B (2:40)

Yeah, great question. So, you know, just to sort of set the scene for, for your listeners, I was at the Office of Science and Technology Policy, which has no formal power, really doesn't have control over any budget, doesn't exercise control over any particularly relevant choke points in the policymaking process. And traditionally the Office of Science and Technology Policy on tech related issues kind of ends up being like the little brother to the National Security Council, which has way more staff, usually significantly more hard power, et cetera, et cetera. And so this is like the fundamental predicament that everyone who's ever worked in OSTP faces. So I did a couple things. I spoke to people like you. I also identified people from former OSTP staffers from every administration going back to Clinton too. That's as far back as I could get. And I talked to them about their various different experiences. And I think that, you know, one of the things that was a consistent message I heard, and that is 100% true, maybe even more true in this administration than others, is that people don't care about like job titles as much in the White House. It's a pretty organic. It's not to say that it's not a hierarchical place. It's totally a hierarchical place, sure. But it's not as though, you know, there's like, for example, one thing that I assumed intuitively would be really important going in is there's this special assistant to the President, deputy Assistant to the President, assistant to the President. These are like these different ranks that the senior staff have. And I was like, oh, that's going to be important. I'm going to have to get promoted to SAP in order for anyone I'm going to have to be the special assistant for AI. Turns out, nope, not at all. No one cares. And the way basically that you gain effectiveness, gain the ability to be effective is not actually that dissimilar from what I experience in the private sector, which is to say it's like just be the guy that people want to have in the room because they think you're going to say something that will be helpful to the conversation. Especially on a technical issue like AI, where lots of people are genuinely pre paradigmatic about this topic. They don't have lots of inbuilt abstractions and ways of thinking about it. And so it kind of just ended up being an extreme willingness to be very outgoing, meeting with people, especially going to agencies, visiting in person. In fact, a former guest of yours gave me this advice to go visit in person all the different agencies that I thought I'd be working with the most. That ended up being enormously useful.

A (5:24)

Excellent. That's always good to hear. I've got tons of questions for you about all of this, but say more about the hierarchy piece because it's interesting to hear. The formal kind of assistant to the President you are is relatively unimportant. So which parts of the hierarchy, explicit or implicit, are the important parts in the White House?

B (5:45)

So, I mean, the things that matter, first of all, like, assistant to the President is the highest ranking one. It goes in order from lowest to highest. It's special Deputy and assistant.

A (5:55)

Okay.

B (5:56)

And assistant to the President matters because there aren't that many of those. Right. Like, my boss, Michael Kratzios was assistant to the President for Science and Technology. The Secretary Rubio, the head of the National Security Council is, well, the President is the chair of the National Security Council, but the acting, you know, is the assistant to the President for National Security affairs. So those are like very, very important, high ranking roles. But then once you get below that and you start getting into like, the policy process of like, the sociological reality of how policy gets developed inside the White House and in the interagency process, what that comes down to is more like who is in the room, who knows about this, who is looped in on this, and who is like actually invited into the conversation. And you know how government is. Plausibly, everyone can be invited to the conversation. Right? Like, plausibly everybody, especially, you know, a topic as, as broad ranging as AI, like tons of people can be invited.

A (6:53)

I don't, I don't know if this word gets used outside of dc. I've only ever heard it used in dc. But if you have equities in a topic, equities, which, which in other contexts you might just say, like, if you have a stake in this topic, you come here. But in D.C. it's, oh, DOJ hasn't has equities on this topic, so we need to bring them into the fold.

B (7:11)

I found it so amusing too, because it sounds like a kind of lefty word to me. And yet totally it was used throughout the Trump administration.

A (7:18)

Everybody says it.

B (7:19)

Yeah, yeah, yeah. Like very, very like through and through MAGA people saying, like, our department has equities in this. And it's like that's kind of amusing, but. Yeah, no, I think that, like, very oftentimes, like, the, the way that power actually works is somewhat more organic and somewhat more connected to, like, who do we think is actually going to create value? Who knows what they're talking about here? Because that's the most important thing is, like, everything in the White House, everything, especially the White House government in general, operates on, like, such incredibly short timescales. The White House chief of Staff's event Horizon is like 10 days, and anything beyond that is, like, impossibly far away. So it's like when those people need information and they need, like, counsel, they need it now. And so it's like, who can deliver that?

A (8:03)

Careful listeners of Statecraft will know that. We've had other folks from the Office of Science and Technology Policy on in the past, most prominently before he was confirmed. We had your boss, Michael Kratzios last fall. And we've also had Tom Kalil, who was the number two in the, in the ostp in the Obama Admin. And actually, if we, if we go back far enough to Tom had the initial idea for Statecraft for this podcast. So I have to shout him out there. The Obama admin ostp, as I understand it, was really organized around certain principles. And we've done an episode with Tom about the whiteboard that they had in their office, and that was these are the principles by which OSTP is run. And it's things like own the paper. You know, whoever owns the documentation for the project is really the person in charge. Always be trying to solve other people's problems for them. Be the person who's kind of, like you said, first comes to mind when someone else in the White House has a problem in your area of expertise. If you think about the OSDP you served in, what are the kinds of organizing principles that come to mind?

B (9:09)

Well, it's funny because I kind of saw this transition over time. I think those organizing principles would not be bad ones. And in that sense, you know, my, my former boss and your former guest, Michael Kratios, is a very intelligent, savvy person, because one of the things that happened early in the administration, day two or three, is OSTP got tasked with the lead role in authoring the AI action plan. So we, in governmental terms, held the pen. And for the vast majority of that process, I personally held the pen. So the person who, in other words, the reason that's powerful is that you have an interagency process and you get like a bunch of feedback. Every AGENC agency says blah Blah, blah, blah, blah. No, not this, this change, this wording, et cetera. Sometimes those bits of feedback are just like purely good, useful things that make the document better. Other times they conflict with what other agencies want. And other times this is in fact the policy debate. Is this conflict. The person who adjudicates what happens there is the person who holds the pen. Fundamentally that's just like, is a mechanical reality. That's just true. So that is definitely one thing. I would also say that I think the idea of making other people's lives easier very, very real as a good strategy in general, I think, in life, but specifically for ostp, typically in the dynamic of most administrations, as I said, is that OSTP is kind of the little brother to the National Security Council. And so one common strategy that I heard from a lot of people was like, look, be friends with your counterparts at the National. Don't view them as rivals at the National Security Council and take stuff off their plate. Because in most administrations, as a practical matter, like in the Biden administration, the National Security Council runs the country, right? Like they're literally doing like all, they're leading all the policy processes or the vast majority of the policy processes that matter. And they're also like the ones who are making sure that our ships, our Navy vessels aren't like running into each other, you know, in the south, in the Red Sea or something. I don't know. But like all that was very true and it served me well in the first month or so. Something that happened that is very contingent in particular with this administration is that in late May, early June, Secretary Rubio completely reorganized the National Security Council, right? This ended up being like, it was a very serious opportunity for OSTP because not because like NSC was, because we had a very good relationship with NSC throughout. Like, there was never any fighting or turf wars or any, any of the stuff I heard from previous administrations did not happen. We weren't in a turf war. But what happened is that because a lot of that reorganization from Secretary Rubio did focus on like tech related things, a lot of policy processes that the National Security Council had been sharing were just transferred by default to ostp. And anything vaguely AI related was transferred by default to me.

A (12:15)

Can you put some color on that? When you say the policy processes were translated to you, what did you suddenly have to do?

B (12:22)

So like, for example, there's an executive order that's in draft that at some point someone thought it was a good idea to do. And now we are taking the draft of that executive order. And we're sharing it with all the agencies that would have to execute that executive order and refining it and figuring out what it should actually be, or maybe it's not an executive order, some other high priority policy process. So the interesting thing about that though is that OSTP became more powerful within the four walls of the White House and then like maybe actually a little bit weaker in the interagency process.

A (13:00)

How so?

