You can find the Palisade Research podcast on Spotify and Apple Podcasts, or anywhere else you get your podcasts.
Dara Lind is a senior fellow at the American Immigration Council. She has reported extensively on immigration policy at ProPublica and Vox, co-hosted Vox’s podcast The Weeds for many years, and has covered immigration in some form since the end of the George W. Bush administration.
Dave Kasten, Palisade’s Head of Policy, talks with Dara about how immigration went from a quiet issue in 2009 to one that helped elect a president: why its salience comes in event-driven spikes, how the post-9/11 split of the old INS into enforcement and benefits agencies shaped their institutional cultures, what Schedule F changes, and what happens when courts stop presuming the executive branch is acting in good faith.
They then turn to what this means for AI policy: the trade-off between salience and bipartisanship, how one political entrepreneur can polarize an issue, DC’s ads aimed at an audience of ten, why “warning shots” is the wrong metaphor, how to keep a field from burning out its people, what reporters actually control in a newsroom, and how much of the law assumes a human being is making the decision. Dara closes with advice for her younger self about the contingency of history.
Transcript lightly edited for readability.
Dave: Hi, and welcome to the Palisade Podcast. I’m Dave Kasten, Head of Policy at Palisade Research. We’re joined today by my very good friend, Dara Lind. Dara is a senior fellow at the American Immigration Council. She has extensive reporting experience covering immigration policy at ProPublica and Vox. She also co-hosted the Vox podcast “The Weeds” for many years. She’s covered immigration in some form since the end of the George W. Bush administration. Without further ado, let’s jump into it.
You and I have talked about immigration for many, many years. There was a time when immigration wasn’t a big deal in DC. It was a quiet issue in politics.
Dara: It’s so funny that you think about it this way, by the way, because I graduated from college in 2009, and I moved to DC for a couple of reasons, one of which was that it was 2009 and nobody else was hiring. But the other was that Barack Obama was going to pass comprehensive immigration reform. He’d promised to do it on the campaign trail, and everyone was very, very excited that it was going to be a priority in the new administration. So I came to DC thinking, “Oh, I really want to be here when this major bill gets passed.” I got hired by an organization that actually brought me on saying, “We may not have you for more than a year, because we may not exist for more than another year. We just started a couple of years ago, but Barack Obama’s going to pass comprehensive immigration reform, and then we’re going to dissolve.” That organization, of course, still exists today, and does not have any of the same staff as it did when I was there. And Barack Obama did not, in fact, spoiler alert, pass comprehensive immigration reform.
But I came to a DC where the first thing I realized was that most people were more like you than like me. Most people did not see immigration as a huge policy priority, and I actually spent the first six months or so that I was here having fights at parties with people who thought that after healthcare, it should be financial regulation, or after healthcare, it should be climate change. There was this idea that there were a limited number of slots on the docket, and that whatever came next after healthcare, which was, of course, being debated at the time, was going to be the only other big bill that got through.
And it was pretty obvious that immigration was losing that fight. But immigration has always had this weird disconnect between how much everyone agrees change is needed to existing policy (and everyone agrees a lot of change is needed) and how much actual urgency there is among the people who are in a position to change things. So in 2008, 2009, everyone agreed that it was not only important, but important enough that you should actually get Congress to do it, but maybe not as important as other things that were going to take Congress the whole two years to do.
Dave: And as someone who was with you at some of those parties in 2009, I can testify that this was an ongoing thing, not only on your issue, although that’s certainly true, but also on a lot of other issues. There was this notion of, “Hey, we’ve got this trifecta. We’ve got two years.” I don’t think people expected as much of a washout as the Democrats actually ended up having in 2010, but I think everyone understood that American politics is thermostatic: there’s always a reaction back in the other direction, and they were assuming there might be one. So there was very much this ongoing thing where everyone would fight at parties. We were all 22 and 24, so…
Dara: To be clear, literally no one who was having these arguments, at least in the rooms that I was in, had any influence whatsoever. It was just that we understood that that was the important fight to have, and many of us who were working on one issue or another felt a lot of existential commitment to, “No, obviously this is the most urgent thing,” and a lot of frustration at not getting other relatively high-information people to see that.
Dave: So I want to cover a couple of things in this conversation. One thing I want to cover is: what is it like to be part of an issue like immigration that goes from being important to a lot of people, but pretty low salience in terms of the political discussion and in terms of what Americans said were their top priorities, to something that’s very high priority? I want to talk about the ways in which you, as a policy advocate and journalist, have done a bunch of things, and the ways that you’ve moved from one to the other, and what people understand your job as being, and then what it actually is, and where some of those disconnects happen, and how that could cause issues. And I want to talk a little bit about what lessons we can learn from all of that for AI policy.
I also imagine that along the way, we’ll inflict on our listeners some of the very friendly brain damage that anyone who begins to try to understand US immigration policy feels. Apologies in advance, folks. It’s going to be a wild one. But I really want to get into the details. So maybe, as scene setting, you can tell us, in very wave-skimming fashion: how is it that immigration policy went from that period of 2008 to 2010, where it was not that high a priority for the Democratic Party, to being one of the major issues, to the point where I think, pretty inarguably, it was something that helped elect someone president, at least once, arguably perhaps twice?
Dara: A thing that I say all the time (which is probably going to be the way I start a lot of sentences during this episode, so apologies to people who talk to me all the time) is that immigration is always a culture war issue, and it is also always a very complex policy issue. And the relationship between those two is, I would say, attenuated at best. So we have a world right now where you can say that immigration was among the more consequential issues of the 2024 election. It didn’t have Congress going into 2025 thinking that it needed to legislate it. And there has been a tremendous amount of change in executive branch policy; I certainly don’t mean to understate that in the least. But it’s also not apparent that a rank-and-file, low-information voter who voted for Trump in 2024, and would have named immigration among their reasons, necessarily understands that there’s a major difference, other than that they voted for the guy, so the guy’s going to do the things that he promised to do. It’s not like people are committed to continuing to follow the issue.
My macro theory of salience on immigration is that it’s very event-driven. If you look at “what is the most important issue to you” polling, you usually see that it’s pretty low, and then there are spikes for a month when something big happens, and then it’ll decrease again. That’s become less true among conservatives and Republicans as it’s become a more important element of Trumpist political identity. But it usually comes in the form of exogenous shocks that the government then has to respond to, or shocks to the system that the government itself inflicts. When Trump tried to end the Deferred Action for Childhood Arrivals program in 2017, for example, all of a sudden immigration became super salient for a second, because a lot of political elites were running around like chickens with their heads cut off: “Oh, this is a big problem. We’re going to have to fix it.”
Other than that, it tends to be stuff that’s visually mediated. And this, I think, is a really big difference between the issue that I work on and the issue that you work on: there are always sob stories in immigration. There are always very obvious human victims, and there are also very obvious, powerful visuals of threat. Usually, when you have mobilization on the right, it’s being driven by something that is visually powerful or metaphorically powerful that’s about a mass of people. The caravans are a great example of this. It’s amazing.
Dave: The big swaths of people perceived as walking towards the border in order to get some sort of immigration status and to enter the country.
Dara: Whereas when the left is mobilized, it’s usually crying children or dead children, to be honest with you: a family that has been separated, a child that has been put through an incredibly dangerous journey, that kind of thing. It’s a sugar rush. There’s very, very short-term interest that’s very powerful, that will totally dominate the headlines, and that doesn’t in itself do anything to generate longer interest in even learning more about how this came to be, much less more durable change.
Dave: One of the things that I’ve learned the most from our conversations in the past is that I, like many Americans, had this high-level, vague understanding of immigration policy as, naively, there is a thing called immigration policy. There is a line that you can get into, and you’re waiting to get to the front of the line. Then when you do, congratulations, you’re in America. And if you’re not in that line, well, there’s maybe a couple of things: if you’re someone seeking to get asylum because you’re fleeing a war, maybe there’s a thing for that, I don’t know. That was my vague level of understanding before I started learning from you on this. And very little of that is in any way accurate when you get underneath the abstraction layer. I don’t want to belabor our entire conversation by going into that, because, as you know, we could burn infinity many hours on that. But could you give our listeners some sense of why the high-level way that Americans think about immigration, and about what a fair process for immigration is, isn’t at all how it actually works under the hood?
Dara: It’s really important to remember that the people who have firsthand experience with American immigration policy in real time definitionally cannot vote. At best, you have a US citizen spouse of someone who is trying to get their green card and ultimately citizenship. If you yourself are a naturalized citizen, then yes, you’ve been through the system. But frankly, naturalization has historically been the thing that the government puts the most effort into keeping from being broken. Even when everything else is absolutely on fire, naturalization will keep running. So you certainly don’t have a critical mass of people who are seeing the bill come in the mail. They’re not dealing with insurance denials, that sort of thing. And so it’s very difficult to get from “everyone who knows about this understands that it’s not working” to making it work.
Two other structural constraints that I would note. One is the way that the executive branch agencies are broken down. After 9/11, the Immigration and Naturalization Service got broken into sub-components, which ultimately became two enforcement agencies: Customs and Border Protection, which is doing enforcement at the border, customs inspection, managing ports of entry, that sort of thing; and ICE, Immigration and Customs Enforcement, which is the low-level detention and removal operations, plus some special agent types. Those two, Congress is more than willing to throw money at. Congress has given hundreds of billions of dollars to CBP and ICE in reconciliation bills over the last two years. It’s ludicrous.
Meanwhile, there is the agency US Citizenship and Immigration Services, which is responsible for processing immigration benefits: the things we think of as legal immigration, but also asylum, and also some of these things that aren’t really immigration status that I’m going to get into in a second. That is theoretically fee-funded. The immigrants themselves are supposed to be paying for it. Congress will occasionally toss a few bucks to USCIS when the backlogs get really, really bad, but in general, it’s expected that that agency is going to run along on its own. And so when there are huge shifts in, say, a lot more people applying for asylum, which doesn’t have an application fee, that’s an unfunded workload. And there’s no structural way to fix that.
The other structural point is about the balance of powers between branches. Congress theoretically is responsible for establishing a uniform law of naturalization. The last time that Congress made major changes to the legal immigration system was 1990. In other countries, there’s a lot more executive branch flexibility here. In Canada, for example, the prime minister’s government gets to say, “Here’s our target for how many refugees we’re taking this year. And yeah, sorry, we’re going to have to take a few away from the stream for children of residents, or the stream for doctors,” that kind of thing. The executive branch does not at all have that power in the United States.
So as these enforcement agencies got more built up, and as it became possible to actually apprehend and remove people who had been in the United States for a long time (and I don’t need to get into this here, but just stipulate that over the 2000s, it also became true that people who were here without papers had been here for longer; it was a more settled population), all of a sudden the question became, “Well, do we really want to just have these people going after everybody? Aren’t there some cases, like people who have been here since they were children, like parents of US citizens? Aren’t there some cases where, yeah, you don’t have papers, but we’d really rather not deport you?” And so this is where we’ve seen this executive branch pendulum over the last few administrations, with Democratic administrations trying to extend what protections they can, which are still short of full legal status, and then consecutive Trump administrations coming in and doing their best to strip those protections from as many people as possible.
