The Economics Aficionado | Ep 94

Intro Voices
Coming to DARPA is like grabbing the nose cone of a rocket and holding on for dear life.

DARPA is a place where if you don't invent the internet, you only get a “B.”

A DARPA program manager quite literally invents tomorrow.

Coming to work every day and being humbled by that.

DARPA is not one person or one place. It's a collection of people that are excited about moving technology forward.

Tom Shortridge
Hello and welcome to voices from DARPA. I'm your host, Tom Shortridge.

What if the same mathematical theories that reshape global finance could be used to protect a nation's economic future? And what if new markets could not only generate private sector profit, but also inherently strengthen national security all at the same time? These are the kinds of questions that drive David Rushing Dewhurst, a program manager in DARPA's Information Processing Techniques Office, IPTO, formerly known as the Information Innovation Office.

He works at the complex intersection of economics, data science and national security, developing tools to help America navigate a world where economic competition is a primary battleground. Our conversation begins with David's first unexpected encounter with the agency he’d one day join.

David Rushing Dewhurst
Back in graduate school, I worked for the MITRE Corporation, an FFRDC, and I worked analyzing market microstructure data. That's sort of really low level, fine grained data about financial institutions and financial market structure.

And I went to an event that was hosted by this weird thing called DARPA. And I never, you know, I didn't really know too much about DARPA. I was, I think, 22 years old, maybe 23 years old. Everybody at the event was wearing suits, dressed to the nines. It was at the Federal Reserve Bank in New York. And then this guy came walking in, a kind of short guy wearing a wrinkled plaid shirt and shorts, and he stood up to a microphone and everybody stopped talking and just turned and looked at him and listened to what he said.

And I remember sitting there saying, I don't know what that is. I don't know what I'm looking at, but I want whatever that is. And so after I asked, you know, who is what's going on here? And this is Wade Shen. He's a program manager in the Information Innovation Office at DARPA. And so ever since at that point, and I don't even think I've ever told Wade this story. So if he sees this or hears this, it'll be interesting for him. I was intrigued by the possibility.

Tom Shortridge
That possibility is now reality. But it wasn't a straight shot to get to DARPA.

David Rushing Dewhurst
It's a very, very roundabout way. So I started off in high school not thinking I was going to go to college. I almost failed out of high school, graduated high school in 2011. And right then the oil boom was happening, particularly in North Dakota. And I said to myself, you know, I'm going to go be a roughneck in an oil crew in North Dakota because I'm going to get paid over $100,000, if you can believe it. And they're going to give me a pickup truck. And so to me, at the age of 17, it's like, wow, that's like the craziest thing in the world, right?

I did end up going to college. And I went to college to study political science, of all things, political theory, okay. And I’m in college studying political theory. And at one point I wake up and I say, you know, all of this is actually just economics. Realistically, everything about this is economics. And so I study econ for a while and I say, you know, if I want to understand any of that, all of this looks to be mathematics put together in some kind of weird ways. Some of them seem a little suspect to me, but how can I really critique it without a knowledge of mathematics? So then I go study mathematics, and I get three majors in poly sci and econand math. And then I say, okay, I actually kind of like math. I just became obsessed with math. So I did my Master's in pure mathematics and functional analysis, which was the sort of arcane area of mathematics.

And in my Masters, I realized in order to be able to use anything that I did, I would have to be able to program at least pretty well. And so I taught myself how to program. And a funny thing happens to some people - data just kind of happens to you. Like, I wasn't one of these people that was really attracted to big data or something. There are all sorts of areas of functional analysis we won't get into, but you can sort of learn approximate functional forms corresponding to data, and you can ask what sort, of among this whole big space of functions, what function best describes this kind of data. And you can do that in a way that is parameterized where you have learnable parameters, and you can do it in a way where there are no learnable parameters. You just try to pick the perfect function. It's beautiful, it's elegant, it's fairly complex. And it also happens to be really applicable to things in the real world. So data science, which is my Ph.D., is technically in complex systems in data science. It just kind of, in some sense, happened to me due to following a thread of interest, I guess, throughout the years.

And so I worked at MITRE for a while. I left MITRE and I worked in in risk management at a large asset manager, a life insurance firm, MassMutual, it’s a great company. The role wasn't exactly what I was looking for, and I left there and went into defense research and development, and I did work there for just about three years at the intersection of actually, again, financial market resilience and risks and opportunities in there, probabilistic programing, which is a niche area at the intersection of programing languages and statistical inference and all the interactions in there.