B (13:01)

Well, because like, like if you have like all of the sort of parts of the White House united around a specific thing, then you have a lot of weight behind you. If you have like everyone in the White House aligned, like all the principals within the White House, all the assistants to the President are aligned around something that's like quite a powerful coalition that it's hard for even a cabinet secretary to go, to go up against. It definitely forces a compromise. It's much harder when you just have like, because just like NSC just isn't there and I see is just like not participating in that anymore, then it's like, well, you're the bigger fish in the, you know, somewhat smaller pond still pretty big pond of the White House, but you are maybe actually a somewhat smaller fish in the grand scheme of the government. That was at least a concern I had. And so that is part of why I doubled down. And it's like, really it is a serious sacrifice of time to like travel out to, you know, Gaithersburg, Maryland to go spend the day at NIST when you're a White House staffer is like, it's like anxiety inducing. I think it raises my blood pressure just thinking about it now because you're.

A (14:04)

You'Re outside of the, of the Eisenhower building.

B (14:07)

Yeah, well, I mean, and truly like, this is so hard to travel. Traveling to Gaithersburg is like traveling to Japan as far as I'm concerned. But it occurred to me like, oh, okay, well, this might be the new dynamic that we're settling into. And so I made it a very big priority of mine to have really good relationships with, you know, not just leadership, but career level people at all the agencies that I worked with the most.

A (14:33)

This is getting to something I've been thinking about a bit recently. I've been reading a history of the Nixon presidency, in particular the administrative reforms. And you can see parallels with Nixon and Trump in that both of them thought very consciously about how do I, from my seat as president, exert more control over this massive organization called the federal government. But there's some very interesting distinctions in how they went about It Nixon's instinct was to really staff up the White House and a lot of the kind of modern, these policy councils in the White House are really creatures of the first and the second Nixon term. These big, much, much, much larger staff within the White House. Whereas my impression from the outside, and I want you to tell me if this is true, is that the, the Trump admin dynamic, especially in this second term has been put a lot more responsibility on cabinet level secretaries as opposed to the staff of the White House. Thinking kind of broadly across the, across the administration. Is that, is that a fair way of characterizing this admin? That there's more authority given or expected from people like Chris Wright or Rubio or I don't know the principles?

B (15:50)

Yeah, it's not universally true, but I think it's more true than not probably that that's the way it's going. So I mean, you know, one thing that, you know, just as a basic numerical matter, the Trump 47 White House staff is significantly smaller than for example Biden like Biden's ostp. My understanding is, you know, I don't have exact numbers here, but this is what I've heard from former from Biden OSDP staff said it was like 150 people. You know, OSTP will grow to be sure from where it is, from where it was when I left. But like when I left OSTP we probably had like 25 or 30. And yet I would actually argue that the OSTP led by Michael Kratzios did like considerably more than the Biden ostp. So I think the relationship between staff and like effectiveness, I would, I would, I would say like that's not necessarily like a one to one correlation but, but like in fact I think adding more staff can sometimes create problems, right? Like I think sometimes like especially in a charged place like the White House, you can create turf wars that are more likely to happen when there's more people. Because you know, for example, if OSTP had like a 10 person biosecurity team and a 10 person biotech team, what would those people do? They would probably spend a huge amount of their time fighting with one another over what issue was biotech and what issue was biosecurity. If we had a robotics team and an AI team, who knows, right? Like I think that there's like a, an element of that to it. But it is also the case that when it comes to agency rulemaking and stuff like that, this administration is more hands off as a general matter than for example Biden was like the basic Story that is told publicly and that I have also heard is that things like the export controls on AI chips in the Biden administration were very heavily influenced by staffers in the Eisenhower building, which is to say the White House. You know, that's not, you know, it's not to say that there's no communication anymore. There's, there's certainly collaboration between the Commerce Department and the White House and this administration. But like, generally speaking, it's like, well, that's kind of like Commerce's rule and so like we're gonna let Commerce do it.

A (18:17)

Yeah.

B (18:18)

And I think that, you know, from the President's perspective, I think the president would rather sort of run the government and exercise control over the government through cabinet officials that he knows really well as opposed to because like the staff of the National Security Council alone is above the Dunbar number. Right. That's above the number of relationships that a, a human can, can reasonably manage. Even the number of like directors and senior directors and blah, blah, blah, blah in and of itself is quite large. So I think like the President manages it like he's managing a board of directors, almost.

A (18:56)

Sure. When we had Russ vote on also last year, who's now the, once again the head of the Office of Management and the Budget in the White House, as He was in Trump 45, he talked a lot about this and this has been, I think, a consistent through line. But he said on statecraft, he said it elsewhere, that he's really against the model that the Biden admin. And that he says many Democratic administrations have tended towards of governance in which you have lots of kind of free floating. He calls them czars. Sometimes they're informally called czars, sometimes they're just, sometimes they have other titles. But czars in the White House who are generally responsible for an issue, who kind of sort of overlap with the actual people in a formal chain of command from a cabinet secretary on that issue. And you've seen this at the State Department as well. A lot of the Secretary Rubio led reorganizations are around trying to strip out some of these cross cutting roles on human rights or the environment or labor or what have you, and trying to just run things through the ambassador through an individual direct chain of command in a given place. I'm oversimplifying, but it's interesting to hear you kind of talk a little bit about that because it seems like a managerial philosophy that the head of Trump's Office of Management also shares.

B (20:14)

Mm, yeah, no, I think so. And you, I mean this is one of the stories that I think is most like profoundly underappreciated. And it's the one that makes me like the kind of the angriest that we don't have whole media apparatus filled with. Santi is doing this like hard, you know, like reporting, God forbid reporting about this. Like there actually is like a new philosophy of statecraft that is like being not just like experimented with in Pilot, but like really done at scale in the federal government. And I would say that the results, like, I mean, you know, we don't know. We haven't, we're very early into the administration still.

A (20:53)

I'm always happy to hear it's too early to tell on this podcast. I'm okay with that.

B (20:58)

But like, I would also say that like my experience of being inside it was like extremely pleasant. Like it was not a soul crushing bureaucracy at all. It was a really, really like dynamic and collaborative congenial place to work. I've worked at think tanks that, and universities that had way, way, way more toxic internal climates and political turf wars and internal bureaucracy than the White House.

A (21:24)

That's fascinating. Will you say a little bit more about what you mean by that new model of statecraft? You said the word statecraft, so I have to push on this point.

B (21:33)

Yeah, I didn't mean to, to, to say the name of the pod.

A (21:37)

It's like when, when they say the name of the movie. In the movie you say.

B (21:41)

Yeah, yeah. So I mean I, I think part of it definitely does relate to this kind of managerial philosophy of like, we're going to exercise more direct control and we're going to like, make sure that the lines of information are like very, very clear. Right. There's also, I think, you know, the way that like ppo, the presidential personnel office is going about its task very frequently. Like the, the question. I, I don't want to, you know, talk about the questions they ask you in any detail, but you know, everyone has to do this. They ask you questions. It's, it's famous. It's not just the Trump administration. This is every administration. They ask you questions, you know, about like trying to make sure that frankly, you know, that you're loyal to the President, that you're going to be a faithful agent of the agenda that the President wants to execute. But like the questions were like so clearly influenced by like, I don't know if the people asking them would think of it as the concept of cybernetics, but they were so influenced by cybernetics. It was so like, are you good enough of an Agent that you can anticipate where the President's going to come down on an issue and make a convincing case for why you believe that, like even if you, we know damn well you haven't talked to the President about like, you know, common law liability and AI. Right. Something like this. Right. Like we know that you haven't. But like, what do you think the President would come down on that? You know, that's really interesting. And I also think that in general there's this willingness in the administration to reconsider. I think it's such a fascinating confluence of events. Let me step back for a second. Or just like there's the obvious fact that the president, you know, has a non consecutive two term, which is the first time that's happened in a long time. It's the first time that's happened, particularly in the era when something like the deep state exists, which is an interesting twist. And the deep state is totally real. Right. Like, it's not conspiratorial to say that it's a real thing. Absolutely. Trust me. I got the memos with the policy recommendation where it's like, you have two hours to clear this. Please let us know what you think. There's a mountain of context on this. We're not going to give it to you. Yeah. You know what I mean? That's how the deep state works. That's how it is. And they're kind of doing whatever they want. So that thwarts a lot of presidents and I think it particularly thwarted President trump in. In Trump 45. Right.

