WEBVTT
00:00:00.959 --> 00:00:02.879
Row Home Productions.
00:00:07.040 --> 00:00:17.679
You're listening to The Hum, a new podcast where we're sharing stories from people across the US who are fighting the incursion of AI and data centers into every aspect of our lives.
00:00:17.920 --> 00:00:20.000
I'm Saul Levin, your host.
00:00:20.320 --> 00:00:25.039
Last time, I talked to eight-year-old twins Oslo and Ivan and their mom Holly.
00:00:25.199 --> 00:00:32.479
The twins shared their strategy that turned out a hundred people at a data center protest in rural Texas to protect their school.
00:00:32.799 --> 00:00:42.320
This week, we talked to Nathan Kim, a PhD student at the University of Michigan fighting a data center on campus and studying tech in Silicon Valley.
00:00:42.560 --> 00:00:53.520
We talk about the fight at U of M, how students discovered that the data center was being built to develop weapons of war, and we link the data center fight to the broader dangers of AI.
00:00:59.439 --> 00:01:02.479
Hey Nathan, thank you so much for joining the HUM today.
00:01:02.719 --> 00:01:03.200
Thanks, all.
00:01:03.439 --> 00:01:04.239
Thanks for having me.
00:01:04.480 --> 00:01:05.280
Absolutely.
00:01:05.519 --> 00:01:09.040
So, Nathan, I'm really curious just to get us started.
00:01:09.359 --> 00:01:15.200
You were involved in fighting data centers as a PhD student, but what about Nathan?
00:01:15.359 --> 00:01:16.879
What about you personally?
00:01:17.120 --> 00:01:19.920
What's the story of how you got into this stuff in the first place?
00:01:20.319 --> 00:01:24.319
So it in undergrad, I think, I uh was interested in social justice.
00:01:24.400 --> 00:01:26.959
This was around uh 2020, 2021.
00:01:27.040 --> 00:01:32.239
There was a period where you know everybody has this question of how can we make the world a better place?
00:01:32.480 --> 00:01:33.840
There's a lot of things wrong with it.
00:01:34.000 --> 00:01:40.799
Well, what are the best ways we can sort of confront uh you know police violence and racism and so forth in order to make the world we want to live in?
00:01:40.959 --> 00:01:52.640
Unfortunately, as time went on, and especially as you know, with the election of with Trump's second election, these Silicon Valley executives were sort of at the front and center of you know so many like kind of evil things that are happening in the world.
00:01:52.879 --> 00:02:01.840
And uh, trying to understand how these different forces converge onto Silicon Valley is what led me to my place as a PhD student at the University of Michigan.
00:02:02.159 --> 00:02:09.840
So that's my um trying to understand the political economy, the motivations, the history of Silicon Valley, that's my main research project at the moment.
00:02:10.400 --> 00:02:17.840
In the December of 2024, the University of Michigan announced the lost a new data center that they were building with Los Alamos National Labs.
00:02:18.000 --> 00:02:21.759
So this is a gigantic uh$1.2 billion project.
00:02:22.080 --> 00:02:28.960
So it's uh at one point in time, yeah, um, you know, this is bigger than uh, for example, the NSA data center in Utah.
00:02:29.199 --> 00:02:33.919
For many decades, this would have been considered the biggest data center in the world at 110 megawatts.
00:02:34.080 --> 00:02:38.879
Today it's it's sort of like trump change compared to some of the open EI Stargate projects, of course.
00:02:39.199 --> 00:02:44.719
But it's still consuming more uh power than the entire town of Intalante, which is where it is being built.
00:02:44.960 --> 00:02:50.319
It's doing all this without community consent or involvement, um, and it's generated a lot of frustration.
00:02:50.560 --> 00:02:59.360
And for me, uh it's sort of hit a particular nerve because uh to me, this is unequivocally a project, uh a war-making project.
00:02:59.680 --> 00:03:02.159
It's a project to develop new nuclear weapons.
00:03:02.400 --> 00:03:11.520
That's Los Alamos National Lab's primary mission at the moment, especially with a new directive that Trump gave in 2018 to develop uh 80 new plutonium pits per year.
00:03:11.840 --> 00:03:19.759
And they need these giant data centers and supercomputers to run all the calculations they need to develop these new weapons and maintain the their existing ones.
00:03:20.159 --> 00:03:21.199
That's fascinating.
00:03:21.360 --> 00:03:28.159
So we hear, I mean, look, we hear from people who are in the fight against AI data centers for so many different reasons.
00:03:28.479 --> 00:03:33.840
You're saying that this AI data center is being used to build weapons of war.
00:03:34.000 --> 00:03:39.039
I think a lot of people are wondering what the AI data center near them is doing.
00:03:39.199 --> 00:03:44.560
What were the steps that you guys took to kind of figure that out and understand what the Los Alamos project was for?
00:03:44.879 --> 00:03:46.240
That's a great question, Saul.
00:03:46.400 --> 00:04:05.199
So here at the University of Michigan, we've filed a lot of FOIA requests and yeah, through public records requests, um, like FOIA requests, as well as uh, you know, just trying to dig through the archives and try to understand how did Los Alamos or the University of Michigan come to this moment historically, we've sort of arrived at our analysis thus far.
00:04:05.520 --> 00:04:32.959
So for instance, understanding the history of uh the Cold War and understanding that you know, beginning in 1992, with the world turning towards a comprehensive test ban treaty and the US discontinuing live tests, Los Alamos's primary mission in order to continue to enforce US nuclear supremacy, was to instead invest in supercomputing so that they could avoid detonating live bombs, but still ensure that all of the bombs that they had in their arsenal were uh up to date and were just as lethal and so on.
00:04:33.120 --> 00:04:46.319
Um and so that gave us the sort of context we needed to understand, you know, given Los Alamos's you know many different needs, their relative distribution of their budget, and the functions that they've performed in the past historically, what is this data center going to be used for?
00:04:46.560 --> 00:04:50.959
And the only possible answer from there was for the maintenance and development of nuclear weapons.
00:04:52.000 --> 00:04:53.120
That's unbelievable.
00:04:53.279 --> 00:04:59.600
Do a lot of people you think who are in the fight against the Los Alamos data center in Michigan come into the fight because of that?
00:04:59.839 --> 00:05:02.319
I mean, there's so many data center fights around the country.
00:05:02.480 --> 00:05:08.240
Do you think some people are coming in just because they're like, I don't want this building near me, I'm worried about the impact on water?
00:05:08.480 --> 00:05:16.720
Or are the majority of people coming in because they're worried about, you know, land in Michigan and other resources going to kill people abroad?
00:05:17.120 --> 00:05:20.720
I think more and more so uh people are starting to make these connections, right?
00:05:20.879 --> 00:05:39.600
They're starting to understand that the reason that uh their land is being taken from them, the reason that their water and electricity uh prices are going up, the reason that all that is happening is so that uh you know that so that the US can launch an operation to kidnap Nicholas Maduro or so that the US can wage war on Iran.
00:05:39.920 --> 00:05:44.959
It has these different needs and functions that Silicon Valley is you know very neatly lined up to serve.
00:05:45.279 --> 00:05:50.319
And as a consequence of that, people's own own resources here in the United States are also being taken for them.
00:05:50.639 --> 00:05:59.199
Often people don't start at that analysis, they might start just at you know understanding that their electricity bills about to go up, you know, 400%.
00:05:59.600 --> 00:06:00.959
Uh, we've seen in some cases.
00:06:01.199 --> 00:06:08.319
But through the course of the struggle, I think, and trying to understand what is it exactly that we're up against, we're able to make those connections together.