And at one point I had a bunch of ideas. I pitched them to a DARPA PM. He introduced me to the office director of a different office, the Strategic Technology Office, that was Dr. Phil Root at the time. And Phil took a chance on me and said, would you like to come join DARPA as a SETA to helping to build a strategic economics portfolio here? And I said, what's a SETA? Yes. And so I did that for about a year until I made the decision to become a government employee and have more initiative over what I could do by becoming a PM, and that's where I am now.

Tom Shortridge
David joined DARPA as a program manager in April of 2024 to design, execute, and transition programs at the nexus of technology and economic strategy. That's a broad mandate. Let's have David break it down.

David Rushing Dewhurst
It really does come down to very, very fundamental, you know, small political questions as in political science questions. You know, here is the state. The state is a thing. And we live in the United States of America. We love the United States of America. Everybody has things they want to change about it. But generally speaking, everybody loves the whole thing. How do we ensure that this country perpetuates, that the country lives on, that the ideas that formed the country live on into the future in perpetuity? So we all have the freedom to do as we wish, to work as we wish, to prosper in the ways that are best to us.

These are these fundamental ideas of the country, as outlined in the Declaration of Independence and enshrined in the Constitution. One of the, if not the biggest asymmetric advantages of America is its incredibly well developed and diverse economy, in many, many areas, in sort of creative industries and information technology. Certainly, my goodness, in financial markets, American financial markets are rivaled by no one.

However, because of our great success and incredible prosperity as a nation in some of these areas, some other areas of the nation have lagged behind in the sense that we have abandoned core capabilities. And this is not a new statement for me. We have core areas  -manufacturing, heavy machinery, things of this nature that have been left to wither because under normal competitive economic conditions, a well-developed country doesn't need those and you can prosper without them, and so on and so forth.

All of these insights from economic theory, but those insights from economic theory rest on fundamental sort of international relations. Political science insights, like, other countries can't be trying to actively destroy your country. Well, if you live in a world in which other countries are actively trying to destroy your nation’s areas of excellence that let it prosper and flourish as a nation, all those things we said we wanted, you have to think about how to defend those really in a directed manner.

Tom Shortridge
So the mission then is clear find new ways to protect and strengthen America's economic advantages. Traditional methods to do so involve what's called economic statecraft, a range of policy tools of both the carrot and stick varieties. But David likes to think about it all a little bit differently.

David Rushing Dewhurst
One of the, dare I say, probably unique in the world aspects about America is that our default position is that the private sector, ordinary people, can probably just do - fill in the blank, whatever it is. So when we say economic statecraft, I prefer instead just to think about creating new mechanisms, new things, maybe they're new market structures, maybe they're new assets, maybe they're new sources of information or ways in which information is being used at very low marginal cost to the government.

That does two things simultaneously. One, that creates positive, profitable outcomes for the private sector and creates positive national security externalities. At the same time, maybe they're interacting in a new marketplace. It's just more competitive for them. Everybody likes that new marketplace better. Well, it turns out that it happens to be a marketplace that brings back market share in some critical commodity to America, as opposed to outsourcing to one of its adversaries.

And we did this all at a very low cost to the US. Wow. Wouldn't that be incredible? The baseline of theory that exists today to say how to create, how do you create one of those markets? How do you get people interested in it that doesn't exist, that doesn't exist. And that is sort of the reason that we that we work in the econ statecraft space, or as I would prefer to say, the sort of theoretical or quantitative geo economic space is to build that capability.

Tom Shortridge
This wouldn't be the first time a seemingly obscure branch of mathematics has completely reshaped the world of finance, and with it, American strategic influence.

David Rushing Dewhurst
In the late 60s and early 70s, the financial researchers, total egghead and mathematicians such as myself, came up with some really important discoveries in an area of math called stochastic calculus as applied to really abstract economic questions. And they essentially describe how to hedge risk in a market where you have an option, the right, but not the obligation to buy or sell something and the underlying something. Fast forward from that time, lots of things have happened. Fast forward, not even to the present day. Fast forward maybe 20 years, and what you have is an incredibly rich, diverse international market.