A (24:05)

Famously.

B (24:06)

Famously. And now we have this like fascinating dynamic where, you know, four years in the wilderness to contemplate, to think about what went wrong, to reflect on it. And then you come back and you come back with like a very detailed plan. You come back also with the knowledge that you don't have a lot of time, you have a plan, you have not a lot of time and you have not a complete willingness to throw out everything about how government has worked, but a much more than is appreciated willingness to not just do things the way they've been done if there are more efficient ways to do it. So like I've heard about how an idea goes from a staffer's brain to the President's desk in the form of an executive order or another sort of policy document. I've heard about how that worked in the Biden administration and in Trump 45 and in Obama. And I can tell you that the process of getting things to the President's desk shouldn't be easy. It wasn't easy, but it made sense. It wasn't filled with things that struck me as pointless procedural blockers. They were like, well, yeah, okay. Like now, you know, these lawyers have to look at it because, you know, it's like that kind of thing. It's like things that made sense, but there wasn't a lot of turf war and pointlessness. Maybe it's also a new administration that's also possible.

A (25:31)

There are these learning cycles in politics, just like how people talk about. You learn the lessons of the last war. When people say that about militaries, you know, overlearning the lessons of the last war, they mean it derogatorily, but you see it in a. In a good way in D.C. often. And I think clearly there was a lot of learning done among Trump folks from the first term. And I also see now lots of folks in D.C. from the last administration, from the Biden admin paying very close attention, I think, to the way this admin has managed to move much faster. I was just having conversations this morning with people who were talking very bluntly about the ways that the Biden admin didn't have as strong a sense of the shot clock, as they put it, as maybe it should have had. And again, all these claims are generalizations. Everybody knows you have a limited amount of time in the White House before you're out one way or another. But it was striking to hear a lot of folks reflecting in retrospect. We could have been much more aware that every day that passes is like one day fewer to set the president's agenda. It's striking how everybody's watching how the other guy does it and trying to adjust. Oh, well, we should have done it that way. We'll do that next time.

B (26:46)

Yeah, I think that's right. And I think to some extent it's the lessons learned. And I think another part of it is just culture. Right. There's a lot of people who maybe they would define the mission somewhat differently if you really got down to it and talked to them each for an hour, but have a very clear sense of we are here to do a job and we have to get this done. And we're trying to affect genuinely transformative change.

A (27:14)

Sure.

B (27:15)

And also like that realization that just the election alone is not nearly enough.

A (27:21)

Yeah.

B (27:22)

That does not itself get you transformative change. All those are lessons that, like, I think that Republicans right now feel viscerally, in a way that's very hard for Democrats because they've been the recipients of the cultural Grace for the last 20 or 30 years. So they get the cultural benefit of the doubt, and they get the political benefit of the doubt. When Republicans violate the const. Or, you know, push against the Constitution in various ways, it is called tyranny. And when Democrats do it, it's called ambition and creativity. It's called statecraft when Democrats do it, and when Republicans do it, it's evil. And, you know, I think that that's. Well, yeah, I think that's just sort of like something that all Republicans are quite aware of.

A (28:10)

Sure.

B (28:10)

And so it affects, like, it affects the culture. I'm telling you. Like, it was cool. It was just fun. It was like. Like being in the Eisenhower building in the spring and early summer of 2025 was like being at, you know, the punk rock club CBGB in lower Manhattan in, like, 1974, you know, like, it was like, man, like, oh, this guy's doing crazy stuff over here. It was like a scene, you know, it's like. And like, all these people are like, there's a lot of really, really smart people with, like, really ambitious ideas, and there's, you know, an entire group of people rewriting the federal acquisition regulation and reducing its length by, like, a gajillion pages, and no one knows about it. There's an executive order. It's all public that this is happening, but nobody really talks about it.

A (28:51)

Yeah.

B (28:52)

In the way that there would be doting pieces about this in the Atlantic if it were a Democrat doing it.

A (29:00)

Let me ask you what you make of one concern one might have about the model of statecraft and the Trump admin. And I'm not sure where I fall on this, but reading about Nixon 2, and there's a book called the Plot that Failed about Nixonian attempts to get a hold of the federal government and the successes and failures there. One concern I think people have, policy agnostic, whether or not they're partisans of one type or another. One concern you could have about a managerial system that runs along direct lines of command and control with principals at the top who are empowered to make lots of calls and to actively oversee people down the chain is that lots of things get bottlenecked at the principal level. So, for instance, a month or so ago, there was reporting that Secretary Lutnick, Secretary of Commerce, had several thousand contracts on his desk waiting for his approval for the Department of Commerce to purchase various procurements. And I don't know. I don't remember exactly if it was printer paper and pens and that sort of thing, but it was, I believe, it was contracts, anything above something like $500, which, for an enterprise, enterprise as large as the Department of Commerce, lots of things that are just totally mundane, you know, cross that threshold. But there was reporting that, that requirement that the secretary sign off on all these very small transactions, by the department standards, we're holding up all kinds of processes downstream. That's the sort of thing you can imagine happening in a bunch of other levels that by trying to centralize and directly create a hierarchy for a lot of decisions that weren't previously made. In that hierarchy, you create these kind of roadblocks at the top, where, of course, the secretary doesn't have time to sign off on all this stuff. But the result is lethargy institutionally. What do you make of that set of concerns?

B (30:54)

I think that's a very real failure mode of this style of management, and not just in. I mean, I think that can be just as true in a. In a corporation. I think the, like, the interesting way that this gets balanced is that, you know, the. The president is not a micromanager. He wants to have the direct line of. He wants to trust that if he wants something to happen, it will happen. Right. And everyone in the administration completely gets that. But he's also, like, once he feels comfortable that the right people are in place and that he's sort of figured out the kind of goals that he wants to articulate, he's not necessarily going to micromanage things. And so, you know, in a lot of ways, it, like, there isn't that kind of procedural blocker because not everything gets bottlenecked.

A (32:01)

But that's at the very top of the system. I mean, I totally take your point. President Trump is not a micromanager. I don't think that's many people's impression of him. Yeah, but you can still have a blocker on. You know, the Department of Commerce is an incredibly large.

B (32:14)

Yeah, yeah, sure, yeah.

A (32:15)

Like, you can go down several levels and still have these massive structures that might have to be gate kept on individual political appointees and, and different Cabinet.

B (32:25)

Secretaries have, like, very different approaches to this, that there are cabinet secretaries that are, like, total micromanagers, and there are others that are, like, quite willing to delegate important things to. To staff. I think there's a way to do that. I think. I think it is true that this sort of. Let's put it this way, I would say that pairing this style of management with a micromanager is a very likely way to create this failure case. And the problem also is not just without speaking about any specific department or anything, just in general things I've observed. It's not just that style of management bottlenecks things at the very top. It also creates an incentive for, like, the direct deputies to also be that way. And so it's like not just one bottleneck, it's actually a series of bottlenecks. If the secretary of a department says, I want every thing like this to come across my desk, well, then like, probably the person's deputy who would have typically been responsible for that probably still has an incentive to, like, be a choke point to make sure that, you know, they're like, there's not just random staffers of theirs going directly to the secretary. So that creates, you know, I'd say the micromanagement is a way to really make this problematic. Unless until we're able to, like, make copies of secretaries and with AI agents or something. But maybe one day, maybe one day, then we'll be able to micromanage as much as we want.