00:06:08.720 --> 00:06:09.279
Got you.
00:06:09.439 --> 00:06:10.879
Let's go back for a second.
00:06:10.959 --> 00:06:20.560
Uh, you guys found out the details, and and I mean, obviously you guys have some smart folks in your PhD program and stuff, but so maybe it's kind of an unusual group.
00:06:20.639 --> 00:06:32.560
But I think that a lot of other folks could learn from you guys and do FOIA requests to find out more about what the AI data center companies are doing and what they're developing these systems for.
00:06:32.800 --> 00:06:37.199
I talked to a reporter yesterday about FOIA requests.
00:06:37.360 --> 00:06:41.600
And what he said was don't ask for all the correspondence.
00:06:41.839 --> 00:06:45.600
He said, figure out who's most likely to be talking about what you're looking for.
00:06:45.759 --> 00:06:50.720
Maybe it's a county lawyer, and ask for a week of his emails at a likely time.
00:06:50.879 --> 00:06:55.920
And that way you can use clues from that request to vote an even more specific request next time.
00:06:56.160 --> 00:07:09.839
In other words, this reporter's suggestion, if you're interested in uh doing FOIA requests to figure out what's going on, is to sort of iterate a little bit until you can figure out what's happening at the data center and which different players are involved.
00:07:10.160 --> 00:07:13.920
You guys successfully gathered information through FOIA requests.
00:07:14.079 --> 00:07:15.360
Do you agree with that analysis?
00:07:15.439 --> 00:07:19.920
And what would you add to folks who are like, yo, I want to be like Nathan and his crew?
00:07:20.000 --> 00:07:25.279
Like, I want to figure out what kind of murky behavior is going on behind the walls of the data center.
00:07:25.600 --> 00:07:26.879
That's a great question, Saul.
00:07:27.040 --> 00:07:29.199
And it's also a great analysis from that reporter.
00:07:29.360 --> 00:07:31.040
It's something I largely agree with.
00:07:31.120 --> 00:07:35.279
And I'll just try to add some context here to understand why is that the case.
00:07:35.600 --> 00:07:41.920
So FOIA requests are uh can actually be super fun, especially when you conduct them as uh the reporter suggested.
00:07:42.079 --> 00:07:54.800
You know, you file small requests at a time, then the fees aren't so large, and you're able to get these little uh hints that and you know, uh a trail of crumbs that you can use to find your way to the gingerbread house or whatever their expression might be.
00:07:54.959 --> 00:08:06.240
In our in our FOIA, for instance, we found uh at first references to a uh non-disclosure agreement, an NDA between the University of Michigan and some of the partners that were involved in this project.
00:08:06.959 --> 00:08:14.639
And understanding that there were references to these NDAs uh was the thing that we needed to actually refer specifically to these NDAs.
00:08:14.800 --> 00:08:18.639
Um uh in other cases we've also received a lot of emails.
00:08:18.720 --> 00:08:29.120
Uh, sometimes these emails reference uh other conversations that people have had, other meetings, other emails that have been exchanged, and we are then able to go in and ask for uh additional emails.
00:08:29.279 --> 00:08:33.279
Um I do want to give some uh I guess caveats there though.
00:08:33.840 --> 00:08:36.720
You know, every uh government agency is different.
00:08:37.039 --> 00:08:40.480
Many of them take a really long time to give uh requests back.
00:08:40.639 --> 00:08:47.120
And so that would be the case for the University of Michigan or a lot of these federal labs, like the Los Alamos National Labs.
00:08:47.200 --> 00:08:51.759
Um and in that case, it's not as uh it's not always possible to do that kind of iterative approach.
00:08:52.399 --> 00:08:59.200
Another uh piece of context that people should know is that FOIA requests are very, very easy to file, right?
00:08:59.440 --> 00:09:06.559
Anyone can and everybody should be filing them because they're our are right as that's that's a tool we have to keep our government accountable.
00:09:06.799 --> 00:09:11.600
And so I I don't want people to sort of restrain themselves like I send only file one request at a time.
00:09:11.759 --> 00:09:14.639
We should be, you know, trying to answer all the questions that we have.
00:09:15.120 --> 00:09:15.759
Sure, sure.
00:09:15.919 --> 00:09:23.519
Yeah, I guess part of the filing one or you know, a small set of requests at a time isn't to say you shouldn't ask for accountability on everything.
00:09:23.600 --> 00:09:32.240
It's just literally like if you accidentally make it too wide, then it's very difficult uh to get it back in a timely fashion.
00:09:32.399 --> 00:09:35.360
Whoever is pulling those records is gonna take a long time.
00:09:35.600 --> 00:09:49.360
But I guess the lesson here is follow your guys' lead of doing some inductive reasoning on the front end and then also start with some a few narrow requests just so you can get on the right track and not have to request something broad and wait for months.
00:09:49.600 --> 00:09:56.080
I want to turn a little bit more to the research that you do, Nathan, and your expertise on AI.
00:09:56.480 --> 00:10:02.879
One of the things that I've been thinking about is that pretty much everything is called AI now, as far as I could tell.
00:10:03.120 --> 00:10:11.440
There's these little uh buggies that go around, I guess, delivering food on sidewalks in DC called Cocos.
00:10:11.679 --> 00:10:12.960
People say that's AI.
00:10:13.039 --> 00:10:26.960
They say that uh self-driving cars uh like Waymo's are AI, as is the updated Google search that for whatever reason is worse than the old Google search is now quote-unquote AI powered.
00:10:27.120 --> 00:10:34.399
Everything from that stuff to you know the robot takeovers depicted in iRobot, which is pretty scary.
00:10:34.480 --> 00:10:40.960
I'm thinking, you know, Will Smith isn't as young as he used to be, so I'm not really sure what our strategy is here.
00:10:41.039 --> 00:10:49.440
But could you just talk to us a little bit about what is artificial intelligence and why are so many things thought of as AI?
00:10:49.840 --> 00:10:50.720
Great question.
00:10:50.960 --> 00:11:00.480
So I think the number one thing that uh listeners and anti-data center activists everywhere should be aware of is that AI is not a single coherent project.
00:11:00.720 --> 00:11:09.600
It is not a single set of technologies, it is not some very neat thing where we can say this thing is clearly AI and that thing is not.
00:11:09.919 --> 00:11:11.919
These definitions are changing all the time.
00:11:12.080 --> 00:11:21.440
And the reason why it is changing is because companies like to use the term AI as a marketing term and as a political as to enhance their own political project.
00:11:21.919 --> 00:11:26.960
When we think of AI, uh listeners can just type AI into Google and just look at the images that come up.
00:11:27.360 --> 00:11:31.679
You see crazy sort of imaginations about what artificial intelligence actually is.
00:11:31.840 --> 00:11:41.919
You see pictures of brains, pictures of humanoid robots, um, and not necessarily a lot of the mundane uh and often very straightforward math that is happening behind the scenes.
00:11:42.399 --> 00:11:55.840
That is very useful for Silicon Valley entrepreneurs and executives because it allows them to pretend that what they are building is something that is uh incomprehensible, that nobody can control, that's there's no point for us to call for accountability over.
00:11:56.000 --> 00:12:00.080
And through that, they then they would have free reign to develop the AI that they want.
00:12:00.559 --> 00:12:21.759
I think that I've heard you refer to AI as a marketing project, which sounds related to what you're saying, but if it's you know marketing and advertising, could you explain a little bit more about how that sort of mystification of AI contributes to the profits of these companies or what they're trying to do with their AI and their data centers?
00:12:22.159 --> 00:12:26.879
So, yeah, let's let's talk about maybe uh about what these data centers are are actually doing, right?