Dave: One thing that particularly strikes me as you talk about this is that there are issue areas where the lines of authority and the lines of responsibility are relatively clean. And I notice it myself as I’m listening to you. Normally, DC is a town that runs on acronyms. We just say the acronyms. You and I have both been in meetings where someone says an acronym and someone asks, “What does that acronym stand for?” And it’s like, “I don’t actually know, but I know they’re the people that do this thing.” This is a very common DC experience. But when you actually spell out what CBP and ICE and USCIS do, they literally have overlapping words in their names. Part of why it is so gosh darn confusing to understand what these different orgs do, I imagine, is that all the way from the very beginning of their restructuring post-9/11, it seems like it’s not especially crisp in terms of who’s doing what, and what the vision is of how they’re supposed to work together to achieve a goal. It seems like, in some sense, maybe overstating it a little bit, there isn’t per se a goal. There’s just a bunch of things a bunch of orgs are doing.
Dara: We could get super into the weeds on this and talk about how it actually took several years after 9/11 to get the structure of the two enforcement agencies that we have, but we don’t need to get there. What I do want to point out is that institutions are made up of people. Existing institutional cultures can be very durable, but huge shifts in personnel can shift institutional cultures. INS, the Immigration and Naturalization Service, the old 20th-century immigration agency, was never seen as a bunch of bleeding-heart liberals. It was seen as a bunch of paper-pushing bureaucrats at one end of the spectrum, or people who were really interested in being stingy toward immigrants on the other.
When you split that out into enforcement agencies and non-enforcement agencies, you still do have plenty of the people who are just very down-the-line bureaucrats remaining at US Citizenship and Immigration Services. But especially as asylum becomes more of the workload, the relevant institutional culture of that agency becomes the culture of the asylum corps, who are the sort of people who are drawn to enforce human rights law for a living. That’s a very, very, very different culture, in terms of self-selection, from the people who were getting hired by Border Patrol in the 2000s, when they had all of this money and suddenly a totally new budget line, or getting hired by ICE on the assumption that they were going to round up immigrants.
And you’ve seen those institutional cultures, and the relationship that creates between rank and file and management, really shape the last several decades of immigration policy. The first time that the Obama administration tried to issue a memo saying, “Hey, maybe don’t prioritize deporting families of US citizens. Maybe if you see somebody in the field, let ‘em go,” the ICE Officers Union held a vote of no confidence in their leadership, which is a really stark illustration. And the Border Patrol Union, for its part, was a very early validator of the Trump campaign the first time, in 2016. So the ways in which those institutional cultures are obviously shaping what those dudes are doing on the ground, but also how they become players in how the government responds to things, aren’t something people are paying attention to in real time. And then there’s a lot of, “Wait, how did we get here? How did we get to a point where the Asylum Officers Union is routinely filing motions against its own administration’s policies?”
Dave: I didn’t know that. That’s fascinating.
Dara: I was going to say, that makes it sound like a Hatch Act violation. Because it’s the union, it’s not. It’s being filed by union officers, not by the federal government.
Dave: Right, in much the same way, from the other end, ICE and CBP are unionized. For whatever reason, in my own filter bubble, I’d heard of some of the ICE union stuff and the CBP stuff, but hadn’t heard of the other one.
One thing that I think about a lot in the context of AI is the point you were making a moment earlier about how institutions have a lot of inertia, and they draw on a particular set of people, but if you really shift the number of people in there, you can shift the culture. I worry a lot about what ends up happening if we suddenly have to really plus up headcount on AI-related tasks, in terms of what kind of culture you suck in. And that could be good, by the way, to be clear. It could be, “Oh, we got exactly the right people. We suddenly hired all of them and built a perfect institutional culture that’s really oriented around the right stuff.” It also could be something like, “Oh, we just hired a bunch of Booz Allen contractors to fill a bunch of seats, and they’re punching a time card. They don’t care that much.” Or it could even be, to your point earlier, a culture that is actively trying to push against the policy of the given administration, which, setting aside the question of whether that’s good or bad (I don’t want to get into that per se here), certainly can have a different dynamic as a result.
Dara: I think in the 21st century we do tend to see more of this, as symbolic goods start taking up a bigger and bigger swath of people’s identities, and certainly their political identities: you’re going to have to assume that your workforce is going to be at least in large part mission-driven, and you need to figure out what that mission is. And, as we were just talking about, that changes the nature of collective bargaining, for example, because you’re not bargaining over bread-and-butter things. You’re bargaining in part over whether you are in fact aligned with the mission as your workforce understands it. That changes the ability to change policies, to change missions from the top. Have we talked about Schedule F on this podcast at all?
Dave: We have not.
Dara: Oh, great. Okay. So a really big difference between the first and second Trump administrations: think about the kind of people who are working in the federal government in, say, the Department of Interior, doing regulation of when people can be in national parks, for example, or doing OSHA, the Occupational Health and Safety Administration, doing work on workplace safety stuff. Those are people who are likely to be enthusiastic about regulation and enthusiastic about enforcing those laws. Which meant that when the first Trump administration came into office and didn’t necessarily have a plan for “do this other thing instead,” but at best was like, “Stop doing that,” and at worst was like, “Oh, we totally forgot this agency existed, so we don’t have our dudes in there,” a lot of those just kept going on autopilot. Or at worst, they didn’t have anything to do, but they weren’t being used to push policy in the direction the administration wanted.
The administration spent those four years really learning the ins and outs of the executive branch. For me, the biggest takeaway from the big Project 2025 document was how incredibly focused it was on which sub-agencies within DHS needed to be elevated and which ones needed to be pushed down so that they could go back to being under the control of the agencies. And it was very clearly a proxy for, “These people gave us trouble. We need to reduce their power.” They were always more focused on immigration than on a lot of other policy issues, but it does appear that the same kind of learning has happened throughout the federal government. And one of the most salient examples of this is that an executive order that they tried to pull off in the last year of the first Trump administration, they’re now actually implementing, which gives the—
Dave: Which is Schedule F, you said.
Dara: Yeah, which is Schedule F, which gives the administration a lot more authority to say, “If you are in a position where you’re making policy, you are, because of that, essentially a political appointee by proxy. You no longer have the full suite of civil service protections. We can fire you essentially for not being mission-aligned.” It allows them to change institutional culture much more quickly, and it also raises some questions: who are the people you put into an agency such as the Occupational Health and Safety Administration? Who are the people who really want to do that job, but want to do it from the perspective of deregulation?
Dave: And I think one of the things that I hope people are getting from this conversation, so I’m just going to say it very bluntly, is that in government, personnel is policy to a much greater degree than in the private sector, even more so than in the tech industry, where I’d say that, relatively speaking, personnel is more policy than it is in, say, agricultural businesses or whatever. Because the US government has set up a system where, by and large, you can’t easily fire most people, because they have civil service protections. We had the notion that we didn’t want people running around trying to shoot presidents because they wanted to be appointed ambassador to France, which is a real thing that happened, by the way, for those listeners of ours who haven’t watched the musical “Assassins.” So we tried to formalize these protections in order to create a non-political civil service. And not even non-political: a non-corrupt civil service. And to be very clear, I don’t mean this in a partisan sense. I mean a civil service that doesn’t take bribes, in a very literal sense. And we—
Dara: Right. Like the way that people think cynically about ambassadorships now: “Oh, of course, the European ambassadorships are going to go to big donors.” We used to think that way about the entire federal government. Postmaster positions: oh, everybody knows that the people you put in the postmaster positions are the people you want to help line their pockets for the next few years.
Dave: Yes, in a very literal sense. And at one point, by the way, postmaster was the most common job in the federal government, when the federal government was much smaller. Or outside of the military, at least. And so we have this system where we’ve created a lot of rigor and difficulty in changing the org chart and changing the list of who’s employed, in order to try to make a bunch of things run more or less on autopilot. Regardless of your political valence, I think everyone on all sides agrees there are upsides and downsides. But as a result, you end up in this situation where, because you can’t easily change the human beings that are in the office, at least not unless Schedule F gets fully implemented, all you can do is change who reports to who and how important different parts of that collection of human beings are to each other, by moving boxes around on the org chart. And this is a very different thing from the private sector, where you can just hire a lot of people or fire a lot of people or promote or demote a lot of people.
Dara: I think the other difference is that it’s not as if these are fallible and biased humans who are being asked to execute an infallible and unbiased system. There is discretion built into the law, and certainly into the enforcement of the law. And there are a lot of assumptions baked all through the way the federal government has traditionally run, that a federal agent could do a lot of things that it might not be worth their time to do.
Dave: Talk more about that, and how those dynamics emerge and how they have changed over time. In part because when we think about AI policy, a common thing, and I talked about this in our episode with our friend Matthew Lipka, is the one weird trick philosophy: the notion that you can find laws already on the books and try to make them do other things than what they might have been intended to do, and then you get to have a fight with the courts about whether or not that’s okay. But I think everyone would agree that right now, at least, there’s a lot of discretion and voluntary behavior going on in AI policy in the United States. This might continue for some time, it might not, but at least right now, that’s where we are. What are the failure modes of that? And for that matter, what are the things that are really good about it that people don’t appreciate?
Dara: The good news about being in uncharted territory, as opposed to the one weird trick school of things, is that you’re not rummaging through the costume bin trying to find things that are going to make you look the most like a wizard, which is kind of the way the one weird trick thing gets done. You look all the way through the books and you’re like, “Oh, actually, there’s this law from the early 20th century that says if there’s an epidemic, then I can do an end run around immigration law by suspending the entry of people.” That was Title 42, which was how we had a complete ban on asylum for the end of the first Trump administration, and then some really wonky implementation under the early Biden administration that I don’t have to get into. Instead, there just aren’t words out there. There is no vocabulary.
Which means there’s a little bit less of that. The biggest difference between the first and second Trump administrations on immigration is that the federal courts, specifically the Supreme Court, are much less willing to step in and say, “No, we don’t agree with your interpretation of this.” One of the fights that is going to hit the Supreme Court this next term is that the administration has taken a provision that says somebody who comes in under these circumstances without inspection shall be detained, and has taken that to mean we don’t have to have bond hearings for anybody who’s ever entered the United States without inspection, even if they’ve been here for ages and ages and ages. That’s not traditionally the way it’s been interpreted. And it’s not just that it’s not traditionally the way it’s been interpreted; it could not have been interpreted that way, for feasibility reasons, because you just don’t have that many beds in detention. Even now, while they have all of this money, actual detention capacity is not increasing at the pace that arrests are increasing. Facilities are being overcrowded. There’s a lot of games being played with temporary holding facilities, et cetera.
It’s not something that the federal government could have taken literally as a “shall.” If you know federal law, you know that there’s a big difference between “may” and “shall,” and with “shall” there’s a lot less discretion, theoretically, except that in practice, it turns out we’ve all kind of assumed that there was discretion there. And the question of whether that discretion really exists or whether it was just read into the law is a very real one.
But there’s also the stuff that really was intended to have discretion. There’s an assumption that you’re not going to assign, say, special agents who are supposed to be doing investigative work to do low-level enforcement, that that’s not a thing you do to plus up your low-level arrests, because you want to have these guys trying to dismantle criminal networks and all of that. And—
Dave: Like preventing rocket parts from getting smuggled to Iran or whatever. That’s a thing that ICE HSI literally does.