A (33:56)

Well, let me change gears here. Most of our listeners are not AI policy wonks, sadly. Maybe they should be. But at this time, the majority are not, I think. So I want to ask you a bunch of nerdy implementation questions about the AI Action plan, which you largely held the pen for. But in order to do that, give us a one minute, your explanation at a high level of what the action plan was for people who are not necessarily in the weeds on AI policy.

B (34:26)

Yeah, so the AI Action plan was created by an executive order the president signed in the first day or two of his administration. Basically, that executive order rescinded the Biden executive order on AI, which was kind of very unpopular in Silicon Valley, and replaced it with a provision that said, go make an AI action plan. So in some important sense, the action plan is kind of a replacement for the Biden executive order. Though it was not itself an executive order. It was more of like a replacement report. But basically the action plan tries to think of, like, all the different conceptual areas or like, themes where America needs to develop a strategy and have a position on AI. So it could be like America exercises a lot of leverage over the scientific enterprise. So maybe a big part of that is like, well, what do we think about AI and science? What do we think about AI and defense? What do we think about AI, and, you know, the infrastructure for AI and the environmental permitting. Right. And then instead of just saying most, most government strategy documents kind of say like, AI is going to be important and there's also going to be risks, and we have got to get the good with the little of the bad. And the bad is going to be defined, you know, however we define it as the government. Right. Either it's misinformation and bias, or maybe it's whatever, any number of things, catastrophic risk. And that's where most of those documents stop. But what's different about the action plan is the action plan is both a strategy document that tries to be quite specific about what America's, at least our government's, strategic objectives are, and then also to identify somewhere between usually like 2 and 6 specific policy actions that federal agencies can take now with existing statutory authorities and existing budget to somehow meaningfully advance the ball on that strategic objective. And so we do this across several dozen strategic objectives with about 90 total policy recommendations. A little more than 90. And that is what really makes a difference, is this emphasis on concrete, actionable recommendations.

A (36:42)

You guys did not run, as I understand it, the formal interagency process in writing the AI action plan say a little bit more about how you actually worked with the other agencies or the interagency. That's another D.C. iSM, by the way. I'm just going to throw that out there. The interagency. I love that people will just refer to the interagency to talk about this.

B (37:02)

Amorphous process concept of other agencies. Right. So again, I assume your readers all have context for this. But like is general matter. When some whippersnapper in the White House gets an idea, that idea is usually executed by someone else and that agency or maybe many other people, or maybe there's many other people who would like to be executing it and aren't because the whipper snapper didn't think of them. And it's like, oh, you didn't know that the State Department also does this and not just the whatever. And so you have to go through a process of communicating what you're doing to those people and then figuring out, like, what it should actually say from the perspective of the people that have to implement it. Fundamentally, it is a good process and is a necessary process. It is much maligned in D.C. because it's where a lot of infighting and whatever else happens.

A (37:51)

Many guests of statecraft, I will say, have maligned it.

B (37:53)

Yes. And so like we ran interagency processes, for example, all the executive order that were signed associated with the action plan, those went through interagency processes that were certainly like, more efficient. I mean, you know, our executive orders didn't take like, they didn't take like months and months and months to, to go from A draft to being on the president's desk. And they didn't end up being like 50 pages long. Like, all the Biden executive orders are like, short, sweet, and to the point. You know, they were written by, by a writer. So there's, there's that. But, you know, the action plan itself, though, I had a lot of latitude to kind of design the process for clearance is the word. That's what you seek from the agency. That is consent, approval, clearance.

A (38:39)

And that's different from security clearance.

B (38:42)

That's right, yes. I'm sorry, I was referring to the concept of approval, not of a security clearance. So the action plan is weird because it's like, it's like a very strong recommendation, but it is not literally. You know, there's no shells in the action plan. The action plan is a should type document. So I had a lot of latitude to design the process by which we would take this thing from a word document to the President. And what I ultimately decided was this thing is going to be hugely wide ranging. It's going to touch on like a million different issues. And we can't have the entire federal government commenting on every aspect of the federal government strategy because that might lead to, if there's some random staffer anywhere, some career staffer who's just grumpy and has been there for 30 years, might just like, throw in a bunch of comments. And, you know, you don't necessarily, as the White House staffer, have context for, like, who they are. Also, it increases the odds of leaking it, so it increases the odds of a long, drawn out process with more interference, et cetera, et cetera. So what I did was I wrote the plan, and then you can actually basically see exactly how it would have worked out. I took the action plan is organized with three pillars, and then under each of those pillars, there's roughly a dozen per pillar, strategic objectives that are in bold text. And then there's a paragraph, and then there's the actual bullet points with the recommended policy actions. I would take those bullet points and the recommended policy actions, and I would, you know, either going by the agencies that were named in them or by just thinking through myself, like, who should have the equities? Who has equities here that's not necessarily named, and then I would just send just those bullets alone with the paragraph of text to the agency. Now, if you're the Department of Commerce, that means you got pretty much the whole plan from the beginning, because the Department of Commerce is mentioned a lot. One other rule that I had was that, like, if there's a series of five bullet points and one agency is mentioned in one of them, then I did give that agency all the bullet points, all five. So that way they could see, you know, the other things that were part of this strategic objective.

A (41:02)

Sure.

B (41:02)

And then we went back and forth on that. So it's kind of like the action plan was really more like, like 20 or 25 inner, or maybe like maybe 40, like interagency processes that were run in parallel. And then at the same time, we did sort of paper interagency process within the White House itself. So within the White House itself, like every. All the components of the White House got the whole plan, like a month or so in advance. And so White House was able to offer complete feedback. And, you know, the other thing I did that relates to this, that made. Made me feel more comfortable at least, is I actually, I would use AI to like, simulate interagency processes, like for. For different policy objectives. So you kind of like, try to get like, okay, what is probably like the first pass, at least going to be of like, pushback or advice or whatever that I get. Maybe try to incorporate some of that. At least be aware of it. Right. Be aware of it. So that when the sort of wizened career person at the National Science foundation or wherever else comes back to you and says, well, blah, blah, blah, blah, blah, this isn't feasible. You can at least have like, anticipated their criticism and be able. Or like their feedback and be able to have a constructive conversation with them. So that was like, basically how we did it. And then it all just came together. And then at the very last minute, agencies got like, the whole thing, but it was really like they were only ever asked to clear the bullets relevant to them.

A (42:27)

That's fascinating. Okay, this gets me to a topic that I want to spend some time on here, which is the implementation of the AI Action Plan. As we've already discussed, OSTP has very little formal power. And you've said that past OSTP folks who've come on this podcast have said it as well. Everybody understands OSDP does not get a lot of shalls, does not get to go to agencies and say, you shall do this. It's reliant on these relationships and political cover and. And all the other soft pieces of the political process. Right. So the AI Action Plan is, I think, notably, for a strategy document from the White House, quite detailed, quite full recommendations. But these recommendations both depend on a lot of state capacity to implement and execute, and they depend on, as you said, the Agency being a fully motivated and willing partner. Because classically, bureaucratically, it's typically the case that agencies don't follow guidance just because somebody in the executive office of the President wanted them to. So I've got more specific questions for you, but at a high level. How should we think about the implementation challenges of something like the AI action plan?

B (43:42)

I mean, you know, it's the hard part, right? It would be like a very difficult part of it. Writing it was hard enough, but it would be, you know, it's the hard part is the implementation. So I'd say like a couple things about that. First of all, the bulk of time that the action plan spent in draft, you know, while I was at the White House, was mostly done in a form that resembles like pretty close to where it came out publicly. Not exactly, but like 75%, 67%, 75% similar in like late April, early May. And then it was like two and a half months of working with my counterparts at agencies to get them. I mean, sometimes it was just like, okay, what exactly should they say? But other times it was like, let's make the case. Like, let's get people bought in here. Let's get the leadership of this agency bought in. Let me explain why. Let me go in person and visit. Right. All that happened. So that helped. I think these ideas were very fully baked. There was nobody in the federal government who was surprised. Well, not to say nobody, but no one in the leadership of any federal agency was like calling the West Wing, being surprised that their agency was asked to do something because of the action plan never happened. And on top of that, I think frankly having like a good rollout strategy helped a lot too.