00:12:27.120 --> 00:12:38.879
We don't recognize enough and make uh understand enough that all the all these campaigns against block cameras are fundamentally part of the same struggle against data centers, um, which need which require uh data centers to operate.
00:12:38.960 --> 00:12:52.000
So these networks of I think uh at this point over 120,000 automated license plate reader cameras, um they are uh able to have the analysis and track people the way that they do because they have models running on the data centers.
00:12:52.320 --> 00:13:20.159
For Flock in particular, I think, which is a very politically vulnerable spot for them, they really want to uh ensure that they can avoid accountability, they can avoid any public scrutiny, which they've obviously failed miserably at, but uh that's where terms like AI are useful for them, because you know, by conjuring this image of AI and human progress and robots, they can pretend that they're leading us into a better world, even as the main function that they're doing is increasing control, both corporate and government control over uh everyday people.
00:13:20.559 --> 00:13:22.879
To me, it seems like AI is dangerous.
00:13:23.039 --> 00:13:32.000
You know, it's it's scary whether we're thinking about the data centers or the flot cameras or all this talk about runaway AI agents.
00:13:32.559 --> 00:13:36.960
Whoever you talk to or whoever we talk to seems to be concerned about AI.
00:13:37.200 --> 00:13:43.039
But how can we understand how AI is dangerous and what the potential harms are?
00:13:43.360 --> 00:13:45.919
So I like to think of this question in sort of two parts.
00:13:46.080 --> 00:13:49.759
You know, the real question that we're trying to answer is why is AI dangerous?
00:13:49.919 --> 00:13:58.639
Is it dangerous because nobody can control AI and it's going to eventually spiral out of control and release, uh, you know, lead to human extinction?
00:13:58.799 --> 00:14:20.480
Or is it dangerous because of the world that we live in and because our society is structured such that you know companies have this profit motive to release AI that they know has serious defects, that they are not able to create any safety protocols for, that they know is you know uh are using massive amounts of resources and being used for surveillance and weapons manufacturing and so on.
00:14:20.720 --> 00:14:22.799
And I think it's firmly that latter case.
00:14:22.879 --> 00:14:35.840
Um the reason why AI is dangerous for us is uh because of the this sort of capitalist world that we live in and the profit motive I think that companies are pursuing so that they each can pursue short-term gains while the rest of us suffer for it.
00:14:36.000 --> 00:14:40.240
And we know that from observing what the actual harms of AI are today.
00:14:40.480 --> 00:14:50.240
It's not necessarily uh we haven't seen AI be uh being used by a nefarious actor in order to pursue some kind of existential human uh project of human extinction, right?
00:14:50.399 --> 00:14:53.440
We haven't seen anybody try to release some kind of new crazy virus.
00:14:53.679 --> 00:15:01.840
But we have seen AI do, just as we've seen uh uh people use technology for over the past uh 40 years, are for these projects of surveillance.
00:15:01.919 --> 00:15:13.039
Um, I'm thinking about cases where you know cops, uh, there's what there was one Texas cop who, in search of a woman who uh was suspected of an abortion, searched over 83,000 flaw cameras.
00:15:13.200 --> 00:15:20.240
And so that's the kind of uh thing that is you need technology uh like AI in order to do to manage data at that scale.
00:15:20.399 --> 00:15:35.039
And yet it's not something that is at the level of human extinction, it's the way it is, and because there are there's limited police accountability in this country because our power structures are in such a way that that we cannot hold anybody to account, either government officials or the tech elite.
00:15:35.600 --> 00:15:43.200
Yeah, I mean, it seems like one of the things that keeps coming up and that we talk about on the hum is that this is a fight for control.
00:15:43.519 --> 00:15:50.159
I mean, it is about the data centers because people hate data centers, and that's valid because they're kind of a horrible deal.
00:15:50.320 --> 00:15:53.919
But this overall fight is like, well, the data centers don't have any rules.
00:15:54.080 --> 00:15:55.279
The AI has no rules.
00:15:55.440 --> 00:16:09.600
If we actually had meaningful regulations that were about uh protecting water, protecting air, making sure that regular people weren't paying the costs of companies making huge profits and weapons of war, then it'd be a very different conversation.
00:16:09.759 --> 00:16:19.919
But because of the lack of rules and regulations, it's making this a lot scarier and a lot more uh, you know, the build-out of AI and the build-out of data centers are both pretty out of control.
00:16:20.720 --> 00:16:23.600
Let's talk, let's let's zoom out for a second, Nathan.
00:16:23.679 --> 00:16:32.960
I mean, every day there's all kinds of articles in every newspaper about this AI economic bubble that's gonna burst.
00:16:33.360 --> 00:16:46.000
That, you know, these companies are leveraging money they don't have to pay for AI data centers, that they're borrowing and renting chips and space from each other with money that each other don't have.
00:16:46.159 --> 00:16:57.360
And because you know, of private equity and other sort of fancy money schemes, they're essentially not being honest or sharing how much money they're making, losing, or spending.
00:16:57.600 --> 00:17:02.879
And that could cause the economy to crash or certainly get very out of whack.
00:17:03.039 --> 00:17:09.839
Uh, how are you thinking about this bubble and what do you think is going to happen in the next few months or years?
00:17:10.240 --> 00:17:18.160
So I think that there's a complicated answer here, I think, because um there's sort of two truths that listeners need to hold in their head at the same time.
00:17:18.799 --> 00:17:21.039
One, there definitely is a bubble, right?
00:17:21.200 --> 00:17:24.000
I I don't think anybody, even in the AI industry, denies that.
00:17:24.160 --> 00:17:32.079
There's this collective understanding that, yeah, OpenAI is saying all sorts of very hyped-up claims about their model capabilities and so on.
00:17:32.240 --> 00:17:42.240
And therefore, that's a lot of the massive amounts of funding that open AI has required, because it depends on these kinds of overblown narratives, is in turn sort of detached from the underlying economics.
00:17:42.400 --> 00:17:48.319
Um there's also this understanding that um the extent to which that is true is is somewhat unknown.
00:17:48.400 --> 00:17:59.759
Um, and that is why the bubble has not collapsed yet necessarily, because the exact amount to which the this claim is overblown can can't actually be asserted, and nobody's ruling to really go on that on that bet.
00:18:00.720 --> 00:18:21.200
So what I mean is that you know, uh for a project like OpenAI Stargate, which supposedly is is will constitute uh you know over$100 billion in overall investment, it's not clear to investors uh exactly how how much that is going to fall short by in terms of the uh the actual uh uh income that OpenAI will make uh and Oracle will make into that deal.
00:18:21.759 --> 00:18:25.680
So that's one truth I think that listeners need to hold in their head.
00:18:25.759 --> 00:18:26.880
That there is a bubble, I guess.
00:18:26.960 --> 00:18:30.880
There is this amount of uh speculation, I think, that and instability within the market.
00:18:32.079 --> 00:18:50.799
But the other truth, I think, um, which is a little bit harder to digest, is that even as a bubble exists and keeps growing larger, these companies, uh especially the hyperscalers, these so-called uh hyperscalers at the center of this, meaning Amazon, Google slash Alphabet, or Microsoft, they are more profitable than ever before.
00:18:51.039 --> 00:19:01.279
And so even as they're spending and borrowing lots and lots and lots of money to try to finance their data center build out, and that is contributing to their instability, they're also raking in more money than ever.
00:19:01.680 --> 00:19:12.480
And so it's kind of a race between these two things to understand you know which is going to get to these companies first, um, you know, the revenue that will follow the hype or you know the consequences of so much borrowing.