Dara: And speaking of institutional culture, when they tried to do that kind of thing during the first Trump administration, you’d get a lot of grumbly leaks from HSI agents, like, “This is bad for us. This is bad for our brand.” Not quite as many of those this time around, which indicates to me, and I have no inside information here whatsoever, just reading the tea leaves, that there’s been some change in HSI personnel between 2017 and now, such that those people are now more likely to be aligned with the Trump administration’s mission. But the presumption of regularity is a term that comes up in court a lot these days, because—
Dave: Define that term for our—
Dara: Yes. The judicial branch has always assumed that the executive branch is acting in good faith, that it’s not openly lying and openly trying to create a paper trail to cover its tracks over the real purposes of things. And that’s really important, because when you’re trying to evaluate whether a policy was conducted in the correct way, whether it was developed according to the federal laws that say you have to have a notice and comment period, you have to get input from X, Y, and Z: if you have a paper trail that shows they did that, but in the background there were phone calls going on, or Signal chats, it makes it a lot harder to say, “Okay, we know exactly what’s going on here.”
Then there’s the willingness of attorneys representing this administration to stand up in court and say, “I don’t know,” and be apparently unable to find out facts that are very, very salient. For example: are the planes to El Salvador still in American airspace? Things that are not only knowable, but knowable by a civilian, and that can, in fact, then be found to contradict what the administration is saying. When that gets worn away, it’s not super clear what replaces it, because so much of the way that government works is predicated on the assumption that people there are trying to do their jobs in good faith and are trying to work together. It relies on the kind of thick tissue of norms that those of us who care about norms saw getting eroded 10 years ago. DC is still, 10 years after Trump was elected the first time, not really in a position of understanding just how much of the way the city always runs was predicated on this assumption of collective good faith. It keeps surprising us. We keep stepping on the rake about it.
Dave: Yeah, and to be clear, this sometimes disadvantages the Trump administration as it’s attempting to achieve its own goals. There are certainly many instances you could point to of courts slowing down or stopping something that they wanted to do purely because they didn’t trust the administration’s representations, and not necessarily because the administration had an underlying basis to anything that violated the law. And indeed, there have been instances where, when they finally opened up the hood and got a bunch of depositions, they said, “Okay, fine, this actually is lawful,” and the restraining order or whatever got dissolved. But beyond that, there’s this sense that I think you’re articulating that DC is like this dance that everyone’s involved in, and everyone is expecting everyone else to be watching out for their own role in the dance, doing that, and not suddenly freestyling.
Dara: I do a certain number of panels. Often it’s like, oh, a local chamber of commerce’s members are coming in for a lobby day, and they want to have some panels about the state of play on various issues that are relevant to their members. So they’ll get someone from an organization like mine, and they’ll get someone from an organization like Heritage, and you’ll make sure that you have a balanced panel that way.
Dave: Right, because the goal is literally just: hey, you’re the Chamber of Commerce from Cleveland. Your job is to genuinely understand what is going on in this debate in DC, not necessarily to be convinced of one side or another.
Dara: Right, and how likely it is that anything is going to happen. If there are things you want to happen, you want to know: is this really worth your 10 minutes with your member of Congress? My first several years in DC, I don’t think there was anybody I was on a panel with, even people who disagreed with me normatively, who not only said things that were empirically wrong, but didn’t nuance it at all when pushed back against. During the Biden administration, I was on a couple of panels with former Trump 1 officials for whom it was very clear that they were seeing this as no different from going on a cable news hit. And those are very different codes of communication for most people in DC. It’s much harder to have a conversation on a panel when your counter-panelist is talking about how it’s really important to shut down asylum because there are tons of people coming who are agents of the Chinese government. What do you do with that? You can say, “Hey, guys, this is wrong,” but it really does look to the audience like a he said/she said.
Dave: Yeah. And part of what I’m proud of in the AI policy world is that, certainly there are folks who are operatives and agents and things like that, but one of the things I tell people all the time is that we need to try to hold onto, as long as we can, the notion that it is very much still a non-partisan kind of thing. Yes, there are issues that have been partisan. Yes, there have been votes that have been taken. But by and large, if you sit down with the average Republican or Democratic staffer, you’re having a real conversation around, “Hey, what are the current capabilities of the models?” It’s not like one side is like, “I think it’s this many FLOPs,” and the other side is like, “It’s my political enemy in every possible way, and I want to focus on that.” It’s like, no. Okay, let’s talk about what the actual implications are for child safety, export control, national security, and they’re really still trying to dive at that. So maybe you can tell me a little bit about what you think the things are that we need to watch out for in an effort to hold onto that. If we’re one of the last islands left where that’s still true, how do we protect our little metaphorical island of good faith debate on this?
Dara: I am not persuaded that there isn’t a real trade-off between salience and bipartisanship. Fundamentally, the reason that immigration wasn’t the top thing on the agenda in 2009, 2010 is that the Democratic Party was setting the agenda, and they didn’t treat immigration as a Democratic priority. It was kind of like, “Well, there’s also enough Republican interest in this that, A, if we pass it, we won’t get credit for it.” And B—
Dave: And then all the proposals were, like, gang of whatever: a bipartisan gang of senators or whatever.
Dara: Right. And B, the base is going to be a little bit hesitant, a little bit suspicious of it. So that is problem one. The good news about where AI is right now is that the X-risk of it all has percolated enough that it’s a little more like climate, or at least like climate among people who agree that climate change is occurring. You can’t really argue that it’s not potentially huge.
But the other thing to watch out for is political entrepreneurship, because it’s very, very easy to get something polarized. You just need someone who has really, really strong favorables and unfavorables to take a strong position on it, and then watch everything fall out from there. Very frivolous example: you know that I’m a big IndyCar enthusiast. When Donald Trump made a big deal of announcing that there was going to be an IndyCar race in DC, all of a sudden this series (I have my IndyCar pride hat that I wear around all the time, and I think of it as not necessarily being Republican-coded in the same way NASCAR is) was something everybody in DC started assuming was all a front for the Trump administration.
But that happens on real serious issues all the freaking time. Arguably, the big turning point on immigration polling was when the Gang of Eight bill was associated with Barack Obama. I have not been able to track this down, and if somebody who has better Google-fu than I do manages to, I’ll be forever grateful. But I absolutely recall there was a poll where it got majority support among Republicans as long as they weren’t told the president supported it. Which tracks entirely with everything we know about negative polarization.
I think the question of data centers is separate enough from the question of AI regulation that this does not necessarily need to be the case. But it absolutely could be the case that data centers are creating a real opening for political entrepreneurship on the NIMBY side of things. And the more that one party becomes visible in support of that and gets rewarded for that, the more polarized that issue becomes, and you cannot undo that ratchet. You cannot depolarize it. Democrats keep failing to learn the lesson that you can’t depolarize immigration just by having a more hawkish policy, because that gets back to what we were talking about earlier: how do people know?
Dave: And I think the other thing I hear you saying is that once a frame gets established that this issue is now a partisan issue, and there are ways in which you’re adopting the, call it, rhetorical and emotional aspects of that, then it’s very hard to break out of. To your point about immigration a moment ago, there’s a part of it that’s like, hey, what are our very technical policies about different types of immigration rules? And then there’s this thing around: do you view immigrants as a threat to the American way of life, government, pick your words? I’ll leave it to our viewers to decide where they stand, but certainly if you do choose to buy into that frame, then you’re inherently aligning with one political point of view. That’s what I hear you saying.
Dara: I think the other thing I would say in terms of watching out for polarization is, to the extent that bipartisanship is really a matter of each party having some constituents in its camp who want to see something happen, those constituents had better have the juice. The Chamber of Commerce got embarrassed by the Republican base in the 2006 and 2007 efforts to pass immigration reform. They just couldn’t stop the fact that Republicans’ phone lines were lighting up with people who opposed, quote unquote, “amnesty.”
Dave: Just to unpack something: you’re saying that the Republican Chamber of Commerce was in favor of amnesty, and then the base was just constantly yelling at their representatives, like, “What the heck? Why are you thinking of doing this?”
Dara: Yeah. The theory of the case during the 2006 to 2013 period of immigration reform was: we know that business and labor both have really strong commitments on future flows of legal immigrants. Let’s put them at the table and have them work something out, and then obviously everyone will be represented, and that’ll cover it. But you then have to have each side blocking and tackling, and that means they have to be capable of winning intra-factional battles. It is amazing in retrospect how juiceless the Chamber was, not being able to stop Republicans from running away from those bills over the early Obama administration and into the second term. Because at that point, it was still assumed that Republicans were the party of business. And yet there wasn’t enough juice to get a rank and file Republican congressperson to align with this thing that the supposedly very influential Republican lobbying interests wanted them to get on board with.
Dave: Yeah, and another thing I’m taking away from our conversation is that we saw a lot of extreme consternation in AI policy early on, when various political action committees got established that were starting to donate to candidates based on their views on AI policy, whether that’s Leading the Future or Public First or some of the others. A lot of folks were feeling like, “Oh, this is going to suddenly push all of the conversation in one direction.” And then in reality, at least our experience has been: no. Sure, there were some political giving things. Palisade stays out of that; we’re a C3, so it’s not our lane. But it didn’t have the impact on policy education and policy discussion that we were expecting it to have, in large part because a push saying, “Just don’t enforce any laws on AI, have a total moratorium,” looked less and less feasible with each passing month. So they ended up having to renegotiate it several times. There still is a draft bill on the Hill for another preemption proposal that’s much more narrow in scope. But it got to the point where, even if a lot of donors were pushing, a lot of interest groups were pushing for it, it just wasn’t sustainable the moment legislators started talking to, not even their base in reality, but just other people. It starts falling apart.
Dara: There’s also this: on new issues, or issues where there isn’t necessarily a whole lot of expertise among congressional staff, and this happens every so often on various issues when the generation who actually worked on a major bill is no longer on the Hill, you have people who have not gotten into the policy area in any major way, and it’s one of five things in their portfolio. That has real downsides in terms of getting people to understand when things are important, but it also creates a certain amount of flexibility in the issue space. You don’t have a situation where not only the member of Congress, but also their legislative director and also their subject matter expert, all have really strong, really well-developed opinions about how things should go because they’ve been working on the issue forever. So that does create a certain amount of latitude. But at some point you do have to educate people about the policy, if they’re going to be writing laws, if they’re going to be writing regulations. And people don’t always want to do that.
I think those of us who find systems very interesting are going to proactively seek out things to learn about, and that is not true for most people, including most people who are high information. A lot of people have gotten very used to having media outlets or feeds that serve them information that is relevant and useful to them, and have not really gone back and built from scratch: how do I get a media diet in the age of short-form video that is satisfying to me? So what you end up having to do is take these opportunities when things do become really, really salient because of events, and cram in as much information as possible.
I used to have a short spiel when I was talking to journalism students about how you cover a Trump tweet. You have to do it; someone is going to get that assignment every morning. The way you do it is you say, “This is what he posted,” and then you have two paragraphs to say as much as possible about the world as it actually is. It doesn’t necessarily have to be a fact check. It can be: here is the controversy that is leading him to have a really strong opinion on this thing all of a sudden, and here is the reality behind that. But you take the opportunity while people are actually paying attention to something to jam some information into their heads before that door swings shut again.