But, may I say, international market dominated by the United States of America in elegant and useful ways to hedge all kinds of risk. And the theory underlying all of that started with these fundamental papers by Black and cSholes and, yes, Bachelier way back in the early 1900s. But his work sort of remained unthought about until much, much later.

I digress into mathematical absurdity. Anyway. So what can happen is that very sort of egg headed scientific mathematic development can take a very long time to mature and turn into incredible tools of, dare I say it, power for the United States of America. Power because our country, uniquely among any in the world, leverages and embraces innovation. But it has to take time to be developed into theoretically rigorous, correct way, and it has to take time to mature.

And a lot of times, as fantastic as the private sector is, the private sector is not the institution that does that best. DARPA is an institution that has a decades long history of doing that. And so that's another example on the total other end of the abstraction spectrum where, you know, we can have great research, turn into something that's really important for national security.

Tom Shortridge
David's National Security Economic Theory program, NASCENT, aims to establish a principled theoretical foundation for geoeconomics, enabling evidence based economic statecraft and defining a rigorous scholarly field of quantitative geo economics. But what does that actually mean?

David Rushing Dewhurst
It's a lot easier to say, I'm going to build a thing in this box. It's a lot harder to say, I'm going to purposefully construct new theory in this area, because what is NASCENT trying to do, what that looks like is sort of on purpose, more diffuse than that would ordinarily be.

NASCENT is trying to build playbooks – generic, and I mean generic in sort of the programing languages sense. It's not one function mapping this thing to that thing. It's a function that's reusable over and over in many different contexts. Generic playbooks. America is in these sorts of macroeconomic scenarios. Here are what its adversaries capabilities are. Here's who the private sector's actors are that are amenable to working with the government. Therefore, here's the new kind of mechanism that should be created. And so that's something we hope that NASCENT can create.

More broadly, we hope that NASCENT can in some sense make the concept of quantitative, rigorous thinking in economic statecraft not strange, not hard, but ordinary, of course. Of course you would think that would, even if people don't know it's ordinary. The bottom line here is that people today in the government, in the executive branch, are doing the best that they can with what they have. We're trying to counter incredible economic and national security threats from nation state adversaries. I have a small toolbox. I have only a few tools in it. I'm going to look out there in the distance, say, I want to look like that goal state. I want to be prosperous. I want to be resilient like that. I don't want to look like this state with, you know, unsustainable inflation or whatever else it is. Right. And I'm going to start from where I am and put my tools together, and hopefully I get to the good one and not the bad one. That's the best you can do, right?

Well, maybe it won't be. And the idea - there's a concept from math. I'll try not to get too geeky. There's a method called dynamic programing that says, let's start at the goal state you want to be. Imagine you're there. What would you have had to do to get to that goal state? Okay, find that optimal action okay. Let's work backward from that optimal action.

Let's find a former optimal action. So that's not quite an accurate description of it, but it's suitable for this discussion. If we have the right theoretical underpinnings for geo economic interactions and modeling.  It's going to be a lot easier to start at the goal state and put the right kinds of tools together, sort of backwards, in order to find what we would call the Holy Grail, is called an optimal closed loop policy, which says, you know, at any point in time, no matter what my adversaries are doing, this is called the subgame perfect equilibrium, I know exactly what I need to do in the next time step. And it is guaranteed to be the optimal choice for me.

Is that actually possible in the real world? Of course, that's not actually possible in the real world, but it is a useful model for us and we can try to approach it in various ways.

NASCENT is fundamentally about hopefully partially automating or massively reducing the marginal cost of creating new economic structures to solve national security problems at low cost to the US government.

Tom Shortridge
Another of David's programs, Anticipatory and Adaptive Anti-Money Laundering, A3ML, takes on the immense challenge of global money laundering, but approaches it as a problem of economics, not just regulation.

David Rushing Dewhurst
Anti-Money laundering is not new, money laundering is not new. And I'm going to come out straight away and say, if I ever told you that I would try to eliminate all global money laundering, that's goofy. That's never going to happen. Human beings want stuff. When human beings want stuff, they're going to go and try to find a way to make it or buy it. That is how people are.

Money laundering is a service. There's a supply and demand for the service, and there's never going to be zero supply because it's never going to be zero demand.

The sort of typical, I would say, sort of quasi legal approach to money laundering is to say we are going to put new regulations, new policies, new compliance procedures in place. And if a financial institution follows these procedures or a high net worth individual follows those procedures, there's going to be less money laundering.