00:19:13.440 --> 00:19:23.599
And the other part of that pill that I think can be a little hard to swallow is that you know, we've seen in this country um that it's hugely dependent on uh a small set of firms, right?
00:19:23.839 --> 00:19:30.559
Even just last year, AI CapEx uh capital expenditures contributed more to US GDP growth than all of consumer spending.
00:19:31.599 --> 00:19:35.519
So we have these like few points of failure in the US economy right now.
00:19:35.759 --> 00:19:57.759
This is a very similar dynamic that existed in 2008, for instance, where uh the US government had a very, very uh real motive to prop up some banks and some industries, because if those went down, then there would be a dominant domino effect where many, many things, uh lots of other companies and industries that were seemingly unrelated to the actual heart of the problem would go down as a result.
00:19:58.160 --> 00:20:20.400
And the government has this sort of too big to fail incentive to prop up, you know, whether that's by you know helping Goldman Sachs buy bear sterns, as happened in 2008, or in uh, you know, potentially in 2030, uh bailing out NVIDIA or providing a sort of a uh interest-free um loan to one of these companies that is that's going under so that this domino effect is stopped short.
00:20:20.559 --> 00:20:34.400
And so I think it's it it because of that, um, and especially because of our current administration and the sort of political climate right now, I think it would be a little naive of us to expect the bubble to uh to pop, I think, and and hurt the capitalists where it hurts most.
00:20:34.720 --> 00:20:46.559
Yeah, I mean it sounds like without some kind of policies or guardrails, the people who created the problem are not going to be held accountable, most likely in that scenario, which is really scary.
00:20:46.799 --> 00:20:50.000
My last question for you is about policy.
00:20:50.240 --> 00:20:55.599
What policy direction would you like to see us go uh based on these concerns that you're raising?
00:20:55.920 --> 00:21:00.880
So I like to break this up into maybe three different types of policy measures, right?
00:21:01.039 --> 00:21:15.680
And they're ordered by that just because of the simple fact that right now, a lot of the solutions that we can be tempted to go for and would be very effective in a functioning uh egalitarian society are really quite harmful in the actual world that we live in.
00:21:15.839 --> 00:21:27.359
So the main example I'm thinking of here is that uh of what we should not do is something like uh you know Bernie Sanders' proposal of just you know, right now today, nationalizing 50% of AI companies.
00:21:28.000 --> 00:21:38.720
It would have a number of harmful effects and backfire on us, um, in part because of what AI under this particular society is structured to do, labor displacements and uh depressing wages and so on.
00:21:38.799 --> 00:21:54.079
Um and of course, not to mention all the environmental harms, but also because, of course, um if we were to nationalize AI the AI industry, you know, under the current administration, I think that that that would probably worsen the problem of AI development and data center development rather than allow greater public control.
00:21:54.400 --> 00:21:59.680
So you're worried about public control because it would sort of tie the fate and economics of the U.S.
00:22:00.160 --> 00:22:04.160
Government to a lot of the problems and instability of the AI industry.
00:22:04.480 --> 00:22:05.119
Is that right?
00:22:05.440 --> 00:22:11.839
Yes, and because the uh so-called public control right now of these companies is not actually public.
00:22:12.000 --> 00:22:13.440
It's in fact state control.
00:22:13.680 --> 00:22:29.039
And given the Trump administration and his control over the state and it's very, very large interest in seeing data centers proliferate, uh a stake in these companies would help the Trump administration uh accelerate the data center development instead of giving the public some levers to stop.
00:22:29.519 --> 00:22:40.079
So it you know, in response to that, I think this kind of bind where a lot of these seemingly best measures that we have, you know, are are sort of doomed to fail because of the society we live in, I think.
00:22:40.400 --> 00:22:48.960
We have to have an approach that is both addressing the problems that we face today and preparing for the society that we do actually live in that will allow us to take these more radical actions.
00:22:49.599 --> 00:22:59.440
So, you know, today I think every action, every policy measure we do has to be designed with the mind of increasing our own power over these tech companies, right?
00:22:59.759 --> 00:23:18.000
So I'm thinking about a nationwide data center moratorium, which would be useful to you know give people time to recruit, time to think of these broader, more capacious strategies, but also certain policy measures to ensure that the these movements which have been very effective in stopping data centers can be protected and can proliferate.
00:23:18.160 --> 00:23:23.920
So right now we have uh an administration which is very hostile to activism of all kinds, right?
00:23:24.000 --> 00:23:28.960
It's uh they've called data center um activists terrorists funded by China and so on.
00:23:29.200 --> 00:23:32.480
They're pursuing felony charges against uh anti-ICE protesters.
00:23:32.559 --> 00:23:37.680
Um so I'm thinking about the Minnesota 15 and the Prairie Land uh defendants, um, as well as the Michigan 8.
00:23:38.000 --> 00:23:51.039
We need some kind of measure of free speech um and you know, blanket protections for the for activists so that all these tactics that we do have to stop data centers can be made more successful, can be encouraged, can be, can proliferate.
00:23:51.440 --> 00:23:57.119
Other measures in that regard might include things like lowering the bar for a ballot initiative to sort of get on the table.
00:23:57.519 --> 00:24:08.480
Right now, a lot of uh counties or municipalities have you know eight to fifteen percent of uh you know the previous elections voters need to put a signature down in order for for a measure just to get on the ballot.
00:24:08.720 --> 00:24:14.240
But that is often an insurmountably large um burden, I think, for many, for an actual democratic process.
00:24:14.400 --> 00:24:26.079
It requires tens and thousands of dollars of funding in most cases and prevents citizens uh who are under-resourced, uh, which uh which, of course, uh these activists find data centers are likely to be given where data centers are built.
00:24:26.240 --> 00:24:30.720
Uh so it poses these very large obstacles, I think, for democratic governance.
00:24:31.039 --> 00:24:31.839
Awesome.
00:24:32.160 --> 00:24:37.920
Nathan Cam is a PhD candidate at the University of Michigan School of Information.
00:24:38.079 --> 00:24:44.720
He's been organizing in the fight against the Los Alamos Data Center being built at the University of Michigan.
00:24:44.960 --> 00:24:47.680
Thank you so much, Nathan, for joining us on the Hump today.
00:24:47.920 --> 00:24:48.720
Thanks all.
00:24:52.079 --> 00:24:57.039
That was my conversation with University of Michigan graduate student Nathan Kim.
00:24:57.279 --> 00:25:06.880
People all over the country are copying Michigan students' strategies for revealing that AI data centers are being built for war and profit, not local benefits.
00:25:07.200 --> 00:25:09.200
You're listening to The Hum.
00:25:09.519 --> 00:25:17.839
We're publishing new episodes weekly, so if you enjoyed what you heard, give us an enthusiastic five-star review that's at least five words long.
00:25:18.079 --> 00:25:24.079
Message us on Instagram at TheHumPod to get connected to organizers in your state or country.
00:25:24.319 --> 00:25:27.599
The Hum is produced in partnership with Rohome Productions.
00:25:27.839 --> 00:25:30.480
Rohome's creative director is Alex Lewis.
00:25:30.799 --> 00:25:33.039
Their executive producer is John Myers.
00:25:33.440 --> 00:25:35.680
Our producer is Emily Rizzo.
00:25:35.839 --> 00:25:41.599
The Hum's video producer is Adultrehan and we partner with the Center for Nonviolent Conflict Research.
00:25:41.839 --> 00:25:44.960
I'm Stol Levin, your host and banjo journeyman.
00:25:45.119 --> 00:25:48.160
Thanks for listening and see you next Tuesday.