A (45:02)

How so?

B (45:03)

Well, I mean, the action plan event was very carefully, you know, worked super hard on it. Well, a lot of people did. You know, we have, there are five cabinet secretaries, the vice President and the President announcing it. It would be hard for the administration to like emphasize Parter or to emphasize more that it cares about this issue than that kind of presence, that kind.

A (45:25)

Of star power, and that matters. And I'll just, I'll just say like political cover. If there's one lesson I've taken away from Statecraft that's just kind of this broad cross cutting lesson. It's political cover matters. Having the principal stand up and say something matters, even if it's not a shall, even if it's a should.

B (45:45)

Oh yeah, no, no, that is very true. And then the last thing is that we were lucky in that the action plan was well Received, which was not part of my model of the world. I did not think it would be. I thought it would be way more criticized.

A (46:02)

Really? What kinds of criticism did you expect to get?

B (46:06)

I mean, I expected to be criticized in every way. I expected it to be criticized for being too concerned with risks, not nearly enough concerned with risk, concerned about the wrong risks. I figured people would throw whatever they could at us. Not just because it's the president, you know, not because of like the content of it, because it's the president. And that it would be broadly characterized as a reckless plan. I mean, and in fact, like, if you go read like the New York Times or the Washington Post, what they'll basically tell you is it's a reckless plan that doesn't care about safety or.

A (46:35)

You know, that wasn't my impression, to be clear. I think the Washington Post actually in the, in the mainstream media, surprisingly to.

B (46:42)

Me, they had a good, they had a good op ed and it was reasonably well received, you're right. But like, I do think you can still go look at like reporter shorthand and follow up stories and like, I've seen things where, like the administration's action plan, which is reckless and also is a, a sign of, you know, turning away from the world order and focusing, you know, turning inward. Right. I mean, that's what they said.

A (47:02)

The New York, the New York Times did, did do a lot of this.

B (47:05)

I will say that's. I think maybe what I meant was MIT, not WaPo. So my apologies, but no, no, no.

A (47:11)

I'm, I'm weirdly obsessive about media, having been a reporter in a past life.

B (47:16)

One other thing, for any younger people who are just interested about media landscape, everyone totally still reads the New York Times, so, you know, I do.

A (47:25)

Everyone.

B (47:25)

Oh yeah. But what I mean is in the, like in the Trump White House. Oh, that's funny. Yeah, yeah, yeah. Like, I mean, it's not like we were obsessed, but like we cared and we noticed. Anyway, I expected it to be poorly received and maligned, but it wasn't. And so that ended up helping too because it made it like prestigious within the administration. And so, you know, the last like two or three weeks that I was there, it's like a market change in the number of people who are like, oh, can we like get OSTP's take on this? We really want to do the action plan. We're really excited about implementing the action plan. And you know, a lot of agencies will do things that are not necessarily directly called for in the action plan, but are certainly related to it. And they'll say, as the action plan says, blah, blah, blah, blah. And that to me, you know, I've. We've already seen that happen. I suspect we'll see more of that. That is exactly what I wanted. That is exactly what I was hoping for, is that this would be a rallying cry, not just for agencies, but like I hope people in the, I mean we actually, it's already happened in the private sector too. Some probably it's just flattery, but like the private sector has done things and they've said like, yeah, actually we were like this, the action plan says this and we think we're aligned with the action plan here. That's the idea. That's what government should do. That's what the leadership part of this is.

A (48:44)

So I guess my questions about implementation fall into two kinds of questions. One is how do you get an agency that isn't formally required to do a thing to try and do a thing? And your answer is combination of careful work from folks at OSDP preceding the rollout, the actual rollout itself, the salutary reception. It got POTUS making a big deal about the AI action. All these things make it easier for agencies to come alongside you. Right? Like that's one, that's one piece of it.

B (49:14)

Yeah.

A (49:14)

There's another question which is just about internal capacity and capability to implement something even if an agency wants to do it. So like let me give an example. In the action plan, the intelligence community is tasked with monitoring the capabilities of Chinese AI labs, Frontier labs. I think that's fantastic. I think it's very, very necessary. But making that happen, working with the Director of National Intelligence to coordinate the ic, which is legendarily hard to coordinate, as we've talked about on here, holding the intelligence community to account for that, building the process by which that happens, and that information gets back into the White House and feeds into the OSTP's decision making on AI. All those things are easier said than done. And they take people and capacity and, you know, some amount of literal staff, even if it's not the size of the Biden admin. I take your point that.

B (50:11)

No, you're right.

A (50:12)

Number of people is not a direct correlate for capacity, but it has some relationship, I think, clearly.

B (50:17)

Yes, yes.

A (50:18)

Let's take that example of getting the intelligence community to monitor the capabilities of Chinese labs and passing that information back with no lossiness to decision makers who have to make really difficult decisions about AI. In light of developments in Frontier Labs in China, how is that going to get implemented?

B (50:37)

Yeah, great question. So I think the parts of my answer that I emphasized probably solipsistically to some extent, were the parts that I myself feel like I am good at. Like I'm fundamentally a communicator. And so that's kind of like, you know, where I emphasize. But you are right that like there is this ongoing role that's not just within the agency, but also, you know, at OSTP of like, how do we actually see to it that these things happen? To use your example of the, the ic, that's like one of the, one of the very hardest ones, I would argue. I think what you need there, and this is, we're now getting into parts of the reasons why, you know, I, I made the decision to leave, is that my feeling is that you don't really need, you need someone who understands AI well to do that. But like me knowing what like a KV cache is, is like not like useful, which is a technical term in machine learning. I'm sorry, but me knowing like the architecture of the transformer or something is not like especially useful to me knowing how to get the intelligence agency to do something. And I'll tell you the truth, like, I don't, that's not my area, that's not my comparative advantage. I think, you know, where, where OSTP is, you know, what I've counseled at least is like go in the direction of implementers, right? Go in the direction of people that'll actually like that know their way around. I think that's what you need. You need people that know their way around the, you know, the relevant agencies that they're going to be working with. The good news is OSTP already has people like that. And so you need people with like relationships. You need people that have established trust and certainly with the intelligence community, you know, I only barely, you know, got to work with them. So that's, that just was, wasn't ever going to be me. And so, you know, you kind of have to know when to step aside to that extent. But yeah, like, I think how you actually do it is that gets you back to this, this combination of like the things that are in the actual plan have got to continue to remain important to enough of the relevant people in the White House. If the White House cares, it'll get done. Because like, this is the sort of thing where, like, if, if, if there's some blocker and it's not happening, then you have to identify what the blocker is. You have to try to resolve it. As I Would say through the front door. So you try to resolve it yourself with the agency and that doesn't work. Then like to go back to the point you made. Top cover is key. That's how things get done. Right. Like it's a call to the chief of staff. That's how it gets done.

A (53:16)

But this is one place where I guess I tend to take the view that staff having more bodies on this stuff does make it actively easier to do implementation. Because it's just really hard for a principal, someone like your boss or like Secretary Lutnick or like Rubio or you pick your principal in the admin to ride a herd on the intelligence community to make sure, you know, to get bi weekly reports to make sure that we're actually implementing this kind of new kind of surveillance the way we want. All that stuff, I mean, just mechanically in any kind of institution require a whole stack of people who are good at carrying out the principal's wishes. Right. It's not even just about political cover. It's not just that. Secretary Lutnick says, I want this done. There's all kinds of resistances and cross pressures within an admin or within a specific agency.