And over time, you hope that you get to a somewhat higher level of understanding. Ten years ago, I could not have expected the average person, even in DC, to understand the difference between CBP and ICE, and I can expect that now. Enough cycles of this have gone on that I do anticipate a certain higher level of policy knowledge. It’s just that it’s nowhere near what would be needed to actually think about, for example, what you would need to do if you wanted to restore a civil service orientation to the executive branch on immigration.
Dave: So if you were in our shoes, and you’re thinking through, hey, you’re going to get a bunch of moments of AI stuff that are going to get bigger and bigger—
Dara: I will cop to this, as someone who outsources most of my AI thinking to you: I was one of the many people who should have known better who took the Hugging Face incident as, “Oh, okay, this is my cue that I really need to learn some stuff myself.” It’s moments like that where you’ve got to (and this is also about social cues) take the opportunity to walk your normie friends from the very beginning of the path.
Dave: Yeah. And this is one thing I think we’re terrible at in AI policy, in total honesty. We’re a lot less terrible than we used to be. I think Palisade is among the orgs that are working on getting better at this and helping drive our field in a better direction. But the way that we used to create AI policy people, or AI governance people (we didn’t call them AI policy people), is that we would tell them to sit down and read 100,000 words. And then when they had read those 100,000 words, maybe they were ready to do some tweets or some posts on LessWrong, and then we would argue with them a lot. And the ones who chose to stick around after getting argued at when they were trying in good faith to express forward-leaning perspectives, they stuck around, and eventually maybe they got a job in the field.
This is obviously just not a way that you can run a society. There’s no way you can possibly expect everyone to read a book’s length worth of content or more on every issue, even if it’s the most important issue in the world. I don’t think the average American read that much on COVID. They absorbed a lot. They absorbed what was important to them, but it was distilled in a bunch of ways. And we cashed in a bunch of credibility that had been developed over decades.
One point that I make all the time is, setting aside your opinion of Dr. Fauci now, and I know this has become a partisan issue: in 2020, he was a guy who, by the way, was in one of those civil service roles. He had decades of built-in institutional credibility. He had become the institution in some sense. Every reporter who covered that beat knew him from the Ebola stuff and Zika and all these other things. They knew him as the guy. AIDS back in the day as well. To the point where there is literally a clip out there from, I think, the 1992 presidential election (maybe it’s ‘88, but I believe it’s ‘92), where George Bush Sr. and his opponents are asked, “What’s a young American that inspires you?” And at that point, Dr. Fauci is named as a young American that inspires George Herbert Walker Bush, for his work on the AIDS crisis. When you’re someone who, for longer than many people have been alive, has been named as an important American, you have a lot of credibility built up, and you can exploit it. But if you’re not one of those and a crisis hits, you’re in a very different place.
Dara: I think the other thing I would add, looking at your role from the outside, is that there are a lot of people with microphones, especially in the business and capital world, who are assumed to know things about business and capital. Some of them actually have been doing the 100,000 words’ worth of reading; some of them have not. And there’s absolutely no way for somebody like me to tell the difference. When a CEO of a company that is not obviously involved in higher-level computing is saying that AI is going to transform their industry, I can’t tell whether they themselves have done the reading, whether they’re listening to someone who has done the reading, or whether they are following the cues that say, “This is important, you should be talking about it,” or the cues that say, “This is a great opportunity to fire a bunch of your workforce. It’s okay, you can hire them back later if you really need them.”
I’m sure I’m not the first person to be frustrated with this on this podcast: most of my peers are much more reflexively Luddite than is helpful, just assuming that any claims made about AI are smoke and mirrors. There are very good reasons for that; I can understand how they get there. People who do words for a living feel more existentially threatened. Hi, I’m someone who spent a long time trying to maximize the amount of information I was presenting to people in a clear and accessible manner. Ask me how I feel about my writing style. So there are ways in which people feel directly threatened, but it’s also that it’s really hard to tell from the outside who is signaling credibly that they understand this issue.
And I think that, in general, people don’t think enough about information ecosystems. There’s a lot of talk about messaging in any advocacy space: how do we best frame? I’m not saying that’s not important. We tell ourselves stories in order to live, yada, yada, yada. But you’re not the only person who is broadcasting a message. What messages are even reaching people, how are those messages interacting with each other, and who comes out of there trusted enough that they even get to continue sending messages: those are really questions that cannot be answered in general terms. You have to get really, really specific, which DC is very used to, because we have all kinds of advertising campaigns that are just geared at Pentagon appropriators.
Dave: Oh, yeah. One of the things that’s amazing is that the average DC person, even if you’re not involved in government in a single way, has an information diet that’s fundamentally different from the average person’s elsewhere in America, in Cleveland, Chicago, whatever. If you’re getting on at Pentagon Metro (which, by the way, you might do even if you don’t work at the Pentagon, because it’s a big bus interchange as well), sometimes the station’s entire ad inventory has been bought out to run ads that are basically targeted at about 10 people. It’s literally, “Hey, Pratt & Whitney,” or whoever, “wants you to buy their new jet engine, because it’s time to replace all the jet engines on such-and-such fighter jet, and when you have to pick which one to upgrade the jets with, you should choose Pratt & Whitney’s,” or whoever’s. There might literally be under a dozen guys making the decision about which engine to pick, and all the ads are for them.
Dara: I think the instinctive response to that is, “Wow, what a waste of money. How much money do you have to burn to do that just for 10 people?” But if you knew that there were 10 people who could make the decision on you getting a contract or not, and you knew exactly where you could find them between the hours of 7:30 and 8:30 Monday through Friday, of course you would want to be there.
Dave: Of course you would. You would spend whatever it costs. Frankly, one of the things that’s funny is that the contracts don’t cost even more. And this is something our listeners can do: you can go and look up the WMATA ad card for different stations, which is public, and you can figure out which stations people have figured out are important for certain types of issues, because the rate for them is much higher. For example, the Convention Center unsurprisingly is higher, but on Capitol Hill, certain stations are more important than others, and it’s not always the ones you would expect.
There’s a very real sense in which, if you’re in DC, you have to learn how to read all these little cues of why things are happening other than the way you think they might be, in order to infer what’s going on. Similarly, this is true of YouTube ads. You and I get served YouTube ads all the time that are clearly intended to target a bunch of congressional staffers or a bunch of White House staffers. It’s literally on the level of, “Previously President Biden, now President Trump, is a leader on [insert issue area that we care about]. He needs to keep on leading by doing the thing that my pressure group wants him to do.”
Dara: I actually think this is something where you do need to distinguish between Trump and previous presidents, because the tone of these ads is much more explicitly, “Donald Trump can save us if he does this, or if he keeps doing this.”
Dave: That’s fair.
Dara: Whereas prior to that, you would have had more of an effort to make an ideological argument or a common-interest argument, something like that. It’s wild how transparently the audience of one has gotten incorporated into this ecosystem, where no one thinks Donald Trump is going to see your YouTube ad, but people who are making decisions based on what Donald Trump wants might.
Dave: I think this is a shift where, in a world where this kind of trend continues, very soon this’ll just be how it is, for better or worse. And in some ways, frankly, to be clear, I think there is something to be said for the honesty of just being very blunt: there is a decider, and we’re just going to make an appeal to the decider directly, rather than masking it in the cloth of seeming to appeal to a neutral party.
Dara: Right, and I think there’s also something to be said for being responsive to events. In a world where there are two ways that Congress does things, and one of them is enormous bills that get put together in two days, and the other is enormous bills that get put together in a year, and Congress has done neither of them, really, for the last few years, you do need to have the ability to exert government capacity in a particular direction. This comes back to what I was saying about the Canadian legal immigration system. There’s a real argument that things have changed since 1990. Maybe somebody should be empowered to say, “We should change things in response.”
Dave: Yep. We’ve talked about this in, I think, every podcast we’ve done so far: one of the things that comes up constantly when you talk to Congress is that they view themselves as having a lot less agency and a lot less ability to get stuff done than they did 20 years ago. When you talk to people on Capitol Hill who are focused on immigration, do they feel like that is also true of them? It seems, from something you said earlier, that maybe it is. But I can also tell a story where it’s like, “Well, my issue became so much more important, so now I can actually get a lot more done than if I’d taken this job 10 years ago, or when I started this job as a junior person 10 years ago.” Does anyone feel that way, or does everyone also feel the same sense of, “I don’t know who’s in charge, but it sure ain’t me anymore”?
Dara: Take all of this with an enormous grain of salt, because I’ve actually structured my job such that I don’t have to talk to the Hill a whole lot. I’m not really characterologically an advocate, so when things really need to be advocated for or against, I’m not really the person you want in the room, because I’m going to say, “Yeah, I understand where you’re coming from, but if you want to achieve those goals…”
But what I have gathered from colleagues and what I’ve seen from the outside is that now that immigration is a polarized issue, rank-and-file Democratic members of Congress are interested in messaging bills. But it took not just the macro-polarization. Frankly, there was also a big, big shift between spring of 2025 and spring of this year, in particular over January of this year. Once Renee Good and Alex Pretti were killed (honestly, it took the second ICE killing), all of a sudden there was a lot more appetite among Democratic members of Congress. The switch flicked from “Trump won on immigration in 2024, so this is a weak issue for us, so we need to run away from it” to “This is something that has our base really fired up, and we should give them something.”
What that looked like in the short term was a bunch of members of Congress who had not expended serious legislative capacity on this in the past coming up with ideas about what they wanted to do, and a bunch of very similar bills getting floated around and introduced, because all of a sudden everybody in a safe seat wanted to have their name on something, and there are only so many ideas for what we can do. So it was a lot of effort that could have been better corralled or coordinated, but also a lot of effort on things that no one really expects to pass.
I think there is appetite (at least, it sounds like there’s appetite) for doing some real oversight of appropriations in particular: certainly for not continuing to give as much in enforcement to ICE and CBP, potentially even ways that Congress could claw back some of the funds that have been allocated but haven’t actually been committed yet. I’m not going to say that I would predict that happens, but I think it’s more of a possibility than legislation, where everybody sees the alternatives as either we do nothing or we pass messaging bills.
Dave: One thing that I’m always befuddled by is that it really does feel like not only are people only doing messaging bills, but they’re not really coordinating anymore on messaging bills. When you and I first showed up on the DC scene, there were certainly plenty of messaging bills then, but (and some of our friends who worked on the Hill at the time would, I think, corroborate this) there was this vibe of people talking to each other about what messaging bills they wanted to do and wanted to deploy. Instead of a random office dropping one, it was much more like 12 offices all supporting this bill. This is broad strokes, and I’m sure there are plenty of counterexamples, but it feels like all the offices are more isolated than they used to be, and they’re all more just doing their own thing and not talking to each other very much, on both sides of the aisle. That seems to have made for a lot of duplication, in exactly the way you described, but also a lot of paralysis as a result, because people don’t start coming together around a shared idea.