00:00:00.959 --> 00:00:02.879
Row Home Productions.
00:00:07.040 --> 00:00:17.679
You're listening to The Hum, a new podcast where we're sharing stories from people across the US who are fighting the incursion of AI and data centers into every aspect of our lives.
00:00:17.920 --> 00:00:20.000
I'm Saul Levin, your host.
00:00:20.320 --> 00:00:25.039
Last time, I talked to eight-year-old twins Oslo and Ivan and their mom Holly.
00:00:25.199 --> 00:00:32.479
The twins shared their strategy that turned out a hundred people at a data center protest in rural Texas to protect their school.
00:00:32.799 --> 00:00:42.320
This week, we talked to Nathan Kim, a PhD student at the University of Michigan fighting a data center on campus and studying tech in Silicon Valley.
00:00:42.560 --> 00:00:53.520
We talk about the fight at U of M, how students discovered that the data center was being built to develop weapons of war, and we link the data center fight to the broader dangers of AI.
00:00:59.439 --> 00:01:02.479
Hey Nathan, thank you so much for joining the HUM today.
00:01:02.719 --> 00:01:03.200
Thanks, all.
00:01:03.439 --> 00:01:04.239
Thanks for having me.
00:01:04.480 --> 00:01:05.280
Absolutely.
00:01:05.519 --> 00:01:09.040
So, Nathan, I'm really curious just to get us started.
00:01:09.359 --> 00:01:15.200
You were involved in fighting data centers as a PhD student, but what about Nathan?
00:01:15.359 --> 00:01:16.879
What about you personally?
00:01:17.120 --> 00:01:19.920
What's the story of how you got into this stuff in the first place?
00:01:20.319 --> 00:01:24.319
So it in undergrad, I think, I uh was interested in social justice.
00:01:24.400 --> 00:01:26.959
This was around uh 2020, 2021.
00:01:27.040 --> 00:01:32.239
There was a period where you know everybody has this question of how can we make the world a better place?
00:01:32.480 --> 00:01:33.840
There's a lot of things wrong with it.
00:01:34.000 --> 00:01:40.799
Well, what are the best ways we can sort of confront uh you know police violence and racism and so forth in order to make the world we want to live in?
00:01:40.959 --> 00:01:52.640
Unfortunately, as time went on, and especially as you know, with the election of with Trump's second election, these Silicon Valley executives were sort of at the front and center of you know so many like kind of evil things that are happening in the world.
00:01:52.879 --> 00:02:01.840
And uh, trying to understand how these different forces converge onto Silicon Valley is what led me to my place as a PhD student at the University of Michigan.
00:02:02.159 --> 00:02:09.840
So that's my um trying to understand the political economy, the motivations, the history of Silicon Valley, that's my main research project at the moment.
00:02:10.400 --> 00:02:17.840
In the December of 2024, the University of Michigan announced the lost a new data center that they were building with Los Alamos National Labs.
00:02:18.000 --> 00:02:21.759
So this is a gigantic uh$1.2 billion project.
00:02:22.080 --> 00:02:28.960
So it's uh at one point in time, yeah, um, you know, this is bigger than uh, for example, the NSA data center in Utah.
00:02:29.199 --> 00:02:33.919
For many decades, this would have been considered the biggest data center in the world at 110 megawatts.
00:02:34.080 --> 00:02:38.879
Today it's it's sort of like trump change compared to some of the open EI Stargate projects, of course.
00:02:39.199 --> 00:02:44.719
But it's still consuming more uh power than the entire town of Intalante, which is where it is being built.
00:02:44.960 --> 00:02:50.319
It's doing all this without community consent or involvement, um, and it's generated a lot of frustration.
00:02:50.560 --> 00:02:59.360
And for me, uh it's sort of hit a particular nerve because uh to me, this is unequivocally a project, uh a war-making project.
00:02:59.680 --> 00:03:02.159
It's a project to develop new nuclear weapons.
00:03:02.400 --> 00:03:11.520
That's Los Alamos National Lab's primary mission at the moment, especially with a new directive that Trump gave in 2018 to develop uh 80 new plutonium pits per year.
00:03:11.840 --> 00:03:19.759
And they need these giant data centers and supercomputers to run all the calculations they need to develop these new weapons and maintain the their existing ones.
00:03:20.159 --> 00:03:21.199
That's fascinating.
00:03:21.360 --> 00:03:28.159
So we hear, I mean, look, we hear from people who are in the fight against AI data centers for so many different reasons.
00:03:28.479 --> 00:03:33.840
You're saying that this AI data center is being used to build weapons of war.
00:03:34.000 --> 00:03:39.039
I think a lot of people are wondering what the AI data center near them is doing.
00:03:39.199 --> 00:03:44.560
What were the steps that you guys took to kind of figure that out and understand what the Los Alamos project was for?
00:03:44.879 --> 00:03:46.240
That's a great question, Saul.
00:03:46.400 --> 00:04:05.199
So here at the University of Michigan, we've filed a lot of FOIA requests and yeah, through public records requests, um, like FOIA requests, as well as uh, you know, just trying to dig through the archives and try to understand how did Los Alamos or the University of Michigan come to this moment historically, we've sort of arrived at our analysis thus far.
00:04:05.520 --> 00:04:32.959
So for instance, understanding the history of uh the Cold War and understanding that you know, beginning in 1992, with the world turning towards a comprehensive test ban treaty and the US discontinuing live tests, Los Alamos's primary mission in order to continue to enforce US nuclear supremacy, was to instead invest in supercomputing so that they could avoid detonating live bombs, but still ensure that all of the bombs that they had in their arsenal were uh up to date and were just as lethal and so on.
00:04:33.120 --> 00:04:46.319
Um and so that gave us the sort of context we needed to understand, you know, given Los Alamos's you know many different needs, their relative distribution of their budget, and the functions that they've performed in the past historically, what is this data center going to be used for?
00:04:46.560 --> 00:04:50.959
And the only possible answer from there was for the maintenance and development of nuclear weapons.
00:04:52.000 --> 00:04:53.120
That's unbelievable.
00:04:53.279 --> 00:04:59.600
Do a lot of people you think who are in the fight against the Los Alamos data center in Michigan come into the fight because of that?
00:04:59.839 --> 00:05:02.319
I mean, there's so many data center fights around the country.
00:05:02.480 --> 00:05:08.240
Do you think some people are coming in just because they're like, I don't want this building near me, I'm worried about the impact on water?
00:05:08.480 --> 00:05:16.720
Or are the majority of people coming in because they're worried about, you know, land in Michigan and other resources going to kill people abroad?
00:05:17.120 --> 00:05:20.720
I think more and more so uh people are starting to make these connections, right?
00:05:20.879 --> 00:05:39.600
They're starting to understand that the reason that uh their land is being taken from them, the reason that their water and electricity uh prices are going up, the reason that all that is happening is so that uh you know that so that the US can launch an operation to kidnap Nicholas Maduro or so that the US can wage war on Iran.
00:05:39.920 --> 00:05:44.959
It has these different needs and functions that Silicon Valley is you know very neatly lined up to serve.
00:05:45.279 --> 00:05:50.319
And as a consequence of that, people's own own resources here in the United States are also being taken for them.
00:05:50.639 --> 00:05:59.199
Often people don't start at that analysis, they might start just at you know understanding that their electricity bills about to go up, you know, 400%.
00:05:59.600 --> 00:06:00.959
Uh, we've seen in some cases.
00:06:01.199 --> 00:06:08.319
But through the course of the struggle, I think, and trying to understand what is it exactly that we're up against, we're able to make those connections together.