B (54:09)

Even OSTP is going to have to staff up. That is undoubtedly true. That's undoubtedly true. And some of it, like, you know, I think a big chunk of it in practice is gonna, there's gonna be turf wars that you're gonna have to adjudicate, you know, that you can kind of punt on and text, but that at some point are gonna be real. You're gonna have to adjudicate on those things. And, you know, I think all that. Yeah, I mean it is, it is definitely going to require staff. It will require something more than that. It will require like skillfulness and, you know, good relationships and, and all that. You know, it's also like one thing that I, I would love to get, you know, I'd love to query the LLM of your guest's knowledge on this question, which is like, because I don't know the answer, like, you know, what is it like for an administration to age over time? Because it occurs to me that I worked in the administration during a time when it was really quite young.

A (55:08)

Yeah.

B (55:08)

But that, you know, as things develop, like people end up occupying whatever their choke point's going to be, they sort of like their soft power maybe like hardens a little tiny bit and then they're, they are where they are and they're not going anywhere. That could have been me. Right. I could have stayed and like been like a kind of not. I wouldn't have characterized myself as a choke point, but I would have been something, you know, and maybe there'd be some other new person with a great idea that, you know, I don't want to listen to because I view him as competition in two years. Right, sure. So I think it's actually kind of an interesting question of like how do these things harden over time? And I don't quite know the answer.

A (55:42)

Totally. This is one of those areas just to cut in where I don't think there's a abstract correct managerial answer. Should you have lots of turnover in an administration or should you have real continuity of bodies of people? I have spoken to many people in this admin who have had conceptually the model that I think you have of I'm going to go in, I'm going to execute the task for which I am best suited and I've got some sort of end date in mind or I will try and know when my comparative advantage has dried up and it's time for me to go out. And I noticed that quite a bit in this admin and I think you could contrast that The Biden admin which famously in right wing circles didn't fire any principles. Right. There was remarkably little turnover at high levels within the Biden admin compared to many other admins and they touted that as a good thing. Right. We have continuity, we build these relationships over time. There's a collegiality here and I think genuinely from without knowing the details of those individual processes, there's not some magic managerial answer about which of those models is the right one. But I think they each obviously have their own failure modes.

B (56:54)

Yeah, I think that's right. But yeah, no, I wish I had something like, I mean, I would say that like the one other thing that's really useful in government is a forcing function. Forcing functions are great. That's why the provisions of the EOs that get implemented are the ones where you attach deadlines. Even better than a deadline is an event where the President is speaking. Event where the President is speaking is a. The you can't. That it has to be ready must.

A (57:22)

Be because the President is going to talk about it. And woe to you if the President talks about your thing and it's not.

B (57:27)

Ready or it sucks. Yeah, like, and, and, and those aren't the only types of forcing functions in the world, but they're two good ones. But there's lots and lots of others that you can be creative about. So I think in some sense, like, I think a very creative. And this is not me. Right. I was good at some aspects of this, but maybe I would have gotten good at. I don't know. But anyway, I think like, one aspect of like effectively implementing this is actually going to be creating those correct forcing functions where like, you know, it could be. There's one thing where we call out, for example, for the creation of a new kind of facility for the, for the DoD to test, to test autonomous vehicles and drones and things like this. Yeah, like an autonomy proving ground. Essentially that's a physical facility or a data center. A high security data center is another one. That's a very tough, that's a very tough one with like, not just like political execution risk, but like technical execution risk on that. And maybe your forcing function is the President or Secretary Hegseth goes and visits, you know, they're going to visit on this day, you know, like. Yeah, it's that kind of thing. You, you have to, if you get the principle personally invested and, or ideally and you create some sort of a forcing function, you can create a sense of urgency that might not otherwise be there. And a sense of urgency is the only way to make anything happen.

A (58:49)

Sure. Let me ask you one more question on this and then we can pivot. But to your point about the security of data centers, which is something that the action plan talks about and that was also mentioned in the old the Biden admin AI roadmap, this is not the most controversial piece of the action plan. I think, I think generally there's interest in data center security so that rogue nation states or great powers can't hack into our precious data centers. But even something as kind of technical as that, there's lots of debate, there's lots of different views on how robust that security should be, on who should be implementing it, who should be, you know, checking that. Is that not the sort of thing where a Dean Ball on the inside, who wrote the plan, who knows relevant players would have a lot of ability to drive the train to make sure that we actually execute that in a way that makes sense versus somebody who has to come in new and get started from, from square one?

B (59:47)

Well, yeah, I mean, it's definitely the kind of thing that like, certainly it can exercise a high degree of, of influence over it from within the White House. I think there's a couple of things. First of all, I just want to contextualize for, for your listeners so they're aware, please, what we're talking about here is not security regulation about Private data centers, but instead about the security of data centers that are specifically contracted by the intelligence community or the Department of Defense. There are obviously already security standards for such things. I think one of the really challenging political risk areas of this is, like, in the nature of those standards, are those standards good enough or are they not good enough? Like, do we need to throw it out completely, or can we. Can we iterate on those current standards or whatever, like, for, like, classified data centers or whatever? You know, I've heard some people say to me that actually these days, private sector data centers might actually be, in a lot of important ways, more secure than classified data centers, because the standards that the government uses just don't get updated. And so it just becomes like a box checking exercise. That's possible. I think there's a complicated issue. I won't get too much into the technocratic details.

A (61:10)

Sure. But this is helpful context.

B (61:11)

Yeah, yeah, yeah. But I do have opinions about this. I think you are totally right that this is the kind of thing that you could really drive from within the White House. But something else worth noting is that the action plan is a snapshot in time, and I myself don't have all the answers to exactly how to implement. There are some things that are vague because it's like, well, we just really need to talk about this. And I couldn't quite do that in three months. You gotta figure this out, need to develop some policy. And so then you get into the question of if there's, like, policy ideas that are necessarily. Because we're talking about emerging technology, it would be weird if all the policy ideas were fully developed. If there's stuff that you need to, like, bake more, what's the best place to do? Is the White House a good place.

A (62:07)

To develop policy ideas, as you said in other conversations? No.

B (62:11)

Yeah. My conclusion was no. And now it's. See, I'm. I'm. But you. You have asked me these questions in such a way that you're now, like, actually getting the narrative eyes, like, the exact sort of train of thought that I myself went through, where it's like, no, I don't think so. I don't think that this kind of environment, which is an amazing environment, it's not a criticism of it all, but it's not conducive to, like, developing new ideas. You have to come in with a lot of ideas, like, fully baked.

A (62:37)

On another great podcast a couple weeks ago, you talked about this challenge, and one of the things that you mentioned was it's very hard to work in public, in the federal government, you don't get to publish a substack and say, hey Twitter, what do you think of this? And you get dogpiled and you update your thoughts. Exactly. You can't do all that back and forth. The mechanism that we do have in the federal government formally, for quote unquote, working in public, is requests for input, requests for proposals. Right. Like these very formalized mechanisms by which you say, hey public, we're working on an AI action plan. Here's the sort of thing we're thinking about. What do you think? And I happen to know the RFI for the action plan got a little over 10,000 public comments, right?

B (63:23)

Yeah.

A (63:23)

And I know you took a look at a lot of them. I'm curious, just back in the napkin, how many of those public comments were actively helpful for developing your thinking?

B (63:35)

Really good question. And actually very is the answer. You know, I would say like a lot of them were very helpful. I would say, you know, probably it does ultimately end up being a heavy tailed kind of an outcome where like a certain number of them were extremely useful because it was stuff I like, man, the people. There are some people that wrote some like very, very discreet, specific and actionable things about like data center power, for example, that I just found enormously useful in getting me up to speed on that issue. There are others that were perspectives on AI that I just never in a million years would have considered myself.

A (64:14)

Like what?

B (64:15)

Like agricultural interest groups, you know what I mean? Like farmers groups that represent farmers from various. The dairy farmers and stuff. Like what do they think?

A (64:24)

What do they think? What do dairy farmers think about AI?