Dara: We were talking earlier about how DC is really still reeling from the rupture of the tissue of norms. Congress is the institution that has failed the most to examine itself in the light of this, and it’s the institution that’s arguably been damaged the most by it. I think there are real consequences of the long-term trend toward wave elections, and what that does for the tenure that a given member of Congress is going to have, not to mention for their most existential threat being not potentially getting waved out of office but, if they play their gerrymandering cards right, getting turfed out by their own party. I think that has consequences for the ability of different offices to have legitimately high-level conversations, where it’s not staff-mediated, it’s not, “I think that my boss is okay with this,” but the people who are actually making the decisions are hashing things out.
That hasn’t fully been metabolized: the extent to which that’s going to mean really weird things for people trying to make careers on the Hill, because it’s not like the civil service. You don’t say, “Oh, okay, I was doing science policy under a Democratic member of Congress, and now that there are fewer Democratic seats, I can just go work for a Republican member on exactly the same issue.” Some committee staff are nonpartisan, but—
Dave: Most aren’t, yeah.
Dara: —member staff really aren’t. So every two years, or every time control of a chamber flips, you have a need for jobs in a policy area where there weren’t jobs before, and hopefully you have people with the policy expertise to fill those jobs. At the same time, you have a bunch of people who have developed some level of policy expertise who now need to figure out, “Am I going to go into the nonprofit sector? Am I going to try to get a job in the federal government on this issue? Or am I going to pivot, not stay current in the field, and lose whatever institutional knowledge I’ve gained?” These are all questions that I would really like to have a sense congressional leadership was in any way engaged with.
And I’m probably part of the problem here. I’m much more interested intellectually in the executive branch than I am in Congress, because the questions I’m interested in, about personnel and institutional culture, are much better fitted to the executive branch. But it means that there aren’t as many people taking the kind of org-soc approach to Congress that they take toward executive branch agencies.
Dave: Yeah, and you see these dynamics in very direct ways. A couple weekends ago I literally went to a yard sale in my neighborhood where the people were moving out of town because they had worked in an office in Congress and then someone had lost an election. They eked it out for a while by pulling things together in terms of other jobs, and then they were just like, “You know what? Actually, we’re done here. We’re going to move back to hometown.” I’m not going to name where, but they’re moving back to their hometown, and they’re like, “I guess I’m doing something else now.” And it was literally just because the system isn’t well designed to handle that sort of headcount transition, because, as you know, ever since the 1990s it’s been much more based around personal offices than around committee membership, and the committees are the stable part of things.
Dara: Yeah. This is a pretty broadly applicable lesson for a field where a lot of people have recently gotten brought on and where even the near-term economic future is very up in the air. What does it look like if… I’m not going to play out a specific chain of events, because I have no idea how this would happen, but I’m sure there are worlds in which six months from now there are substantially fewer people employed in AI policy, and two years from now there are more than there are right now. What happens to the people who are losing their jobs in the next six months, and then 18 months after that? And, oh my gosh, 18 months is so much time for you guys.
Dave: 18 months is really far away.
Dara: Worlds can change. So the question of how many people are staying abreast, or retaining the knowledge, or continuing to affiliate their identity with “This is my issue”: those are not things that you can even really incentivize centrally. That really is the aggregate of a lot of individuals making a lot of individual decisions.
Dave: Yeah. In total honesty, I do think there’s an increasing awareness among funders in our space that, hey, personnel shifts do happen. In some ways the DOGE thing first sensitized a bunch of funders in this space to that. And I generally have the sense from some conversations that people are tracking how the midterms might affect that to some degree. But I’m not sure we’re on the ball on this. There are some folks who just happen to work in an office but are happy to work in a nonpartisan capacity. They’d love to go work for a nonprofit. Are we actually making sure that we’re putting opportunities in front of them? I’m not especially sure that we are. I think—
Dara: Oh, here’s the other thing that is so obvious that I wasn’t even thinking about it when you were talking about the shift from committee to member staff. The easiest way to expand policy knowledge and capacity in Congress is to get somebody who currently works for an executive branch agency on detail and have them be in your office for six months, so you have someone you can ask, “Hey, how does this law work?” or, “Hey, am I reading this right?” In a world where the federal government itself has less institutional knowledge and less of a civil service orientation, that is harder. It is harder for a Democratic member of Congress to get a Republican staffer in this administration to be their helpmate on it. And also, on an issue area where you can’t necessarily expect there to be a whole lot of… it’s not like there are decades of executive branch competency on AI either. So that’s a couple of dimensions on which you guys are really getting caught up short.
Dave: Yeah. So what have you seen in your space? It sounds like it’s been imperfect, but what are the areas where you’re like, “Hey, the immigration community really did do a good job of trying to skill people up and teach people to understand the details of immigration policy to a much greater degree than previously”? What are the lessons that we could just steal from you guys?
Dara: Unfortunately, in anything that is an area of law, the pipeline is built for you, because there are a lot of smart early-20-somethings who don’t know what they want to do with their lives but know they’re good at reading and writing and arguing, and decide that they should be lawyers. You and I are the two people we know who successfully resisted that call, but it took a whole lot of quarter-life crises on both of our parts.
Dave: Several, yes.
Dara: So the traditional way that you develop an interest in immigration policy is you go to law school, you do your immigration clinic, and then you do some pro bono or low bono work doing direct representation. That actually creates some massive, massive biases in the space. Whenever we—
Dave: Such as?
Dara: Whenever we talk about the process for asylum seekers: people don’t understand what asylum is. Certainly people who are trying to get asylum don’t necessarily understand the nuances of the Refugee Convention. So how you make it accessible to them, in a world where there are a lot more people than there are lawyers for them, is a very live question. And some of what’s been proposed is, well, maybe you get rid of the adversarial process entirely. Maybe you have someone who’s helping them navigate it rather than somebody who’s on the other side of the table. People who have done direct rep are generally really strongly opposed to that, because they have seen the importance of representation for immigrants, and the importance of an adversarial system in arriving at truth. So they don’t want anything that could get in the way of robust representation.
And I understand that, but as someone who has not had that life experience, I’m not thinking about this from the perspective of somebody who’s representing the 33% of immigrants who have representation. I’m thinking about it from the perspective of (and these are old, Biden-era stats) the 67% who don’t, who are now being thrown into this legal system that they have no awareness or understanding of, and being told to—
Dave: Figure it out.
Dara: Being told to put together the strongest application possible.
So having a set feeder system certainly does have problems, but it does mean that you haven’t had to cultivate pipelines. There’s a lot of concern about people burning out, in the same way that there’s concern about people burning out in anything that’s bleeding-heart affiliated. But at the end of the day, there are enough of us who are just willing to stick with this until it becomes less totally borked as a policy. I already told the story about when I first got a job in DC, but there was also a period (this is more like 2013) where I was thinking, “Okay, you know what? We’re going to pass immigration reform, and then I can go back to school, get another degree, and then come back and fix mass incarceration.” But in the absence of the thing being fixed that I moved to DC to fix 20 years ago, it’s not just pot commitment. It’s also that this is actually the thing that I care about enough to fight for. And it just seems like there’s a critical mass of people who have developed that same orientation, mostly through being in the trenches, but in plenty of cases through direct family experience, and in some cases through just applied Judaism. The extent to which a secularized Judaism just ends up embodying social justice politics, you know?
Dave: Yeah, for sure. And there’s a very real sense in which one of the things that’s quite unique about the AI policy space is that it is another deep refuge of people who often don’t have significant religious practice anymore but are deeply informed by the values they learned from religious communities. We have this joke that the average person in the AI policy space is either a liberal Jew (liberal in the sense of practice, not necessarily in terms of politics) or someone who used to belong to a fairly intense Christian faith community, whether it’s evangelical or something that has very strong commitments, like a Jehovah’s Witness community or something like that. There are very few people who are just like, “Yeah, I was kind of loosely Protestant. I occasionally went to church when I wasn’t going to hockey.” Very few of those people show up in our community, and I think it’s because there’s something about taking ideas really seriously, and the ethical call they have on them, that seems to really pull people into this space in a particular way.
Dara: Yeah. A thing that I’ve been thinking about over the last six months is that you and I came of age at the end of the very weird period in human history where, if you were good at words and logic, that opened up doors of power for you. Historically, that was not the case. It’s arguably much less the case now, and it arguably isn’t going to be the case at all, depending on how things shake out in the next few years. But people who think that ideas should matter and ideas should shape the world need to remember that we’re the weird ones. We think that in large part because we’re good at ideas, and we’re good at using our words to shape ideas and to shape other people’s concepts of what the world is. We can’t expect everybody else at the table with us to have that, and it’s useful to think a little bit about what other reasons there are that someone who is not necessarily all in on the cosmology of it might be brought along on understanding the importance.
Dave: Yeah, and there are people I know, whom I respect very deeply in this space, whose true core objection, when they hear AI safety ideas, is, “I’m sorry, I just don’t believe that something can destroy the world, because that’s not my theology. God has promised a certain set of things. God has revealed a certain set of things. That is what is going to happen. If you tell me a story of how this fits into that, okay, we can have a conversation. But at the end of the day, there’s only one person who gets to call the end times, and that’s the big guy upstairs, so I’m waiting for that to happen.”
I think people’s true objection when they encounter AI policy stuff is often not, “Oh, this feels weird,” or, “This feels too science fiction,” in just a generic sense, but rather, “Hey, I actually have a very deeply ordered sense of how things are supposed to progress in capital-H History, and this does not accord with that narrative.” And by the way, that’s not only a religious thing. You also see it with the folks you were talking about earlier who are Luddite. I think some of them are like, “Well, no, my understanding of what’s supposed to happen here is that capitalism is supposed to make things shittier until eventually we have the glorious workers’ revolution or whatever.” It’s a not-uncommon thing you hear on the left. And it’s like, yeah, maybe that’s the story, but actually it might be something else, much more complicated, as with a variety of movements that were reactions to the nature of their times. Maybe we just need to take this moment on its own terms, politically.
Dara: Yeah, but telling people who are good at ideas, words and logic, and therefore probably good at pattern matching, that they should pattern match less…
Dave: Extremely difficult. This is one of the things where I genuinely think: let’s say we get to a good ending on policy. We get to a point where we figure out how to build things that are as smart as or smarter than humans, and have the right government system in place to ensure that the risks are mitigated all the way through, and at the end, we genuinely have an aligned superintelligence. And we sit down with it and we’re like, “Hey, Deep Thought. ‘42,’ we knew that already. But beyond that, tell us what was actually the thing that mattered to get us here?” I think there are entire worlds where the answer is just: the comms stuff didn’t matter that much, the policy stuff didn’t matter that much. It was just things happening in terms of technological advancement that triggered people to realize, “Oh, this isn’t fake. This is real.”
Dara: This is what I’m talking about: events exogenous to the policy process. There’s a line that’s a cliché in DC, and probably not a cliché outside it, that’s attributed to Rahm Emanuel, which is, “Never waste a crisis.” That’s usually seen as this really icky, exploitative thing: oh, it’s gross to take events and use them to push your preexisting agenda. But the flip side of that is that nobody cares about your preexisting political agenda until they do. And they won’t care about it for very long. And if you manage to get them to care in the way you want them to care, and you understand the things they need to do to pressure the powers that be, man, that’s—
Dave: Gold.
