Stephen Miller Is Finally Getting the Crackdown He Wanted (w/ David Wallace-Wells) | Bulwark Daily

Because he's not obviously sociopathic. Like, that makes me distrust him. [laughter] It's like, it's like all you other guys I can tell are evil. What's happening with this one?

Tim, great to be here.

I mean, at the moment, I think the politics of AI is the most exciting and dramatic political story of the summer. I mean, I would say there was polling that just came out today, I think, or yesterday from Heat Map showing that now 75% of Americans oppose the construction of data centers in their communities. Something like 60-plus percent oppose them strongly. That is more than would oppose the construction of a nuclear power plant. And if you think back to what the opposition to nuclear power did to the, you know, buildout of nuclear power decades ago, it looks like we're heading for an incredible collision course here because we're building out data centers, trying to construct a massive compute empire at breakneck speed, and this opposition is rising, growing really, really quickly. And that's been happening for, you know, 9 months, 12 months, depending on how you want to clock it. But the really significant thing that's happened over the last couple of weeks is that a lot of political leaders are aligning themselves with the opposition here. And Kathy Hochul in New York was the first one, but you know, Greg Abbott in Texas of all places has now signed a kind of data center moratorium. We're seeing, you know, Josh Shapiro, who is basically campaigning for president as a like, I'm going to build stuff guy, 'Get shit done' as his motto, he's imposing a data center moratorium. And so you see across the political spectrum politicians like, you know, getting a read of which way the wind is blowing and putting themselves in the camp of data center opposition. And this is, you know, completely novel as far as I'm concerned, you know, signing up with the populists against the sort of economic interests of states and communities. And I think the sort of decision there, the sort of contrast there, the showcase there is really, really powerful when you dig into it. Just a few brief little case studies. In Loudoun County, Virginia, they're now making half of their tax revenue from data centers. They've cut property taxes by 30% because of the revenue coming in from data centers. In, um, in down in Louisiana, where you are, Richland Parish, I think it's called, um, they've given elementary school teachers $50,000 bonuses. These are women and men who make less than $50,000 a year, all of them. They got bonuses because of the construction of the new Meta data center there. And you see similar stories all around the country where you have this incredible flow of money coming into local communities when data centers are built there. And yet the local populations are unwilling to accept those projects even with that money promised to them. And increasingly you now see state and national leaders signing up not with the corporations who want to do that work but with the people who are opposed to them. And this is, I think, just a really big deal, a really new pattern in American politics and a sign of where populism is in the summer of 2026 that we're now seeing people like Greg Abbott or Josh Shapiro, you know, on the protest side of the equation.

I mean, you would think that that would be an absolute golden ticket appeal in American politics. You can pay less taxes and your local government will be much better funded. They'll be able to provide more social services. You would think that would be the kind of appeal that, you know, pitchmen and politicians would dream of. And yet...

I see it in roughly those terms, too. It's a sort of a catch-all objection to people feeling out of control and powerless, feeling that their lives, their communities, their futures, their economic prospects are all being determined by other people, not by them. And they've been feeling this way for a while. They've expressed that frustration in a variety of ways. Um, but the AI buildout is a sort of useful opportunity to express all of those frustrations. And people bring different things to it. I mean, you mentioned a lot of the particular objections. There's also old-school NIMBYism. There's old-school environmentalism. There's new-school environmentalism focused on the climate impacts. There are people who are worried about electricity prices. There are people who are worried about the, you know, the fact that these data centers might shadow their whole small town and change the physical landscape of their environment. There are people who are worried about AI being used for surveillance. There are people who are worried about unemployment, job loss, even existential risk. It's such a big messy bucket. It's such a big messy coalition that I think you're right. We shouldn't like think about it piecemeal. When you think about it piecemeal, you almost get to a place where you say, 'Well, there could be some bargain struck here. Um, we should be able to persuade people. We should be able to make sure the sound isn't as loud. We can make sure that the electricity prices are actually going down rather than going up.' There are a lot of things that can be sold to local populations. But when you have 75% of the country saying that they are opposed to this, I mean, this is just an astronomical number. It is not the kind of number that we see in American politics about anything. Three-quarters of the country opposed to data center buildout. I mean, three-quarters of the country. Those numbers, they're just almost too big to imagine. A lot of people associate it. They think in their minds, oh, this is NIMBYism or it's opposition to renewables. People support the construction of renewable energy by 65 to 30...

Right? People oppose the construction of data centers by 75 to 20. This is a supermajority coalition larger than anything we have seen in American politics in a generation. And I think some people, their opposition is thin. Some people, it is informed by misinformation, disinformation. I'm not guaranteeing that this resistance will succeed and stall the data center buildout. There are a lot of forces moving in the other direction, but it is striking. That number is so large that you can lose quite a lot of people from that coalition and still have an incredible command of political power. And I think that's where we're heading because every time you see a new poll, the number in opposition goes up. It's not going down.