B (64:28)

It's interesting. Well, no, I mean the interesting thing is that unfortunately it's one of the areas of the action plan that we didn't quite ever go into.

A (64:36)

The dairy section, the agriculture.

B (64:39)

I wanted to do an agriculture section, but I didn't have enough time even to develop. There wasn't enough stuff that I felt high conviction in that I felt like, yes, this is the right thing to do. And it did. You know, it does take a lot of time to do that. The amount of time you have to develop policy in the White House is non zero. But most of the research that went into the action plan was done before I joined the White House. At the same time, most of the research that I did for the actual plan. I should say when it comes to agriculture though, like, I mean, we're actually like remarkably close to being able to automate like big chunks of farms. Like we kind of already did that in many ways, like we've. It's been a continual progression.

A (65:15)

It's a process. And my understanding is that several folks in this admin have been interested in pushing that further as a way to substitute for labor, especially illegal labor.

B (65:26)

Spy. It's complimentary. Yeah. Another like little lesson I will say is like, you know, just for the intellectual entrepreneur out there listening, the action plan is ultimately a document of many different intellectual entrepreneurial ventures undertaken at the same time. If you can find things that will get different people rowing in the same direction, that's really good. You should totally do that. You should totally look for things like that that get weird find like weird, you know, Kanye west style mixes of different coalitions that you can bring together into one. Yeah, I mean Kanye west stylistically, not politically, to be clear, but as a producer of music.

A (66:04)

Yes. I figured when you say that, that makes me think about one of the places that I think there might be the least amount of overlap between different kinds of intellectual entrepreneurs in the action plan. And that's there's a lot in the action plan about data center build out. There's a lot of interest generally on what you call the tech right. In data center build out. And that's a place that I see especially when you talk about data centers on federal land, special compute zones, as we've talked about at ifp, where you see a huge amount of pushback from other parts of the Trump coalition, folks who are otherwise bedfellows in the president's coalition. But you see really different instincts, really different philosophical views on the role of federal land at all, on concerns about all kinds of data centers, whether it's energy usage or what have you. How are you thinking about that general political tension now that you're out?

B (67:03)

Wonderful question. So I mean, I think one of my bets on like how anti AI sentiment will manifest itself in the political world, one of my highest conviction beliefs is that it won't be new regulation, it will be data center NIMBYism of various kinds. So it will be exercised in local decentralized fashion for the most part. I think that's probably true. At the same time, you know, probably the Trump administration probably places somewhat less emphasis in the action plan on federal lands than did the Biden administration. It's not like a full area of. Because like the Biden administration had like a 30 page executive order that was like all about federal. It wasn't. There was like the geothermal thing they did, but it was basically about federal land.

A (67:47)

Some good geothermal stuff. Just want to Say we. We're very happy with some of that at ifp, the geothermal work.

B (67:54)

Well, sorry, we rescinded it, but. But although at the same time, the. The administration, as you know, the Trump administration remains quite enthusiastic about geothermal.

A (68:02)

That's right.

B (68:02)

But anyway, the approach that we took with the AI infrastructure EO that we did was to focus much more on private and public lands. I do think that's a very real issue. I don't know exactly. Like, I think there are some people that are using this idea of federal lands as, like, the idea that we are. The public is now making some sort of sacrifice to construct the data centers. And so, like, we're paying for it. And that. That's somehow wrong. And that probably mostly has to do with animosity toward the people that they view as the beneficiaries of the data center, which they don't view as themselves and their children, and they instead view as distant billionaires in California. And that's a broader political issue about which I, in the short term, can do very little. But I would also say I'm a practical person. My fear of data centers on public lands is actually kind of like IFP had a version of this that I'm more okay with. But, like, what IFP wanted to do? The reason that, like, Tim Fist, the IFP Scholar, was like, so into this was that he wanted to attach riders to the contracts to the leases. He wanted to make them do stuff in exchange for the lands. The thing.

A (69:15)

Security. Security guarantees.

B (69:17)

Yeah, yeah. Cyber cybersecurity guarantees. Right. Which is, like, legit. That's like a real thing. But the political economy of that is that it's just taxed on a lease to be whatever you want. And it's the second. The second your landlord is the federal government. You know, real estate developers say the federal government is a terrible tenant. I can only imagine how they are as a landlord. And so I'm kind of like, well, look, we. The federal land authorities are there and they're activated. And like, you know, if agencies know they can use them because there's that eo and if they want to, that's like, great. They should, if there's some tactical reason where it makes sense. But I really. I personally, like, want to turn down the volume on the importance of federal lands for AI because I don't think land is really the rate limiter. And I don't like the political economy of it in the long term. And if it also happens that some people in the Republican coalition don't like it for like, sentimental reasons, then like, fine. I don't think anyone wants to build a data center in the Grand Canyon, right?

A (70:20)

The Yellowstone XI training run.

B (70:24)

No, but every inch of America is beautiful and so whatever, like, you know, I guess I just don't like if it actually entails a cost. And also on, like, as a policy matter on the merits, I'm like, a little skeptical of it. Then I'm very happy that we turned down the volume on that. And then the volume seems like it's now at its appropriate place, which is like 10%.

A (70:45)

Okay, let me go to my grab bag of questions at the end. You've done yeoman's work dealing with my implementation. Hang up here. I want to ask you just a specific policy question. There's been this fight over export controls. Export controls of the H20 chip that Nvidia produces. Should we sell a bunch of them to China or not? There's been backs and forths within this administration on exactly what we're going to be doing. It's unclear whether the Chinese even want the H20 anymore. So there's a kind of debate influx right now. Meanwhile, Nvidia announces they're rolling out the B30A, a new chip which is. Gets around the existing export control rules on the h20 and is half performance, half price. So theoretically, the Chinese can just buy twice as many of them and they're as well off as if there were no controls at all. How should America think about the export of the B30A?

B (71:40)

Yeah, so I think that the fundamental gap that I see in, in how the administration has approached. Well, there are. There are a number of reasonable criticisms one could make of, of the current approach to export controls and like, just specific tactical stuff in this administration. I mean, I don't mean going back. You know, certainly there's much bigger critiques you can make of the Biden administration and, you know, even the first Trump administration. But let's set all that aside and just talk about tactics for a second. I think there's a legitimate case to be made for, like, we want to sell more chips to China. Like the case that Jensen Huang, CEO of Nvidia, will make is that China should be dependent on US technology, they should be dependent on US Chips rather than developing a domestic semiconductor ecosystem, because if they do that, then they're going to be able to flood the market and out compete all of our companies in ways that we've seen before. So the correct strategy in that model of the world is basically we need to double down on controls. For semiconductor manufacturing equipment. But then we should also sell China some of our top of the line Nvidia, AMD and other, you know, AI acceleration hardware. I think that is a very reasonable strategy to take. But to go back to your implementation point, you have to actually implement it. And to implement that strategy requires more than just saying that's what we're doing. And like taking the tweet length version of that. I of the argument I just gave you, you have to think a little bit more carefully. One thing you have to think about, for example, is what sum, what does sum chips to China mean? We have ways of making this intelligible. We should do that. We should have a framework that is a piece of policy where we say we're going to allow this much. We want a chip that is X percent better than the best chip in China. And once that if an American company makes that chip, we're going to sell China Y number of that chip that is X percent better. Give the American people, give the relevant businesses, give everybody predictability. You also, you therefore can make sure that like we're not allocated because at the end of the day, the rate limiter here is tsmc. It's how many, how many chips can actually be fabbed, not how many Nvidia will sell as many as they make, at least over the next couple of years probably. So it's like that way you can, you can have some reasonable sense of, okay, like we're talking about giving roughly this percentage of, of TSMC's chip output of AI chips to China. That seems like the way to do it.

A (74:20)

Should we give American companies the right of first refusal on those chips? Because as you say, the bottom, there's more demand than there is supply of these chips and a lot of American companies would love to buy them.