Dara: That’s the Konami Code everybody’s trying to unlock, and nobody can actually unlock it. But that’s really the closest thing I have right now to an actual theory of change.
Dave: So this is what scares me. I was actually talking about this on an unrelated call today with some folks I’d been introduced to via some friends. They’re asking for opinions to try to understand DC policy on AI stuff, and I was happy to chat with them. And they were kind of like, “Hey, it seems like when we talk to all of you guys, you’re all kind of waiting for the thing that really shifts policy, and that might be a disaster or something like that. Are we missing something?” And I basically had to tell them, “Well, I genuinely and sincerely pray that that’s not the case.” I hate the term warning shots, in part because it’s not a warning shot if there’s a body count. Even if it’s a warning shot in some sort of weird, messed-up utilitarian sense relative to the entire human race, it’s not a warning shot.
Dara: You and I both know enough about elementary marksmanship to know that very, very few people are actually good enough to fire warning shots that don’t hit anybody.
Dave: Okay, so we’re going to digress on this for a moment, then, because this is actually a really important point. The concept of a warning shot is not a thing that people are trained to do. In fact, they are trained not to do it. There is a limited sense in which—
Dara: Any firearm safety class in the country will tell you, “Do not point your weapon at anything you do not intend to destroy.”
Dave: Yes. In a very limited sense, in certain rules of engagement, the US military is occasionally told to do warning shots. But even then, it is considered kind of a bad idea when you talk to people who have been in these crowd control situations. Because unsurprisingly, you know what tends to trigger a riot when you’re worried about there being a riot? Having a gun go off. And I think there’s a very real sense in which a metaphor is powerful, and I really dislike the notion that we are using a metaphor that actually doesn’t really map onto reality.
Dara: Well, or to the extent that it does, who is firing these warning shots? Is it an entity over which you have complete control and are dictating the rules of engagement? And is it an entity that is competent enough to execute those rules of engagement perfectly? Maybe the answer to both of those questions is yes at this particular point on the METR curve, but I’m not sure.
Dave: I’m sure not sure. I think it is objectively true that you shouldn’t waste a crisis, whether it’s a big one or a small one. And I kind of have this theory that there’s the silly crisis, there’s the medium-sized crisis, and then there’s the big national security crisis. And then, by the way, there’s thing number four, which is you don’t get a crisis at all: you just wake up dead one day, because it turns out that something went out of control and you didn’t realize, and it’s just too late, and you just lose. But you can’t do that much about the fourth one, other than trying to prevent ending up in that world in the first instance. It’s more the first three.
In the silly season, it’s not a warning. It’s just, oh, there was nothing going on in the news, and so the Cuyahoga River caught on fire and people said, “Wow, water on fire. How funny. Let’s pass the Clean Water Act.” Water was catching on fire for decades in Cleveland because of all the industrial runoff, and no one had done anything. So, in fact, there were many warning shots that we’re not taking advantage of, if you’re using this framework.
Then there’s the medium-sized thing. You and I both know that bad things happen all the time and they don’t happen to catch our attention. Sometimes they do, but sometimes they don’t. And even if they do, you don’t necessarily fix it.
Dara: And by the way, for people who are not used to working in areas of public policy, this is the thing that burns people out. This is the thing that actually makes people crazy. Most of the people I know who are dealing with secondary trauma aren’t dealing with secondary trauma from the things you know about on immigration that have happened in the last decade. They’re dealing with secondary trauma from things that were just as horrifying, that they saw up close, and that they then couldn’t get the public to care about.
Dave: And one of the things that I worry about a lot when it comes to AI policy is that we’re going to burn our people out. With the silly small crisis, okay, maybe it’s a little bit stressful, in the sense that suddenly you have an opportunity and you have to scramble a little bit, but eh, it’s a small one. At the end of the day, if you only catch glimpses of it as it goes by, you don’t feel like it’s a big deal. The medium-sized crisis and the really big ones, those are the ones where people are going to feel like a lot’s on the line, and if it doesn’t all go the way they want it to, they’re going to blame themselves.
And maybe sometimes that’s fair. Sometimes, in retrospect, you’ve learned some things and you’re like, “Yep, I would totally play that one differently if I was five years older when it happened.” But a lot of the time, this just happens in policy and politics. Sometimes it’s just the case that, for dumb, stupid reasons, things don’t go the way you want them to. And I’ve seen a lot of people, as have you, dash themselves on the rocks of overwork on this. Do you have any advice, just bluntly, for either our community or for individuals who are starting to feel this way, on how to avoid that?
Dara: This actually ties into something that I wanted to point out, talking about the vibe of waiting. Because we also have a vibe of waiting in our issue space, certainly with Congress, and it’s not because we’re waiting for an event to happen. It’s because everybody knows that Congress doesn’t do anything in the fall until the midterms. And then obviously pieces have been moved into place for which things are legislative priorities, but DC has a much more luxurious attitude toward time than the AI world does. And that is deeply ingrained, and it’s not quite as simple as public sector versus private sector. Part of it is a high-minded belief in deliberation over time. Part of it is just that we’re used to laws about how long things have to take, on the assumption that that’s how meaningful input is generated. But what it means for something to be urgent is just a wildly different timescale.
And I think accommodating that is probably at least an important step to thinking about this. It took me until May of 2017 (when I was a full-time immigration reporter and writing a daily generalist newsletter, at a website that only had five politics and policy reporters, so I also had to be our James Comey person for three weeks) to realize: okay, you know what? The pace of news is not slowing down. I need to stop acting like I’m just having a hard week. I started working out, and was a little bit more okay with not necessarily filing something that I hadn’t been asked to file on a Friday night. But that requires reorienting your attitude toward time, and I think it’s really, really hard, from what I’ve seen of folks in the AI community, to feel morally comfortable with doing that.
And maybe just see it as: you’re interacting with a lot of institutions that have slower metabolisms and longer life spans. You’re going to have to speak whale. And maybe you don’t necessarily have to be slowing down for your own sake, but you do have to be slowing down if you’re going to be intelligible to those people. And by the way, it’s also going to help you not feel, six months from now, like you’ve lived an entire decade.
Dave: I know a bunch of folks who already feel like they’ve lived a year in the month and a half or so since OpenAI Hugging Face popped, and maybe it’s two months now; I’ve lost track. But I think one thing that gave us was a big shakedown. In some ways, preemption was the best thing that ever happened to the AI policy community, because it gave them a series of ongoing threshold events: crisis, then a pause, and, well, okay, energy goes down. Threshold event, and then energy goes down again. And people ended up having a shakedown of figuring out, okay, where are all the bugs in how we communicate with each other and relate to each other.
And we had a bunch of other small things. The whole Moltbook thing was kind of a mini “is this anything?” in terms of a rogue agent crisis. Then you had Anthropic and the Department of War, you had Mythos, Fable, and export control, and then you get to OpenAI Hugging Face. And by then, we’ve actually had a bunch of times to figure out how to communicate better and plan better and all of that. But I don’t think we’ve yet gotten to the point of figuring out how to avoid burning out our people better.
And one of the things that I’m finding kind of shocking in the group chats I’m in is that you can very quickly figure out who has gone through a cycle of personal and professional burnout before and who hasn’t. Because the ones that haven’t are basically saying things along the lines of, “We can keep pushing till 3:00 AM,” and then all the people who have are like, “And what changes if we keep pushing till 3:00 AM?” And the answer is sometimes something.
Dara: Right, or why is tonight the night to keep pushing until 3:00 AM?
Dave: Why is this night different than all the others? And look, a lot of the world, whether it’s DC or the US military or hospitals or consulting or banks, pick your thing, or journalism for that matter, runs on 20-somethings who don’t need as much sleep as people later on in their lives do. But though it is a seemingly endless resource, it is actually a non-renewable resource, because every time you do it, you’re rolling dice on the odds that you’re going to just burn someone out.
Just because there are really important things that need to be done doesn’t mean that we can assume that it’s all going to be over in one spasm of policymaking in the next two to three years. It might be, but it might be that it’s a false sort of overhang for some reason. Who knows why? And maybe we’ll be sitting here 10 years from now being like, “Well, it’s still a Palisade podcast.” It’s episode number, I don’t know, 500 at that point, and we’ve had entire swaths of people in AI policy come in and out and burn themselves out and go work on fusion policy instead or something. And if that’s the case, we’re going to have to figure out how to sustain our people across that and not burn them out.
I think one thing I am going to try to take out of this conversation is really trying to push people more on taking better care of themselves, not only personally, but also building the processes institutionally. How do we have a callback rotation? How do we set norms of when this is actually an important message versus not? By the way, on the latter one, we at Palisade have done a bunch of that, but I think a lot of our other orgs don’t do that. How do we make sure people are getting fed and getting their exercise in and all that stuff? The stuff that everyone thinks they can get away with until they can’t.
Dara: Right, and then the meta level of how you set boundaries that aren’t obviously fake. In theory, there are rules about when people can and can’t schedule meetings at my organization, and there are definitely times when it’s like, “Yeah, okay, I know that we’re not supposed to have any meetings before 10:00 AM, but something enormous just happened.” So you want to make sure not only that you’re setting norms that people will actually follow, but that to the extent that they don’t follow them, that’s not a missing stair. That’s not a thing everyone comes to expect.
Dave: And I do think one of the downsides of AI policy until recently largely being incubated in a moderately large but not giant number of small nonprofits (the average nonprofit is under 30 people, by far) is that we end up having this weird situation where a lot of people just never had the institutional scale to understand how to handle this stuff well. And now we’re starting to get to where there are certainly orgs that have more than 30 people: ARI, FAI, RAND’s AI folks. I’m sure that if you go to the AI companies, there’s certainly more than that on their policy teams these days, or if not, they’re doing something really wrong. And we’re starting to hit all the lessons you learn when you start to scale. You saw that wave of people coming in, getting really excited, thinking it was going to be two years, then it wasn’t two years, and you had to settle in for the long slog—
Dara: Yeah, I actually feel like my more salient experience here is working at a media startup.
Dave: We haven’t actually talked about that nearly as much as I was hoping to. So let’s talk about that in the context of this, and then I want to talk about it more broadly.
Dara: I’m not spilling any specifics here, but organizations that don’t have HR departments often learn the hard way why you need HR departments, because something happens that needed to be an HR incident and wasn’t. Vox Media itself was only coming into the form in which I understood it shortly before vox.com launched. It had just doubled its headcount, essentially. And by making this flagship brand, it was really supposed to be putting its mark on the map as an entity, as an organization, instead of as a bunch of different sites. And on the learning curve, each site was in its own cycle of burning talented 20-somethings into the ground and then getting more talented 20-somethings.
Then it hit institutional scale, where there were both standardized pay scales and a collective bargaining effort. And the collective bargaining effort was really instructive here, because you could see the people who were mobilized enough to be on the front lines of it, because they had the most urgent grievances, setting the tone for the priorities of the rest of the workforce, which was maybe not as diverse as a workforce where people actually had been working for more than 10 years at the company, but was a little more diverse than that. But then that sets the tone for: okay, when there are trade-offs between higher salary floors and more generous parental leave, what are the talented 20-somethings going to pick?