00:06:08.720 --> 00:06:09.279
Got you.
00:06:09.439 --> 00:06:10.879
Let's go back for a second.
00:06:10.959 --> 00:06:20.560
Uh, you guys found out the details, and and I mean, obviously you guys have some smart folks in your PhD program and stuff, but so maybe it's kind of an unusual group.
00:06:20.639 --> 00:06:32.560
But I think that a lot of other folks could learn from you guys and do FOIA requests to find out more about what the AI data center companies are doing and what they're developing these systems for.
00:06:32.800 --> 00:06:37.199
I talked to a reporter yesterday about FOIA requests.
00:06:37.360 --> 00:06:41.600
And what he said was don't ask for all the correspondence.
00:06:41.839 --> 00:06:45.600
He said, figure out who's most likely to be talking about what you're looking for.
00:06:45.759 --> 00:06:50.720
Maybe it's a county lawyer, and ask for a week of his emails at a likely time.
00:06:50.879 --> 00:06:55.920
And that way you can use clues from that request to vote an even more specific request next time.
00:06:56.160 --> 00:07:09.839
In other words, this reporter's suggestion, if you're interested in uh doing FOIA requests to figure out what's going on, is to sort of iterate a little bit until you can figure out what's happening at the data center and which different players are involved.
00:07:10.160 --> 00:07:13.920
You guys successfully gathered information through FOIA requests.
00:07:14.079 --> 00:07:15.360
Do you agree with that analysis?
00:07:15.439 --> 00:07:19.920
And what would you add to folks who are like, yo, I want to be like Nathan and his crew?
00:07:20.000 --> 00:07:25.279
Like, I want to figure out what kind of murky behavior is going on behind the walls of the data center.
00:07:25.600 --> 00:07:26.879
That's a great question, Saul.
00:07:27.040 --> 00:07:29.199
And it's also a great analysis from that reporter.
00:07:29.360 --> 00:07:31.040
It's something I largely agree with.
00:07:31.120 --> 00:07:35.279
And I'll just try to add some context here to understand why is that the case.
00:07:35.600 --> 00:07:41.920
So FOIA requests are uh can actually be super fun, especially when you conduct them as uh the reporter suggested.
00:07:42.079 --> 00:07:54.800
You know, you file small requests at a time, then the fees aren't so large, and you're able to get these little uh hints that and you know, uh a trail of crumbs that you can use to find your way to the gingerbread house or whatever their expression might be.
00:07:54.959 --> 00:08:06.240
In our in our FOIA, for instance, we found uh at first references to a uh non-disclosure agreement, an NDA between the University of Michigan and some of the partners that were involved in this project.
00:08:06.959 --> 00:08:14.639
And understanding that there were references to these NDAs uh was the thing that we needed to actually refer specifically to these NDAs.
00:08:14.800 --> 00:08:18.639
Um uh in other cases we've also received a lot of emails.
00:08:18.720 --> 00:08:29.120
Uh, sometimes these emails reference uh other conversations that people have had, other meetings, other emails that have been exchanged, and we are then able to go in and ask for uh additional emails.
00:08:29.279 --> 00:08:33.279
Um I do want to give some uh I guess caveats there though.
00:08:33.840 --> 00:08:36.720
You know, every uh government agency is different.
00:08:37.039 --> 00:08:40.480
Many of them take a really long time to give uh requests back.
00:08:40.639 --> 00:08:47.120
And so that would be the case for the University of Michigan or a lot of these federal labs, like the Los Alamos National Labs.
00:08:47.200 --> 00:08:51.759
Um and in that case, it's not as uh it's not always possible to do that kind of iterative approach.
00:08:52.399 --> 00:08:59.200
Another uh piece of context that people should know is that FOIA requests are very, very easy to file, right?
00:08:59.440 --> 00:09:06.559
Anyone can and everybody should be filing them because they're our are right as that's that's a tool we have to keep our government accountable.
00:09:06.799 --> 00:09:11.600
And so I I don't want people to sort of restrain themselves like I send only file one request at a time.
00:09:11.759 --> 00:09:14.639
We should be, you know, trying to answer all the questions that we have.
00:09:15.120 --> 00:09:15.759
Sure, sure.
00:09:15.919 --> 00:09:23.519
Yeah, I guess part of the filing one or you know, a small set of requests at a time isn't to say you shouldn't ask for accountability on everything.
00:09:23.600 --> 00:09:32.240
It's just literally like if you accidentally make it too wide, then it's very difficult uh to get it back in a timely fashion.
00:09:32.399 --> 00:09:35.360
Whoever is pulling those records is gonna take a long time.
00:09:35.600 --> 00:09:49.360
But I guess the lesson here is follow your guys' lead of doing some inductive reasoning on the front end and then also start with some a few narrow requests just so you can get on the right track and not have to request something broad and wait for months.
00:09:49.600 --> 00:09:56.080
I want to turn a little bit more to the research that you do, Nathan, and your expertise on AI.
00:09:56.480 --> 00:10:02.879
One of the things that I've been thinking about is that pretty much everything is called AI now, as far as I could tell.
00:10:03.120 --> 00:10:11.440
There's these little uh buggies that go around, I guess, delivering food on sidewalks in DC called Cocos.
00:10:11.679 --> 00:10:12.960
People say that's AI.
00:10:13.039 --> 00:10:26.960
They say that uh self-driving cars uh like Waymo's are AI, as is the updated Google search that for whatever reason is worse than the old Google search is now quote-unquote AI powered.
00:10:27.120 --> 00:10:34.399
Everything from that stuff to you know the robot takeovers depicted in iRobot, which is pretty scary.
00:10:34.480 --> 00:10:40.960
I'm thinking, you know, Will Smith isn't as young as he used to be, so I'm not really sure what our strategy is here.
00:10:41.039 --> 00:10:49.440
But could you just talk to us a little bit about what is artificial intelligence and why are so many things thought of as AI?
00:10:49.840 --> 00:10:50.720
Great question.
00:10:50.960 --> 00:11:00.480
So I think the number one thing that uh listeners and anti-data center activists everywhere should be aware of is that AI is not a single coherent project.
00:11:00.720 --> 00:11:09.600
It is not a single set of technologies, it is not some very neat thing where we can say this thing is clearly AI and that thing is not.
00:11:09.919 --> 00:11:11.919
These definitions are changing all the time.
00:11:12.080 --> 00:11:21.440
And the reason why it is changing is because companies like to use the term AI as a marketing term and as a political as to enhance their own political project.
00:11:21.919 --> 00:11:26.960
When we think of AI, uh listeners can just type AI into Google and just look at the images that come up.
00:11:27.360 --> 00:11:31.679
You see crazy sort of imaginations about what artificial intelligence actually is.
00:11:31.840 --> 00:11:41.919
You see pictures of brains, pictures of humanoid robots, um, and not necessarily a lot of the mundane uh and often very straightforward math that is happening behind the scenes.
00:11:42.399 --> 00:11:55.840
That is very useful for Silicon Valley entrepreneurs and executives because it allows them to pretend that what they are building is something that is uh incomprehensible, that nobody can control, that's there's no point for us to call for accountability over.
00:11:56.000 --> 00:12:00.080
And through that, they then they would have free reign to develop the AI that they want.
00:12:00.559 --> 00:12:21.759
I think that I've heard you refer to AI as a marketing project, which sounds related to what you're saying, but if it's you know marketing and advertising, could you explain a little bit more about how that sort of mystification of AI contributes to the profits of these companies or what they're trying to do with their AI and their data centers?
00:12:22.159 --> 00:12:26.879
So, yeah, let's let's talk about maybe uh about what these data centers are are actually doing, right?