B (74:31)

Yeah, I think that's a, that's a really interesting idea. It seems like the kind of thing you could lawyer your way around pretty easily on either side of that. So like how exactly you write that into whatever BIS rulemaking statute. I don't know how you would do it, but however you do that, I think that would be like, that'd be a great way to do it. I think that would be like, I mean, what could be more America first?

A (74:54)

How can we implement this idea of selling China a certain amount of chips and no more in a world where buyers in Malaysia famously buy a bunch of chips and then turn around and pass them along to their, to their partners in mainland China?

B (75:14)

So to return for a moment to the action plan this is an area where like the action plan does not actually say anything specific about what the substantive content of AI compute export control should be.

A (75:27)

Right.

B (75:28)

What it says is that whatever the compute export controls are, we are saying this is American law. So damn it, let's enforce it. Let's enforce it evenly and robustly throughout the world. And so that people, you know, like take us seriously, like, you know, big navy and stuff, right? Like don't screw with us. So make sure that we know when you're violating the law. Because it doesn't matter how big your guns are if you don't know that the law is being violated or where or by whom. So yeah, I mean, I think one of the things that you know, in that regard that the action plan goes into is, you know, this idea that we need to be partnered with the intelligence community more thoroughly between the Bureau of Industry and Security and the Commerce Department, which actually writes and enforces the export controls and the intelligence community. So much information sharing that could be happening there that isn't. Yeah, the boggles the mind when you think about it, about how much low hanging fruit there is there. There's also this question, you know, maybe somewhat one of the more controversial parts of the action plan of location verification. The action plan is very sincere in what it says about that, which is like, it is worth exploring. I'm not technical enough to know what is feasible and what is not. Is that creating a security vulnerability? I think there's probably versions of it that do. Maybe there are versions of it that don't. I'm not really sure. But I think it's like the White House should get its people together and really explore that. So those are all things. But one other piece of this that I just as a prescriptive matter that I think is important to emphasize that's not in the action plan because BIS writes its own rules. But now I can comment on such things because I don't work for the White House anymore. So I don't just have to say BIS writes its own rules. This is where the benefit of something like a global licensing rule actually does make quite a bit of sense and is actually good for American industry. The Biden diffusion rule was an attempt at this, but it added a lot of complexity and it grouped the countries of the world into strange classifications that I think offended a lot of people. Still do offend a lot of people, frankly. But you can have a global licensing rule that doesn't do any of that stuff. But does a make sure that the bulk of the AI compute is preferenced to the United States. You can also make sure that to the compute that's not built in the United States, we're preferencing American companies to build that compute, which is obviously very consistent with the export promotion agenda that we did talk about in the action plan. All this stuff can be put together into a harmonized way that gives again, chip makers, data center operators, financiers, et cetera, other foreign governments, everyone. Confidence and predictability, which is like, I think it's just very much lacking in this AI infrastructure build out and it is a problem. I think to some extent we need, we need more of that both internationally and domestically. We need more confidence and predictability.

A (78:26)

Yeah. Okay, last couple questions, Dean. One is about a piece that you wrote in November, so close to a year ago on the US AI Safety Institute, or AC as it's known. You laid out a roadmap for how the Trump admin could take AI seriously. And you suggested maybe this admin will be a better home for AI safety concerns than putative Kamala administration. Yeah, about a year on little less. How do you reflect on that argument?

B (78:59)

Oh, I feel great about that thesis. I feel so good about that thesis.

A (79:04)

Go on.

B (79:04)

So, so part of this is like a political realization about just the ways that the Democrat and Republican parties are hooked up in reality. So you know, like if you want to go. And I've talked to people who tried to like make the case for like AI catastrophic risk being a problem. And sure there are like good hearted Democrats who will just be like, yeah, that sounds like a problem for humanity. That's an issue. But the actual way that the Democratic Party works is that if you want to get the Democratic Party to like be supportive of like catastrophic risk things, well, you got to have a story about why biorisk is a problem for the teachers unions and for the ACLU and, and for the school bus drivers union and all the other various constituencies that Holy Roman Empire like make up the Democratic Party. Whereas Republicans. It's not like we have interest groups in the Republican Party, but like we're motivated by like big, like what was the president's like objective in the action plan for the action plan Achieve American AI dominance. That's it. Period. Like big objectives or in the campaign, you know, it was human flourishing and free speech. Like absolutely, that's. Those are big ideas. And it's not a bunch of jumping through hoops and like all this tortured stuff. It's just like this is, this is very simple and what we want to do that lends itself well to caring about these big picture issues, particularly because it turns out that many matters of AI safety and you know, like the doomers or like people that are worried about such things that they ultimately care about comes down to like values and alignment and control and like maintaining human control over these things. And those are all issues that, while I think it's easy for Republicans, I think one failure mode for Republicans will be repeating the social media policy battles with AI. They're not the same at all. But there are some structural similarities in that we have broadly aligned concerns about both of these different kinds of technologies. And so I think that there's actually a real opportunity there and Republicans are quite primed to think about it, to think about these issues in ways that are not the same as, but, but isomorphic to how like AI safety people think about it. And I've seen that happen in the real, like that's a real thing I've observed that is no longer a theory. I have observed prominent Republicans do this.

A (81:29)

Fascinating. Well, Dean, there's a lot of things I could ask you here, but I want to be conscious of time and I want to ask you anything I should ask you before we close out. What should we talk about that we haven't tackled yet?

B (81:45)

Okay. There's like two. Well, okay. All right. One of the questions that I can't actually answer, I don't have a good answer, but it's like one of the ones I want to is why can't the White House use Google Docs?

A (81:55)

Yeah.

B (81:55)

And more broadly, why can't the whole government use Google Docs? I was thinking about that a lot in the interagency process of like, why don't we all have one Google? Like, why can't we do that? We super. Can't. Yeah, definitely doesn't work that way. The other thing that like, again, and just like, maybe, maybe it's more of a future thing for us to talk about. But an underrated issue in D.C. is how rules and laws governing the flow of information profoundly affect the way the government works and the incentives of public employees. So FOIA is a good example of this. FOIA hugely impacts the way that the federal government is organized and what federal government employees do and do not say and do and do not do for that matter.

A (82:40)

And when you say hugely impacts my senses, you mean for ill?

B (82:45)

I mean sometimes it's for ill, sometimes not. Sometimes it's neutral. But like, I think the bigger point is that it's an artificial imposition on the activity. It's like, it's like this thing. It's not like a law of nature that every email is a record. Right. It's not like that's like we discovered that in quantum mechanics or something. That's like a thing we chose to do to ourselves. And I think that after having done it.

A (83:09)

Well, but most laws are.

B (83:11)

Yeah, no, most laws are, but like after having done it for 50 years, we should decide whether we still want to or whether we want to make any changes to that. The other one is classified information. Classified information. I don't know if you've ever done any like, episodes specifically on that, but like, man, the political economy of classified information is something that I could write a damn book about.

A (83:30)

It is. The institutional statecraft position is that we classify way too much stuff. And I think John Askinus has been very influential for me in this regard.

B (83:40)

No, we absolutely classify too much stuff. But the reason we do it is so interesting, like the, the various reasons that it's done, one of which is we classify information to keep people out of the room. It's about the interagency process because some interagency processes are classified. And so you do that. If you want to limit the number of people who can be in the room, you put it at the top secret level and you put it in a skiff and then like, I don't know, does NSF even have a skiff? I don't know. Like, you know, like who's. How it literally makes it harder because, like, you massively limit the number of people who can be involved. That's one example of many of ways in which the classification system is used tactically in government as a move and not necessarily used just to protect that information. A lot of classified, like, I promise you that, like half of what's on the front page of the New York Times today is like classified information. You know what I mean? Like, like technically. Right? Like you could find it on the high side is my point.

A (84:39)

Right. Well, Dean, it's been a real pleasure. Thank you for indulging me.

B (84:45)

Of course. Thanks for having me. This was fun.