And understanding when those inflection points are happening for the organization, and making deliberate decisions about them, saying, “You know what? Just because this is where our demographic curve is right now, is that the kind of organization we want to be in 10 years, or do we want to keep these particular people?” That’s going to require making decisions that might make it less attractive for your current workforce, in the hopes that the people who stay will stay forever. Those are just things that I don’t see institutions thinking about a whole lot, because they need a whole lot of revenue runway, or at least funding runway. But how do you design an institution that is going to not just recruit the people you want, but retain them? And how do you do so without pissing off your current employees?
By the end of my time at Vox, I probably could have gotten a role at Vox where I was just doing some really fancy immigration projects and just reporting on that. But in the context of, again, a team of five, maybe at that point eight people, that would have been really bad team-player behavior. It’s not good to have a small team where not everyone is in the rotation for another primary debate or another Fox News town hall. That’s not a system that you want to encourage, because that means that ultimately those rotations are getting boiled down to the two or three people who don’t have the wherewithal to say no. And making that decision individually was easier because I’d seen the way that the institutional incentives were turning. But it also meant that by not doing that, I was not making that path possible, or easier on the margin.
Whereas in my current organization, I’m in a role where I have a senior title, but I’m not managing anybody. And it was actually important to the organization to have an example of somebody who was senior as a subject matter expert, or as a non-manager, to make it clear that you don’t need to be developing management skills in order to deepen your expertise and to pull in more money and more seniority.
Dave: One of the things I always find fascinating about journalism is that it’s this community where, if you look back to the middle of the 20th century and look at just the things that people say casually in their memoirs, that they at the time clearly did not think were a brag: they were getting paid, by modern terms, an absurd number of cents per word. They had incredible expense accounts and much lower expectations in terms of publication. And it was because all advertising was print advertising at one point. Well, radio was already there, but in the middle of the 20th century, it was mainly print advertising. TV took a while to really get cooking, and even once it did, print was still pretty large until the rise of the internet.
And the set of patterns and expectations that got established, the “Devil Wears Prada” era, is gone. There’s a reason in “The Devil Wears Prada 2” they have to do all this very careful hanging of a lantern on the notion of, “Yeah, we don’t have as much budget as we used to,” kind of thing, clearly to avoid getting yelled at by everyone in the fashion industry. And this seems to be even more true in mainstream media production. That feels like a thing a lot of people don’t understand about journalism outside of journalism-adjacent communities. Is that fair?
Dara: The macro of that is extremely true. If you look at census data for how many people are telling the census that they’re employed at newspapers or magazines, it is a Wile E. Coyote cliff curve from the beginning of the 21st century to now. And it’s much worse in local. There’s a lot of attention that is paid to the New York Times qua New York Times, and part of that is probably a little bit of what I was saying earlier about people not being very thoughtful about their media diets. But it’s also the case that the New York Times is hiring, year on year, a larger share of the entire journalism industry than the year prior, because it’s the only entity that is not in at least occasional existential crisis.
And the thing is that the long-term trajectory of that is very obvious and has been very obvious to everybody. When I went into journalism in 2014, it was a joke: nobody goes into journalism in 2014. But it wasn’t like anybody had figured out how a website would make money at that time. There were some ideas that people tried, some of which didn’t pan out, some of which panned out for a while.
And the thing that I think is actually really relevant, because I see this mistake getting made in, again, these industries that are not themselves AI industries that are trying to deal with AI as an exogenous thing, is assuming that the thing that will work for the next six months is going to be the thing that works forever. Vox did a lot of really, really smart work, both with articles that were designed to perform well on social media (and you can talk about rage bait or curiosity clickbait headlines all you want, but I had a lot of trouble getting people to share things that were just informative or just sad and not enraging) and also with search. Because when Google had algorithmically prioritized things that were already getting good traffic from those searches, it was a very good idea to build evergreen articles, because those would continue to generate clicks over time. And—
Dave: Hence the Vox explainer sort of thing.
Dara: That wasn’t even initially the thinking behind it, but yeah. It was great for a brand that wanted to really allow people to come in at their own level of knowledge and come away with a higher level of knowledge. And once even the main search algorithm started saying, “No, you have to update the URL in order for us to treat this as salient,” the entire incentive behind generating evergreen articles goes away. There was a joke in media for a long time about “What time is the Super Bowl?” being an article that everybody would write, because people wanted to search for it, and if they searched for it, they would click on it to find out. And that was dumb, but there were also a lot of articles where you were deliberately writing for people who were searching for something because they wanted the answer to it, and you could provide the answer for them.
And obviously we are not in an age where there is a connection between search engines and clicking on a URL anymore. There is so much less incentive to produce text about the way the world is in September 2026, because the reason you would produce that text would be that you would get impressions on your text for it. And without that, there’s a certain amount of eating the seed corn, in terms of there being an expectation that the words will just continue to be there to describe current reality.
And as somebody who works in a field where things change all the time, and where people have always used Google to be their immigration lawyer and are now using large language models to be their immigration lawyer, I think a whole lot about how you ensure that the inputs are strong enough, and what are the decisions that you’re making now on the expectation that those incentives will always remain, when in fact all it takes is another entity making a business decision six months from now, and now we’re all pivoting to video?
Dave: Yeah. And one thing that strikes me about all this is that not only is there a notion that you can feel the sun setting on a certain type of journalism, at least, maybe all of journalism, but there’s also a thing you hear if you talk to people who were at Google at a particular point in time. It is now well established, there’s been a bunch of reporting on this, and you can also hear it if you talk to people who were at Google then, that Google (for those who don’t know, Google basically invented the concept of the transformer-based architecture that now informs most LLMs) intentionally slow-rolled its external application, because they had basically done the numbers on it, and they were like, “Well, this is obviously going to destroy our entire business.” And then, of course, once ChatGPT got launched, they were like, “Oh, never mind. Code red. Someone else is going to do it to us if we don’t do it to ourselves first. We might as well try to do it to ourselves first and figure out how to monetize along the way.”
Many people have written on this. There are some folks who have talked about it publicly, others less so. But it wasn’t like Google didn’t see the end in sight. They understood that if you have a summarization of their information that is that good, it removes the entire incentive to click on articles. To the point now where you ask a question of Google and it pops up the answer box, and sometimes it even hides the search results from you altogether. It’s just the answer box now. And it’s because they’re all in on trying to figure out how to make that sticky for users, because they realize that if they don’t, ChatGPT or Anthropic or Perplexity or whoever will just eat their lunch, which is not a great place to be.
Dara: I was joking about this earlier, but there is a very real daily existential question of what good is someone who has learned to digest information and make it more accessible to people, in an age where the first thing the machines got really good at, after coding, was that exact style of writing.
But the flip side of that is that when a new regulation comes out, for the analysis of what that new regulation means, the actual lawyers who are my colleagues do a first pass and produce an internal document, and then as often as not, I’m the person who’s responsible for turning that into something that we can then publish on our website. We don’t have bylines, so I’m not saying that this is my product. But that’s what’s then getting slurped up. And there are so many discretionary, political decisions, made, we assume, by human beings, in what the text of law looks like, that I think it’s very easy to assume that you can just skip both of those middle steps and just slurp up the text of the regulation and have what it means come out. And maybe. But A, is that going to be the same way that a human being is going to interpret it when they actually make the policy? And B, how confident are you that you’re not just imposing a version of the world as you wish it were, in terms of the transparency of words?
Here’s an example of how disconnected words and reality are in immigration policy. You can have a work permit to work legally in the United States and not have legal presence in the United States. You can have legal presence in the United States and not have legal status in the United States. That was a really salient issue when young people who had deferred action were trying to get driver’s licenses. Some of those state laws said you had to have lawful status. And deferred action was lawful presence, but it wasn’t lawful status, so they couldn’t get driver’s licenses.
Dave: And it’s not clear that anyone had that in mind when they wrote that law, or that they intended anything like that.
Dara: No. It’s just not like, “Oh, yes, well, this is a very important distinction to draw.” And you can have legal status and not be a legal immigrant, in terms of: you are neither an immigrant nor a non-immigrant in the eyes of United States law. And when the demographers do their counting, a lot of people who had temporary protected status, for example, got counted as unauthorized immigrants, because they weren’t in the immigrant bucket and they weren’t in the non-immigrant bucket. The unauthorized bucket is what remained, even though they were in active compliance with federal authorities.
So I use this not just as an example of how immigration law is absolutely insane and contrary to common sense, but also as an example of how you have to have a certain understanding of that, and a certain willingness to accept that the connection between words and logic is a little attenuated, that it’s subject to interpretation, and that that interpretation is subject to power dynamics, in order to do anything in my field. And that makes it awfully hard to imagine how that seamlessly gets automated, because every step of the process is built on the assumption that a human being is making a decision.
Dave: Yeah. Or put it another way: let’s say we have a model that is capable enough to make those determinations as well as (which is a huge hand wave of a phrase there) a human being. We’re still shifting where the locus of power goes in a really fundamental way. And maybe that’s good, maybe that’s bad, but certainly it’s a thing that wasn’t contemplated. I don’t think anyone, in a serious way, when they were writing immigration law in 1990, was like, “Well, of course, in the year 2026, we’re going to have near-AGI-level models and they’re going to be able to do all these things, and we should really think in advance about how the balance of power might shift if we end up suddenly having much more ability to rapidly process things on the executive branch side by using an LLM instead of a human.”
Dara: Yeah. This is something that I talked around a little bit earlier, but I think it’s worth talking through. There are a lot of ways in which federal statute empowers low-level adjudicators to make value determinations. This is not necessarily a low-level adjudicator, but it’s the example that comes to mind. If you’ve been in the US for more than 10 years, and you don’t have a criminal record, and you have a US citizen immediate family member, like a child or spouse, and you’re in removal proceedings, in front of an immigration judge, and it’s possible you can be deported, there’s a thing that you can seek called cancellation of removal, which is essentially the backdoor to get legal that everybody thinks exists. Everybody thinks, “Oh, yeah, if you’re here for a long time and you don’t break any laws, of course you can get legal.” This is the only way to do it, only it’s very, very limited, both in terms of how many people can get it a year and in that you have to demonstrate that your US citizen immediate relative would suffer exceptional and extremely unusual hardship.
For the definition of exceptional and extremely unusual hardship, obviously there’s a body of case law, and it doesn’t work the way that federal case law works, where there are opinions that work their way up. It can do that, but it’s also all under the Department of Justice, so sometimes the attorney general just writes a decision that says, “Actually, the law is this now.” And that trickles down.
Dave: Yeah, they’re judges, but they’re not judges in the way our listeners think.
Dara: Right. There are some cases that obviously do not meet the definition of exceptional and extremely unusual hardship, such as: these kids have known nothing but the United States, they don’t speak English, they’re brilliant and they’ll absolutely fall apart in a school in a foreign country, and this is the only country that they’ve ever known and it’s the only country where they have citizenship. That doesn’t count as exceptional and extremely unusual hardship. But it is absolutely supposed to be in the power of the individual immigration judge to determine whether a case meets that level or not, and it’s just not at all easy to see how you’d get around that without changing the text of the law. And sure, maybe not in this particular regard, but we all agree that there are large swaths of immigration law that need to be rewritten. But then you have to go through Congress for that. There are human beings one way or the other on this.