Because he's not obviously sociopathic, that makes me distrust him. [laughter] It's like, it's like all you other guys I can tell are evil. What's happening with this one? And I think it's tied up in this broader story which is that, you know, the last 20 years of American life, people have had material gains. They've been sort of unequally distributed, but by many measures, Americans are doing better than they used to. And yet, we've had this incredible growing unhappiness and distrust with our society. And that's especially true when you ask those questions about the internet and social media which seemed 10 or 15 years ago to be generally speaking forces for good which most Americans appreciated and now feel deeply corrosive. In fact, a lot of people suspect that they kind of ruined a generation and maybe ruining the entire human race. I'm not a doomer on those points. I think a lot of those arguments are overstated. But Americans feel in some deep way that their experience of the world is getting worse. That is in part because more and more control and power is being exhibited by a small number of people, many of them in Silicon Valley and many of them centrally positioned now to be determining the shape of our collective future. And when people look at that possibility, they don't think, 'Here are generals I want to follow into battle. Here are leaders who I want to emulate.' They think these people's vision of the future is completely distasteful, in certain ways anti-human, it's certainly anti-worker. Yes. And we just want to reject that out of hand. And you know what's striking to me most about this phenomenon is that if you had spoken to me about this six or nine months ago, I would have outlined that story in almost exactly the same way. But I also would have said one of the deeply unhappy things about this moment is that so many people feel out of control and powerless to do anything about it. And that even if there was this incredible upwelling of resistance, probably it would be accompanied by the sense of fatalism, like, what are we going to do? These people are so rich, they've already bought the land. They've already bought the politicians, etc., etc., etc. And so it's really amazing to see a huge movement, something equivalent, I think, to Occupy or the Tea Party springing up in a time of even more fatalism and despair about our politics. It's a sign that people are maybe feeling frustrated, maybe feeling alienated, but responding to that not by retreating into despair, but trying to fight back. Perhaps in some irrational ways, perhaps in some ways that'll burn out relatively quickly, but as an expression of broad American distrust of the powers that be, I don't think that we've seen anything this strong and this wide in many years.

I mean, I come back to the example I cited at the beginning. I mean, in Richland Parish, Louisiana, elementary school teachers who have annual salaries of like $30 to $40,000 got a $50,000 bonus. And still that community is not happy.

Yeah. I mean, I think this Hugging Face incident woke some people up. So, there's been a new little wave of alarm. The people who are working very seriously on AI safety found an opportunity to weigh in again and show their faces in public. But in general, when we first encountered ChatGPT in 2022, 2023, people were really worried. It felt like an existential challenge. It felt like, you know, maybe it wasn't that we were on the brink of Skynet or human extinction, but there was something really deeply profound about the world that we were about to live in. And people worried about biorisk and they worried about terrorism and they worried about, you know, yes, the effects of like mass unemployment and, you know, complete transformation of the economy. And all of those worries are in some ways still present in the zeitgeist, in the American psyche. But, you know, when you look at the front page of the New York Times or you watch cable news or whatever, people are much more likely to be talking about things like, is it possible that Anthropic can earn back its capex so that its investors will be made whole? And those are important questions. If they can't, it does mean something significant for the medium-term of the American economy since this country has bet so much money on artificial intelligence that if it doesn't work out, not just in general, but if it doesn't work out in precisely the pattern that the bets were made, suggesting we're in for a kind of rocky road ahead. So those, I don't want to diminish those things, but it's not the same thing as saying, you know, this antisocial outcast at the local high school can now design a superbug that will wipe out the entire town or possibly the entire world, which is the kind of worry that people were having three or four and five years ago. And to some degree, that change is because the people monitoring safety have basically thought that things are going a little bit better than they worried. And yet every few months, in some cases every few weeks, you have one of these incidents where something quite alarming happened. So in this case, they were doing some training runs, the models were not connected to the internet. They were tasked with finding some cyber vulnerabilities. And that's a kind of an open-ended task. And so they found their way onto the internet even though they weren't technically connected to the internet originally. They weren't, I don't think, disobeying any explicit orders. So it wasn't as though they'd been told not to get to the internet. It seems like they were trying to follow the order in the most extreme way possible. It's like because they weren't on the internet, the model could not identify the bugs that it was supposed to identify. And so anyway, that seems like what happened. But it is still, I think it's still important to keep in mind with all of these sorts of stories, like these models that we're freaking out about the same as we did a year or so ago when we heard that they were blackmailing executives or, you know, that sort of thing. These were models that are in training runs being tasked with particular tasks. And I think on some big profound meta kind of a level, we have to keep in mind that there is this distinction. We should keep in mind this distinction between what models are doing on their own and what they are doing in response to things that we have asked them to do. And when we talk about them, at least in this moment in 2026 with the models, with the capabilities that they have now, when we talk about them as totally autonomous, as though they are seeking out their own goals, I think we're letting ourselves off the hook, not just in the context of the training environment, but more generally. Everything that AI is doing now, humans asked it to do. I think that is likely to be the case forever after. But at least for now we should be really, really careful in attributing true agency to AI agents. They are AI agents because humans asked them to undertake a task, not because they are coming to decisions on their own about ultimate goals. And so, you know, this was nerve-wracking for cybersecurity people. It's nerve-wracking to some extent about people who are worrying about, you know, what AIs can do. But I think at a base level, we should remind ourselves as often as we can that when we're worrying about disasters that might strike from artificial intelligence, we are still overwhelmingly today thinking about disasters that humans will task AIs with undertaking.