00:12:27.120 --> 00:12:38.879
We don't recognize enough and make uh understand enough that all the all these campaigns against block cameras are fundamentally part of the same struggle against data centers, um, which need which require uh data centers to operate.
00:12:38.960 --> 00:12:52.000
So these networks of I think uh at this point over 120,000 automated license plate reader cameras, um they are uh able to have the analysis and track people the way that they do because they have models running on the data centers.
00:12:52.320 --> 00:13:20.159
For Flock in particular, I think, which is a very politically vulnerable spot for them, they really want to uh ensure that they can avoid accountability, they can avoid any public scrutiny, which they've obviously failed miserably at, but uh that's where terms like AI are useful for them, because you know, by conjuring this image of AI and human progress and robots, they can pretend that they're leading us into a better world, even as the main function that they're doing is increasing control, both corporate and government control over uh everyday people.
00:13:20.559 --> 00:13:22.879
To me, it seems like AI is dangerous.
00:13:23.039 --> 00:13:32.000
You know, it's it's scary whether we're thinking about the data centers or the flot cameras or all this talk about runaway AI agents.
00:13:32.559 --> 00:13:36.960
Whoever you talk to or whoever we talk to seems to be concerned about AI.
00:13:37.200 --> 00:13:43.039
But how can we understand how AI is dangerous and what the potential harms are?
00:13:43.360 --> 00:13:45.919
So I like to think of this question in sort of two parts.
00:13:46.080 --> 00:13:49.759
You know, the real question that we're trying to answer is why is AI dangerous?
00:13:49.919 --> 00:13:58.639
Is it dangerous because nobody can control AI and it's going to eventually spiral out of control and release, uh, you know, lead to human extinction?
00:13:58.799 --> 00:14:20.480
Or is it dangerous because of the world that we live in and because our society is structured such that you know companies have this profit motive to release AI that they know has serious defects, that they are not able to create any safety protocols for, that they know is you know uh are using massive amounts of resources and being used for surveillance and weapons manufacturing and so on.
00:14:20.720 --> 00:14:22.799
And I think it's firmly that latter case.
00:14:22.879 --> 00:14:35.840
Um the reason why AI is dangerous for us is uh because of the this sort of capitalist world that we live in and the profit motive I think that companies are pursuing so that they each can pursue short-term gains while the rest of us suffer for it.
00:14:36.000 --> 00:14:40.240
And we know that from observing what the actual harms of AI are today.
00:14:40.480 --> 00:14:50.240
It's not necessarily uh we haven't seen AI be uh being used by a nefarious actor in order to pursue some kind of existential human uh project of human extinction, right?
00:14:50.399 --> 00:14:53.440
We haven't seen anybody try to release some kind of new crazy virus.
00:14:53.679 --> 00:15:01.840
But we have seen AI do, just as we've seen uh uh people use technology for over the past uh 40 years, are for these projects of surveillance.
00:15:01.919 --> 00:15:13.039
Um, I'm thinking about cases where you know cops, uh, there's what there was one Texas cop who, in search of a woman who uh was suspected of an abortion, searched over 83,000 flaw cameras.
00:15:13.200 --> 00:15:20.240
And so that's the kind of uh thing that is you need technology uh like AI in order to do to manage data at that scale.
00:15:20.399 --> 00:15:35.039
And yet it's not something that is at the level of human extinction, it's the way it is, and because there are there's limited police accountability in this country because our power structures are in such a way that that we cannot hold anybody to account, either government officials or the tech elite.
00:15:35.600 --> 00:15:43.200
Yeah, I mean, it seems like one of the things that keeps coming up and that we talk about on the hum is that this is a fight for control.
00:15:43.519 --> 00:15:50.159
I mean, it is about the data centers because people hate data centers, and that's valid because they're kind of a horrible deal.
00:15:50.320 --> 00:15:53.919
But this overall fight is like, well, the data centers don't have any rules.
00:15:54.080 --> 00:15:55.279
The AI has no rules.
00:15:55.440 --> 00:16:09.600
If we actually had meaningful regulations that were about uh protecting water, protecting air, making sure that regular people weren't paying the costs of companies making huge profits and weapons of war, then it'd be a very different conversation.
00:16:09.759 --> 00:16:19.919
But because of the lack of rules and regulations, it's making this a lot scarier and a lot more uh, you know, the build-out of AI and the build-out of data centers are both pretty out of control.
00:16:20.720 --> 00:16:23.600
Let's talk, let's let's zoom out for a second, Nathan.
00:16:23.679 --> 00:16:32.960
I mean, every day there's all kinds of articles in every newspaper about this AI economic bubble that's gonna burst.
00:16:33.360 --> 00:16:46.000
That, you know, these companies are leveraging money they don't have to pay for AI data centers, that they're borrowing and renting chips and space from each other with money that each other don't have.
00:16:46.159 --> 00:16:57.360
And because you know, of private equity and other sort of fancy money schemes, they're essentially not being honest or sharing how much money they're making, losing, or spending.
00:16:57.600 --> 00:17:02.879
And that could cause the economy to crash or certainly get very out of whack.
00:17:03.039 --> 00:17:09.839
Uh, how are you thinking about this bubble and what do you think is going to happen in the next few months or years?
00:17:10.240 --> 00:17:18.160
So I think that there's a complicated answer here, I think, because um there's sort of two truths that listeners need to hold in their head at the same time.
00:17:18.799 --> 00:17:21.039
One, there definitely is a bubble, right?
00:17:21.200 --> 00:17:24.000
I I don't think anybody, even in the AI industry, denies that.
00:17:24.160 --> 00:17:32.079
There's this collective understanding that, yeah, OpenAI is saying all sorts of very hyped-up claims about their model capabilities and so on.
00:17:32.240 --> 00:17:42.240
And therefore, that's a lot of the massive amounts of funding that open AI has required, because it depends on these kinds of overblown narratives, is in turn sort of detached from the underlying economics.
00:17:42.400 --> 00:17:48.319
Um there's also this understanding that um the extent to which that is true is is somewhat unknown.
00:17:48.400 --> 00:17:59.759
Um, and that is why the bubble has not collapsed yet necessarily, because the exact amount to which the this claim is overblown can can't actually be asserted, and nobody's ruling to really go on that on that bet.
00:18:00.720 --> 00:18:21.200
So what I mean is that you know, uh for a project like OpenAI Stargate, which supposedly is is will constitute uh you know over$100 billion in overall investment, it's not clear to investors uh exactly how how much that is going to fall short by in terms of the uh the actual uh uh income that OpenAI will make uh and Oracle will make into that deal.
00:18:21.759 --> 00:18:25.680
So that's one truth I think that listeners need to hold in their head.
00:18:25.759 --> 00:18:26.880
That there is a bubble, I guess.
00:18:26.960 --> 00:18:30.880
There is this amount of uh speculation, I think, that and instability within the market.
00:18:32.079 --> 00:18:50.799
But the other truth, I think, um, which is a little bit harder to digest, is that even as a bubble exists and keeps growing larger, these companies, uh especially the hyperscalers, these so-called uh hyperscalers at the center of this, meaning Amazon, Google slash Alphabet, or Microsoft, they are more profitable than ever before.
00:18:51.039 --> 00:19:01.279
And so even as they're spending and borrowing lots and lots and lots of money to try to finance their data center build out, and that is contributing to their instability, they're also raking in more money than ever.
00:19:01.680 --> 00:19:12.480
And so it's kind of a race between these two things to understand you know which is going to get to these companies first, um, you know, the revenue that will follow the hype or you know the consequences of so much borrowing.
