limited GPU capacity they are allowed to use. That would likely leave data centers with older chips as fallow white elephants, which would be a huge drag on local economies where they are. I frankly think that is the most likely scenario. Data center moratoria would help with that.
In any event, it's all academic. No AI non-proliferation treaty will happen. The data center fight, on the other hand, is very much happening every day. And my critique of Karma's piece was the elitism implicit in suggesting that local community concerns are trivial--which would be condescending even if the concern he had was real.
An example I'd like to borrow is the perchlorate contamination of groundwater issue, common during the late cold war era. On one hand the nation justifies these occurrences as national defense necessity; on the other hand whatever justifications that be, there is no justification for poisoning people's food and water and subjecting them to diseases.
Thus I say geopolitical and community concerns are two aspects of national AI mobilization, not two levels. The word "level" would invite the slippery thinking that higher level concerns ought to overrule lower level concerns. It is not necessarily true in every case, and not true in the case of today's AI construction spree. That slippery thinking was the norm, at the height of the cold war. This is why these lessons become the mandatory knowledge of university students, so that we know not to repeat these mistakes.
On the first point, it make sense, and in general I get the problem with thinking AI is magical, but I don't see what opposition one can have generally to the proposed non-proliferation framework? What is your general position about that?
"But making them find other places to build is the whole reason for the grassroots opposition to data center construction." But where are those other places? Climate apocalyse is a real possibility, even a reality now, even if superintelligence apocalypse is not, and data centers are going to accelerate it, or make things worse with respect to the environment.
So to base everything on what 'the people want' and what their reasons for opposing AI centres are, doesn't seem like a great idea to me. My guess is that not many in Wisconsin will be protesting against data centres being built in other parts of the world, where the people have lesser political agency, and those are going to mess the whole world up anyway, it seems like..
Non-proliferation of data centres that essentially no people in the world are asking for, or need, is a good idea maybe? Or are the climate concerns around data centres also unfounded?
Hi Ayush! If it was a *data center* non-proliferation treaty that would be one thing. But what he's talking about is essentially throttling the use of chips and data centers to limit the power of the models that can be trained. That doesn't get rid of the data centers. In the best case for the communities involved, they would have a *longer* useful life extracting resources from their communities, etc. Even in the best case scenario, you'd end up with comparable water consumption, emissions, etc, just stretched out over a longer period of time.
More likely, in my view, the AI bubble will burst and leave many data centers and chips as stranded assets, and in that case, the slow burn of throttling the number of chips that can be used could have an even worse effect--until the bubble bursts, companies would rush to build new data centers with the most efficient chips (bear in mind that data center construction and chip purchase agreements are made at least a year and often 2 or more before they come online, so the data centers with the most efficient chips are at the earliest stage in that pipeline) so that they can squeeze the most utility out of the limited GPU capacity they are allowed to use. That would likely leave data centers with older chips as fallow white elephants, which would be a huge drag on local economies where they are. I frankly think that is the most likely scenario. Data center moratoria would help with that.
In any event, it's all academic. No AI non-proliferation treaty will happen. The data center fight, on the other hand, is very much happening every day. And my critique of Karma's piece was the elitism implicit in suggesting that local community concerns are trivial--which would be condescending even if the concern he had was real.