No, people are going to find new workarounds around the procedure. This is this really, really sort of legalistic approach to the problem. And it just hasn't worked. It's publicly reported North Korea launders over half of their funds for the nuclear program. Money laundering is a huge problem.

Let's just think about it like, you know, econ 101 student, how do I move the supply curve of money laundering to the left? What does that mean? It means I just reduced the global supply of money laundering. Okay. That's interesting. Like, that's the fundamental concept. If I can do that, that just means I'm reducing the global supply. I'm shifting the curve to the left. What does that mean? That means the price is going up, right, at the intersection with the demand curve. The demand is eventually going to dry up at some point. There's never going to be zero demand, but it's going to be nearly impossible if I can sustain a very high price for global money laundering. Well, if that price is comparable to what the adversary will be gaining by laundering, it washes out. I'm not even going to do that bad thing anymore. Yay! We made America more secure.

Okay, that's A3ML. What is A3ML actually doing? A3ML has two technical areas in it. In the first technical area, I don't know how to say this. Folks are going to try to find money laundering. And so they're going to come to the table with tactics, techniques and procedures - borrow that term from cyber -  or TTPS of how threat actors launder money today.

And I say launder money. What I really mean is transfer liquid value because it's not always money. It's securities. It's cars, it's gold, it's Tide Pods. There was a thing in prisons where people used to pay for stuff with Tide Pods in the US. I'm serious. Look it up. We're not worried about tide pods on A3ML.  The point here is, you know, we're going to go look for stuff.

In the second technical area of A3ML we get to, you know, be the bad guy I've always wanted to be and pretend to launder money. We create these hypothetical illicit finance tactics, techniques and procedures for the purposes of red teaming the TA1 algorithms. For TA1 performers, they all have different approaches and they all say, hey, we're going to create TTPS that look like a threat adversary. So they have to do that. So pick your specific favorite threat adversary. I'm going to pick TTPs that look like that one.

And the TA1 is going to search for them. And then they're going to make totally new ones. Hey ,how would you launch of money if you really wanted to? Please don't, please don't. And inject those into data and go try to search for instances of those as well. And so the TA2 metrics are ,how well can you fool the TA1 performers, but also how realistic is what you've created.

And we're hoping with this sort of kind of recursive game theoretic approach that they play off each other so the TA1 gets better and better as it finds harder and harder TA2 stuff. TA2 gets better and better as TA1 gets better. We'll see if that works. A3ML is still a pretty new program. It's looking promising right now, but you know, check back in here.

Tom Shortridge
Hunting for patterns of illicit finance across vast global data sets raises an immediate and critical question how do you target bad actors without compromising the privacy of innocent people?

David Rushing Dewhurst
Privacy entered the conversation on like day one or day two of, “oh man, I think we have to do A3ML.” So A3ML comes at the problem in the following way. We have what we call sort of the Zen of A3ML. The tactics, techniques and procedures used by the adversaries will always change. You will never have all the data. You will never have all the data in one place. And it's this third one that is so important for privacy. Across the US government, and particularly in the intelligence community, Department of War, we often try to like centralize all the data. I'm going to wrap my arms around all the data. A, that destroys privacy. B, no, you're not, because there's always data you're not going to have.

So in A3ML, we say, everybody keep your own data. That's fine. We don't want to see your private data. You have all your private customer data. You have all your private US persons data. We don't want to see it. We're going to develop analytics that work fully asynchronously. And we'll send a version of those analytics to your private data store that return results, we might say aggregations or views of that complex sort of semantic structure that you have there. And those views can report how many of which different types of illicit finance TTPS are there.

Here's some new type of behavior that we think is associated with illicit finance. It might be a new TTP. Those can be reported across other sources of sensitive data, but never does the algorithm report the SourceA sensitive data to Source B or vice versa. The way we've designed the technology system is that by default, nothing that reveals privacy or private information is ever shared. Only abstracted insight about the kinds of threat finance occurring in different subsystems is shared.

I am personally sort of a privacy advocate and used to advocate very strongly for this and sort of in my private life and still do. And so I've talked to some very, very skeptical people about A3ML, and they generally have come around to saying, well, that's at least a heck of a lot better than the way it's on the day.

Tom Shortridge
David brings, perhaps a more pragmatic perspective than we're accustomed to on this podcast.