Dave: Yeah. It seems like one question that we’re going to come across as a policy community (or maybe better to say, as a society) real soon now is: to what extent do we actually really want a human to be in the docket making the decision? And in what instances have we said we really cared about a human making the decision, and then when you actually look at it, you’re like, “Hmm, this human is basically just implementing a deterministic process”? They aren’t really, in a meaningful sense, putting a thumb on the scale or exercising some judgment in a way that we would want, in some dignified way, to have happened. It’s just like, “Yeah, we could have used a computer. We didn’t have a computer, so we put a guy in a robe and said, ‘You’re an administrative law judge now.’”
Dara: There is a lot of disagreement among people in my field as to whether it would be good to have a point system, where if you meet this criterion, then you get five points, and anybody with 50 points gets a green card.
Dave: And Canada basically does this.
Dara: To a certain extent, yeah. There’s a pool of people who couldn’t qualify based on points. But one of the arguments against it is that you do sometimes need to be able to see, okay, the reason that this person was not making any money was because they were going to school. They don’t have the degree yet, but they’re about to get a degree that’s going to make them much more valuable. Or other things where there are mitigating or aggravating circumstances that it is easy to imagine a human being noticing before a point system notices them.
And that’s hard, because the flip side of that is that in order to demonstrate that it’s a sufficiently discretionary process, you do have to have things rejected for opaque reasons. In one of the lawsuits over one of the executive protection programs, the lawyers were filing denials that had come back with no explanation as evidence that you weren’t going to get deferred action just by meeting the stated criteria. There was also this black box policy going on, which clearly meant it was discretion on a case-by-case basis, and therefore in line with the law. But you do have to have a certain amount of opacity, of arbitrariness. You can’t have everything able to be judicially reviewed, for example, because that totally gums up the works. If you have discretion in a system, it does, by necessity, make it a little more opaque and a little more arbitrary. But are you comfortable being confident that you can determine which inputs are relevant enough that you don’t have a human being maybe providing a check on some of those, in terms of mitigating circumstances?
Dave: So one thing I’d love to talk about a little bit is: what is it to be a journalist in the 21st century? I think a lot of people don’t understand what it is when they’re interacting with journalists on a policy level. I think people are pretty good at giving quotes and being interviewed, whatever, in the AI policy space. But I don’t think either folks in AI policy or people more broadly understand the job of a journalist in the correct way. And I know that all the time you and I have conversations (less so now in your current role, but in your previous roles) where I’d be like, “Well, surely you…” And then you’d be like, “That’s never been my job. That’s never been under the control of a reporter. That’s an editor thing or a photo person thing or whatever.” I would love for you to just riff for a couple minutes on what you think people in America get wrong about what journalists do and don’t do, and how we could relate to them better.
Dara: Yeah. So, look, this is in large part a problem that newsrooms bring on themselves. I tend to talk about newsrooms because it’s really an organizational culture thing. There’s a set of norms in the profession, but the institution that’s putting those norms forward is the newsroom, and the organizational dynamics of that matter a lot. When you put one person’s name on a story, the assumption is that that person has authorship of the story in the same way a novelist has authorship over a novel, where it’s their idea, it’s their characters, they had input from some other people, but they ultimately get to make the decision about what direction the story goes. They haven’t designed the cover, but they have input on the cover. None of that is true, as a rule, in newsrooms.
It varies a whole lot based on the size of the organization. I had two kind of unusual experiences, insofar as I was never in a large legacy newsroom and was in somewhat smaller organizations. But even there, there was just this huge disconnect in terms of: Are you responsible for putting in the URL? Are you responsible for selecting a photo? Are you responsible for making sure that we have the rights to that photo? Are you responsible for workshopping a headline and putting in some suggestions? Are you responsible for finding out the search engine optimization of that headline? Even in the two newsroom jobs that I had, there were completely different answers to those questions. And generally, it is not the reporter’s job to do any of that. Generally, it is the reporter’s job to come up with the text of the story, or to record themselves with the audio of the story.
And that happens at the beginning of the process as well. You can’t just file a story on something you want to write. You have to persuade your editor that it’s a good idea. And your editor is also coming back to you all the time with assignments or suggestions, or, “I was at dinner and my friend mentioned this.” And you’re like, “I’ve been trying to explain that to you for five weeks.” So the agency that people think individual reporters have is wildly hypertrophied relative to the way things actually are.
And it’s a special problem in the 21st century, because the byline is on the thing, and that’s the locus that readers have, so journalists are seeing a whole lot more audience feedback than editors are. I think of this especially in terms of some of the culture war fights about what term you use for an unauthorized immigrant in a story. Do you use illegal? Do you use undocumented? A lot of reporters spent years getting a lot of flak from their audience about using terms that the audience deemed dehumanizing. But the people who could actually make the decisions about whether to change the style guide were editors, and not just desk editors, standards editors in some organizations, who weren’t hearing any of that, weren’t seeing any of that. And therefore you have this weird extent to which reporters are getting flak for, and have to publicly defend, or at least be publicly neutral on, the same things that they’re then turning to their editors about and going, “We have to fix this because I’m getting hung out to dry. We have to fix this because my sources will stop talking to me if they’re upset about the way that this goes down.”
That’s not usually about symbolic things, but at one point I was in an internal newsroom conversation about footage that had been obtained by FOIA that showed someone dying, and the family did not want that footage to be published, but they didn’t get the final say. And the fact that that was an after-the-fact conversation about “Should we have done that?” was not ideal, but it is pretty illustrative of the assumption that people who came into journalism in the 20th century have, as a profession: “Well, of course we’re on the side of the powerless against the powerful. That’s the whole definition of it.” And a failure to interrogate that. And that’s a problem on the journalism side, but it’s rarely a problem with the reporters themselves. The reporters themselves are usually the people who are most sensitive to that, because they actually have to be dealing with sources.
I was just talking to a reporter a couple of weeks ago who got told by her editor in the late stages of the process that her source was not allowed to be anonymous. That happens. I’ve been docked substantially internally because I promised, and did not give in on, not including a subject’s last name in a story. I knew people who had stories killed that were very well developed, because the assumption is if you have nothing to fear, you have nothing to hide. And that assumption somehow hasn’t been interrogated yet, or is being interrogated entirely too slowly. And in the meantime, the people who are actually trying to do right by everybody are caught in the middle.
Ninety percent of the problems you think are problems with reporters are really problems with editors, is the oversimplified way to put this. But if you are dealing with a reporter and you are dealing with some things that should be stonewalls, or some things where you’ve worked with a reporter enough that it should be obvious to them and they don’t appear to be, it’s rarely them that you’re dealing with. Had I been in a major newsroom, I shudder to think how often I would have looked at the final copy of something and said, “I understand why this got reworded. I’m sure it got reworked so that it sounded less jargony. The problem is you’ve just stated something that’s wrong.” Because words have different meanings in immigration policy. Luckily, in small newsrooms, I didn’t have any firepower left for “Oh, I really like that sentence,” but I had enough firepower for “No, we can’t accidentally say things that are wrong.” But that does happen in newsrooms where not only are you not the last person touching the copy, you might not even be the next-to-last person touching the copy. You might not even be the third-to-last person touching the copy. And it means that there’s just so much of a gap between the name on it and the actual agency that reporters often retreat into the professional norms of, “Well, I don’t have to be accountable to you.”
Dave: Yeah. And one thing this is making me realize: on Bloomberg’s articles, at least at one point (I think they still do, but certainly at one point), for at least a subset of their articles, they would have the contact info for the reporter, but then also the contact info for the editor. And I always assumed that that was kind of condescending, that reporters wouldn’t like that and would be annoyed that they could be called to account in that way. But actually, probably the reporters like that, because it causes their editors to get yelled at about some stuff that they shouldn’t be yelled at about.
Dara: Yeah. You’re more likely to see that on really fancy pieces, where it’s like, “Oh, if you’re going to be on our Pulitzer nomination, then we’ll put your name on the page.” And in fairness, an editor does have a really big role in shaping large projects too. But day-to-day, for the reporter whose name is on the story, there’s a less than 50% chance that they were the ones who first had the idea for this story. That disclaimer is never up there.
Dave: Yeah. That is a useful frame that I had not really thought about. Okay. So, Dara, one question we’re trying to ask all of our guests in this series is: what’s advice that you wish you could give your younger self in terms of thinking about DC and policy broadly?
Dara: I mean, the problem here is that when I was in my 20s, I was like, “There isn’t anyone who’s 10 years older than I am who has a job that I want and has gotten it through a replicable career path.” Because the people I saw who had jobs that looked similar to jobs that I wanted had gotten there through weird means. And then I became one of those people. I had this whole career U-turn into and then out of journalism. And it’s awfully hard to imagine a world where I’m where I am now without that kind of massive exchange of economic capital into social capital, and then social capital back into economic capital. So it’s just hard. I don’t know that I have any advice that, if I’d taken it, would actually have left me super-duper better off. I think that I probably could have spent less time agonizing over whether to go to the Hill or go to grad school. But the only reason I was able to say that is because a third option opened up at exactly the time when I was running out of runway on not doing either of the other two. So I guess just the contingency of history, question mark.
Actually, this is something that probably is an elder millennial problem, because we came into a job market that was over-determined by the Great Recession, and so it’s been very easy for us to see ourselves as the victims of circumstance, and I’m sure that’s also true for people coming into the workforce now. But that is a really terrible way to live your life. I often say that nobody wants on their tombstone, “Well, in fairness, it was a really hard time to be alive.” You want to actually do things. You want to have agency. You want to have mastery over your life. And that’s fair, and so in general, it’s not good to just be like, “Yeah, vicissitudes of history.” But also, contingency means that you cannot obviously predict the opportunities that are going to come up, and you can’t be sure that those will not come up for you in the future.
Dave: Yeah. Probably a really important lesson for many of our friends who are in this workforce in this exceptional time. Dara, I’ve really appreciated this conversation. You and I have talked about a lot of these things before, and I still feel like I learned so much from this that we’ve never had a chance to talk about before. If people want to follow up with you on this or on anything else, where can they find you?
Dara: Anyone sufficiently motivated can probably find ways to reach me. But the easiest way is probably Signal. My Signal username is daralind.18. Yes, of course I picked 18. Any Jew who has to pick an arbitrary number, that’s the first arbitrary number we’ll pick. But there’s that. Bluesky is my primary social media platform. I technically still have a formerly-known-as-Twitter account, but I never log on to it. I haven’t logged on to Facebook since my wedding. And I will also happily hand out my email address to anyone who asks, for people who are into longer-form communication than that. But if people think that they have thoughts to share with me or want to discuss things, this is a group of people where I suspect I would learn at least as much from you as you would from me, so please hit me up.
Dave: Awesome. And I’ll also add, of course, that if people do find you on Bluesky, they also get amazing photos of your cat, Penny. And on behalf of Palisade, formally, I want to say that we endorse Penny, and we think she can do no wrong.
Dara: She finds this wholly unnecessary because she is totally convinced of it. But I look forward to seeing whether you continue to say this after cat-sitting her in the near future.
Dave: Absolutely. I’m very excited for that. So again, thank you so much, Dara, and we’ll catch you again soon on the Palisade Podcast.