The way that I think about it is that we've been told a very particular story over the last four or five years about AI as a business proposition, which is to say we looked at these charts that had AI capabilities growing quite quickly. We saw plotting of future AI development beyond the present into the future and we saw that at some point those charts started to look pretty vertical, kind of exponential. And the basic case that was made to us by the frontier labs but also by, you know, to their investors, to the American public, was that at a certain point artificial intelligence would be so productive that the companies who had moved fastest and made the most progress would be just drowning in profits from how transformative the tools that were coming out of their labs would be. And that meant that the race was really, really important. And the race was important between Anthropic and OpenAI and other players who have sort of since fallen behind. And it was really important between the US and China because getting past that point of true exponential growth, sometimes called recursive self-improvement, sometimes called artificial superintelligence. Um, people... Yeah. AGI, AI, AGI, ASI, whatever. You would just be in an entirely different world on the other side of that boundary as you are today. And I just think for a number of reasons that is starting to seem at best questionable. And we are seeing remarkable improvements in many capabilities of artificial intelligence. We are also seeing relatively slow effects on productivity, relatively slow effects on economic growth. Some of that is about the slow pace of diffusion through the economy, but a lot of it is because the truth is we haven't figured out how to make these things super useful. And there are probably many things that it can't be all that useful in doing at least for the time being. Another big part of that challenge is that, you know, if it is the case, if it needs to be the case that, you know, a frontier lab passes that threshold and is on that exponential curve for them to really, really profit and earn back the investment that has been made in their capabilities, then the fact that there's a whole sea of alternative models, mostly in China but not exclusively, that can do like 90, 95% as well as the true frontier models at 5%, 20%, 30% the cost, that's a big challenge to the basic business case because there may still be some, you know, opportunities and there may be some applications that require the absolute best in class, but most of the world, and not just me and you writing emails with it or doing research with it, but businesses trying to do analysis with it, they don't necessarily need the absolute first-class version if they can get something that's almost as good for a quarter of the price. I mean, I think about the way that investment banks work with their research analysts who are just out of college. They pay them a fraction of what they pay their senior people. They work them like crazy hours, so their brains aren't completely functioning, but most of the work they're doing is just churning through spreadsheets. AI can do that. Maybe a better AI can do it a little bit better, but if you could find one that does it almost as well for a fraction of the price, many people are going to go for that. And that leaves the businesses that have been built on the promise of AI monopoly kind of in the lurch. And that is really the abyss that Anthropic and OpenAI are staring down right now where they've even started openly acknowledging that they're more focused on price competition with one another and with open source models than they are in terms of improving the frontier of, you know, the actual capabilities. And I think it's, you know, it's notable that over the last few months, they've made that acknowledgement. A number of tech leaders have essentially started to welcome open source into their lives because it makes startups easier. It sort of circulates, you know, is easier to use.

Let's do ICE. I just read about it.

Yeah, it's probably going to set a new record in August. Um, deportations are likely to set a new record this year. Um, certainly deportations of people seized within the border are going to be way above what they've ever been before. Um, and this comes in the wake of this confrontation in Minneapolis, which was sort of observed by the country in real time, um, and seemed to me and I think a lot of people I know to mark a real shift in the mood of the country about immigration and about Trumpism. It looked like the citizens of Minneapolis stood their ground against an invading...

horde of federal agents and there was conflict. There were a few deaths. There was a lot of protest activity and a lot of media attention. And then the Trump administration retreated with its tail between its legs and Greg Bevino was forced out and yeah, Christine M was was forced out as well.

But you know, Steven Miller is still in the White House and he's still maybe the second most powerful man in the country. And what happened, as I understand it, is that there was a kind of a strategic conflict going back a couple of years between people who really wanted to like stage spectacular confrontations with immigrants.

They wanted the ice porn. They wanted the ASMR videos with the with the immigrants and chains and they wanted to to rub it in their face. And as as part I think as a strategic point of hoping that if they did that, people would self-deport and leave. And also in part because there's some sadists in in the White House.

But there were also people including Miller who was like, you know, we can do this in a much more strategic way. The ultimate goal is actually getting these people out of the country. The standard they set for themselves on a campaign trail was 20 million immigrants deported. And we're nowhere near that. So we have a long, you know, we have a long way to go to the the absolute worst case scenario. But if you were horrified by the brutality that we saw on our phones last year, in many ways things are worse today because they are in order to keep up with their quotas and hit their numbers, they're actually targeting immigrants who are complying with immigration law.