David Rushing Dewhurst
As an adult, I don't read science fiction, and I don't know if that says something about me, or if that says something about the way technology has progressed. I find myself surrounded by incredible things all the time, I find that we too easily move by small imperfections, and that we too easily want something big and incredible and new for the sake of novelty.

Nothing I've said in this entire  - however long we’ve been speaking - is about being new for the sake of novelty. NASCENT is about new things. New things do matter, but a new consideration or a new use of old things. A creative repurposing matters as well. Imperfection matters as well. Imperfection is incredibly important to us in defense R&D, in the intelligence community. It's in some sense our lifeblood.

I read much more history. I read much more sociology, things of that nature, things that can't be perfected, things that aren't going to be the ultimate vision of something. Things that have to be worked with the way they are.

Tom Shortridge
That's not to say that David's programs, his ideas, are any less audacious than others at the agency. He understands the weight of expectation and the impact they could have.

Eventually.

David Rushing Dewhurst
DARPA barely exists. What I mean by that is DARPA is an idea more than it is anything. The SETAs live here on, you know, the contractors live, but the PM's go, the office directors go, the director of DARPA goes. We are all so ephemeral. And that's on purpose.

The programs end, everything ends. Everything is permanently, you know, collapsing into the pit of the void, as you know, Roberto Boloña would say. And so, like, success is almost never realized while you're here. And you have to be okay with living with that, you know, and failure is also part of it. You have to come to terms with saying, you know, here's here's the NASCENT program, right? It might just fail. Like, I might just be nine months into the program, like, guys, it turns out we just can't make that kind of theory. Okay, okay. We tried.

I mean, what does success look like for me? I mean, I think there's a couple timelines. For A3ML and NASCENT I mean, A3ML is going to take a long time. You know, we can have complete technical success. Again, you know, DARPA doesn't make policy. I'm not the person that goes to FinCEN, the Financial Crimes Enforcement Network, and says, you know, “you have to use the A3ML program.” I mean, they don't have to use the A3ML program. And so it's going to take creative people at FinCEN, at other places across the US government, maybe even, you know, other places around the world to say, hey, we really think this is a better way of combating global money laundering. I don't know how long it is. Four years. Five years. I hope not too long because it's a real problem.

With NASCENT, the success, if I'm lucky, you know, I have three little kids, six, four and three. And by the time my six year old is graduating high school, maybe I'll know if NASCENT was successful. Maybe.

Because success for NASCENT is walking into the Eisenhower Executive Office Building, talking to the NSC the National Security Council, National Economic Council or something, and having them say, “yes, so in our, you know, framework for this, that we, you know, we push this button and, you know, this framework shot out and we applied it to this scenario. We really think the optimal actions are the following. And we're going to create these six markets.”

I will not say how laughably not close to reality that is today or ever has been.

That's what success looks like. That's not coming anytime soon because even if NASCENT is wildly successful, we made the best theory in the world, like, that theory still isn't going to be directly used. It takes maybe more DARPA programs, maybe programs else where in the US government, certainly it takes buy in from corporate culture to be okay with this as well. And that's going to take years and years.

People have an idea about what a successful DARPA program looks like. And a lot of times that's, you know, I have this great idea. We do a bunch of work for two phases. And then in the third phase, I transition to this part of the Department of War. And then it was a win.

The goal of DARPA is to create and prevent strategic surprise for the United States of America. The warfare America is sort of besieged by and must engage in today, it's not just multi-domain, it's almost like pan domain. Creating or preventing strategic surprise for the US doesn't just look like that thing there, that thing there. It can be changing ideas. It can be changing culture. It can be widely promulgating work.

One of my fellow program managers, they've done a really great job at open sourcing lots and lots of, you know, privacy increasing technologies. Now that's interesting for the following reason. You say, well, open source, oh my gosh, our adversaries could use it. Well, I mean I guess I guess they can. They can use it to increase the privacy of their own people. Golly. How horrible. You know, whereas, you know, okay, if that's what happens, that's something that we want. Yay. People can now be private and, you know, not be besieged by a totalitarian regime.

So it's creative and elegant approaches to how your technology makes a difference in the world, often far after your tenure that I that I hope people think about very carefully before they come.

Tom Shortridge
That's all for this episode of voices from DARPA. For more information on David Rushing Dewhurst and his programs, including NASCENT and A3ML, please visit DARPA.mil. We'll have links in the show notes. As always, thanks for listening.