00:19:13.440 --> 00:19:23.599
And the other part of that pill that I think can be a little hard to swallow is that you know, we've seen in this country um that it's hugely dependent on uh a small set of firms, right?
00:19:23.839 --> 00:19:30.559
Even just last year, AI CapEx uh capital expenditures contributed more to US GDP growth than all of consumer spending.
00:19:31.599 --> 00:19:35.519
So we have these like few points of failure in the US economy right now.
00:19:35.759 --> 00:19:57.759
This is a very similar dynamic that existed in 2008, for instance, where uh the US government had a very, very uh real motive to prop up some banks and some industries, because if those went down, then there would be a dominant domino effect where many, many things, uh lots of other companies and industries that were seemingly unrelated to the actual heart of the problem would go down as a result.
00:19:58.160 --> 00:20:20.400
And the government has this sort of too big to fail incentive to prop up, you know, whether that's by you know helping Goldman Sachs buy bear sterns, as happened in 2008, or in uh, you know, potentially in 2030, uh bailing out NVIDIA or providing a sort of a uh interest-free um loan to one of these companies that is that's going under so that this domino effect is stopped short.
00:20:20.559 --> 00:20:34.400
And so I think it's it it because of that, um, and especially because of our current administration and the sort of political climate right now, I think it would be a little naive of us to expect the bubble to uh to pop, I think, and and hurt the capitalists where it hurts most.
00:20:34.720 --> 00:20:46.559
Yeah, I mean it sounds like without some kind of policies or guardrails, the people who created the problem are not going to be held accountable, most likely in that scenario, which is really scary.
00:20:46.799 --> 00:20:50.000
My last question for you is about policy.
00:20:50.240 --> 00:20:55.599
What policy direction would you like to see us go uh based on these concerns that you're raising?
00:20:55.920 --> 00:21:00.880
So I like to break this up into maybe three different types of policy measures, right?
00:21:01.039 --> 00:21:15.680
And they're ordered by that just because of the simple fact that right now, a lot of the solutions that we can be tempted to go for and would be very effective in a functioning uh egalitarian society are really quite harmful in the actual world that we live in.
00:21:15.839 --> 00:21:27.359
So the main example I'm thinking of here is that uh of what we should not do is something like uh you know Bernie Sanders' proposal of just you know, right now today, nationalizing 50% of AI companies.
00:21:28.000 --> 00:21:38.720
It would have a number of harmful effects and backfire on us, um, in part because of what AI under this particular society is structured to do, labor displacements and uh depressing wages and so on.
00:21:38.799 --> 00:21:54.079
Um and of course, not to mention all the environmental harms, but also because, of course, um if we were to nationalize AI the AI industry, you know, under the current administration, I think that that that would probably worsen the problem of AI development and data center development rather than allow greater public control.
00:21:54.400 --> 00:21:59.680
So you're worried about public control because it would sort of tie the fate and economics of the U.S.
00:22:00.160 --> 00:22:04.160
Government to a lot of the problems and instability of the AI industry.
00:22:04.480 --> 00:22:05.119
Is that right?
00:22:05.440 --> 00:22:11.839
Yes, and because the uh so-called public control right now of these companies is not actually public.
00:22:12.000 --> 00:22:13.440
It's in fact state control.
00:22:13.680 --> 00:22:29.039
And given the Trump administration and his control over the state and it's very, very large interest in seeing data centers proliferate, uh a stake in these companies would help the Trump administration uh accelerate the data center development instead of giving the public some levers to stop.
00:22:29.519 --> 00:22:40.079
So it you know, in response to that, I think this kind of bind where a lot of these seemingly best measures that we have, you know, are are sort of doomed to fail because of the society we live in, I think.
00:22:40.400 --> 00:22:48.960
We have to have an approach that is both addressing the problems that we face today and preparing for the society that we do actually live in that will allow us to take these more radical actions.
00:22:49.599 --> 00:22:59.440
So, you know, today I think every action, every policy measure we do has to be designed with the mind of increasing our own power over these tech companies, right?
00:22:59.759 --> 00:23:18.000
So I'm thinking about a nationwide data center moratorium, which would be useful to you know give people time to recruit, time to think of these broader, more capacious strategies, but also certain policy measures to ensure that the these movements which have been very effective in stopping data centers can be protected and can proliferate.
00:23:18.160 --> 00:23:23.920
So right now we have uh an administration which is very hostile to activism of all kinds, right?
00:23:24.000 --> 00:23:28.960
It's uh they've called data center um activists terrorists funded by China and so on.
00:23:29.200 --> 00:23:32.480
They're pursuing felony charges against uh anti-ICE protesters.
00:23:32.559 --> 00:23:37.680
Um so I'm thinking about the Minnesota 15 and the Prairie Land uh defendants, um, as well as the Michigan 8.
00:23:38.000 --> 00:23:51.039
We need some kind of measure of free speech um and you know, blanket protections for the for activists so that all these tactics that we do have to stop data centers can be made more successful, can be encouraged, can be, can proliferate.
00:23:51.440 --> 00:23:57.119
Other measures in that regard might include things like lowering the bar for a ballot initiative to sort of get on the table.
00:23:57.519 --> 00:24:08.480
Right now, a lot of uh counties or municipalities have you know eight to fifteen percent of uh you know the previous elections voters need to put a signature down in order for for a measure just to get on the ballot.
00:24:08.720 --> 00:24:14.240
But that is often an insurmountably large um burden, I think, for many, for an actual democratic process.
00:24:14.400 --> 00:24:26.079
It requires tens and thousands of dollars of funding in most cases and prevents citizens uh who are under-resourced, uh, which uh which, of course, uh these activists find data centers are likely to be given where data centers are built.
00:24:26.240 --> 00:24:30.720
Uh so it poses these very large obstacles, I think, for democratic governance.
00:24:31.039 --> 00:24:31.839
Awesome.
00:24:32.160 --> 00:24:37.920
Nathan Cam is a PhD candidate at the University of Michigan School of Information.
00:24:38.079 --> 00:24:44.720
He's been organizing in the fight against the Los Alamos Data Center being built at the University of Michigan.
00:24:44.960 --> 00:24:47.680
Thank you so much, Nathan, for joining us on the Hump today.
00:24:47.920 --> 00:24:48.720
Thanks all.
00:24:52.079 --> 00:24:57.039
That was my conversation with University of Michigan graduate student Nathan Kim.
00:24:57.279 --> 00:25:06.880
People all over the country are copying Michigan students' strategies for revealing that AI data centers are being built for war and profit, not local benefits.
00:25:07.200 --> 00:25:09.200
You're listening to The Hum.
00:25:09.519 --> 00:25:17.839
We're publishing new episodes weekly, so if you enjoyed what you heard, give us an enthusiastic five-star review that's at least five words long.
00:25:18.079 --> 00:25:24.079
Message us on Instagram at TheHumPod to get connected to organizers in your state or country.
00:25:24.319 --> 00:25:27.599
The Hum is produced in partnership with Rohome Productions.
00:25:27.839 --> 00:25:30.480
Rohome's creative director is Alex Lewis.
00:25:30.799 --> 00:25:33.039
Their executive producer is John Myers.
00:25:33.440 --> 00:25:35.680
Our producer is Emily Rizzo.
00:25:35.839 --> 00:25:41.599
The Hum's video producer is Adultrehan and we partner with the Center for Nonviolent Conflict Research.
00:25:41.839 --> 00:25:44.960
I'm Stol Levin, your host and banjo journeyman.
00:25:45.119 --> 00:25:48.160
Thanks for listening and see you next Tuesday.
