September 24, 2024

Podcast

Lav Varshney Connects AI Research, Executive Policy, and Public Service

By Lav R. Varshney, Jason Lloyd

In this installment of Science Policy IRL, host Jason Lloyd goes behind the scenes of the White House Fellowship program with Lav Varshney, associate professor of engineering, computer science, and neuroscience at the University of Illinois Urbana-Champaign. Varshney served as a White House Fellow from 2022 to 2023, where he worked at the National Security Council with Anne Neuberger, the deputy national security advisor for cyber and emerging technology.

In this episode, Varshney describes the day-to-day experience of working at the White House, gaps in the innovation system that science policy can help fill, and how making artificial intelligence systems more transparent could define the future of AI applications.

SpotifyApple PodcastsStitcherGoogle PodcastsOvercast

Resources

Transcript

Jason Lloyd: Welcome to The Ongoing Transformation, a podcast from Issues in Science and Technology. Issues is a quarterly journal published by the National Academy of Sciences and Arizona State University.

I’m Jason Lloyd, managing editor of Issues. On our series Science Policy IRL, we talk to people in science policy about what they do and how they got there. On this episode, I’m joined by Lav Varshney, an associate professor at the University of Illinois Urbana-Champaign, whose research focuses on artificial intelligence. Lav recently served as a White House fellow from 2022 to 2023, where he worked at the National Security Council with Anne Neuberger, the deputy national security advisor for cyber and emerging technology.

On this episode, Lav describes the day-to-day experience of working at the White House, gaps in the innovation system that science policy can help fill, and how making artificial intelligence systems more transparent could define the future of AI applications.

Lav, thank you for joining us today!

Lav Varshney: Thanks Jay. I’m excited to be here.

Lloyd: I wanted to start with what we ask all our guests. How do you define science policy or do you have a working definition of science policy?

What I saw when I was serving at the White House is that scientific research and the development of science can impact all kinds of levers of state power.

Varshney: I think everyone probably goes into the “using science for making policy” versus “policy for science.” And let me focus first on the policy of science. I think there’s the resources and funding aspect of it. But I think what’s even more interesting and what I saw when I was serving at the White House is that scientific research and the development of science can impact all kinds of levers of state power. So what is being researched can have impacts on the missions of a variety of the departments and agencies and government. It can have impacts on nearly every societal and industrial sector, but also it can play a role even in international relations. Working together with multilateral or bilateral partners can itself be something that’s valuable, but it can also help drive different policies. I think it’s actually fairly broad and getting broader in a sense because science and technology seem to be entering into nearly everything you can think of. On the flip side, there is the use of science in policymaking, kind of evidence-based policymaking, and I think that’s also kind of a compelling application. And I think for areas like health and wellbeing, I think it’s especially important to be drawing on scientific evidence as one makes policy.

Lloyd: Yeah, so that leads well into the next question about what you were doing at the White House. I’m really curious about the fellowship program. So what kind of science policy do you do either in your daily life in your day-to-day work? Or what were you doing day-to-day as a White House fellow?

Varshney: So let me start with what I was up to when I was serving in Washington. Just as a little bit of a background, the White House Fellowship is a program that was started in the Johnson administration, so it’s been around for nearly 60 years, and between 11 and 19 people are brought in to serve at the highest levels of government. Most of them are not scientists. In my cohort, three were from the military, there was a police officer, a city councilman, a community organizer, some doctors, a dentist, and so on. And so, it’s a nice mix of people and you’re kind of parachuted in your placement into being a special assistant to either at a senior staff at the White House or cabinet secretary in the departments and agencies. And so I was serving in the National Security Council staff and I was working very closely with the deputy national security advisor for cyber and emerging technology, Anne Neuberger.

The White House Fellowship is often taken on by people who don’t have so much experience in policymaking and it gives you a exposure on how to do that.

The portfolio that I was looking at was focused on AI policy. And that especially picked up once ChatGPT came out in November of 2022. I had started in September of 2022, and then also did a fair bid on wireless communication policy and also looked at a few things on the geopolitical side with respect to the role that science and technology play as these kind of levers of state power in various ways. I was able to do a variety of things, which was really amazing.

In terms of my professorial and startup life that’s historically been less connected to policy. And in fact, the White House Fellowship is often taken on by people who don’t have so much experience in policymaking and it gives you a exposure on how to do that. But when I was a grad student at MIT, I actually did take a science policy class, so at least I knew a little bit about it. And over the years I’ve done a little bit of scholarly work in science policy as well, or policy relevant scientific research depending on which paper you’re looking at. So, yeah, definitely have continued to be involved and going forward, now that I’m back, I’m keeping some connection to the policy world in Washington, but also pursuing some scholarly research that is policy relevant and then also some policy-focused research.

Lloyd: Cool. Can I ask some very practical questions? So when you got this fellowship… Okay, first of all, did you apply or were you asked to apply? Were you selected?

Varshney: Yeah, someone I knew through the National Academy of Engineering, in fact, encouraged me to apply. And so there’s a written application, then there’s regional interviews, then there’s national interviews, and if you’re selected, you go back again for another week for interviews for placement. You interview in many of the offices in the cabinet secretary’s offices or the senior staff offices to determine where you’re placed. So it’s actually a fairly long process. The application is due at the beginning of January, and then you’re selected roughly in June, and then your placement interviews are July. And you might not find out where you’re placed until maybe just a couple of weeks before you start, which is the end of August.

Lloyd: Oh, wow. So when you were doing the placement interviews, do you then kind of rank where you would prefer to be based? Or is it kind of they choose where you end up?

Varshney: Yeah, it’s kind of a matching process. So it’s a little bit like medical school residencies. You provide some rankings and the offices provide some rankings and then some magic happens in the background.

Lloyd: So did you get your top? Were you interested in going to the NSC?

Varshney: Yeah, yeah, it was definitely some place I wanted to go and I’m happy that I ended up there, gave me a lot of exposure to a lot of different things.

Lloyd: So on a day to day-to-day workflow, was it a lot of meetings? Were you doing research? Was it a lot of… Were you writing white papers? What is sort of your daily activities look like?

Varshney: Yeah, yeah. So the day often started with reading intelligence briefings. One nice thing with the serving in the National Security Council is that you get access to the intelligence community, pretty much all the products that are relevant, whether it’s from NSA or CIA or pretty much anything that you care to learn about, they provide. So that was very helpful especially… the first few weeks I just spent reading and learning what’s going on in the world beyond what you read in the newspapers. And then once I settled in, it was a mix.

The role of the White House is very much in convening, so the White House itself can’t actually do anything.

The role of the White House is very much in convening, so the White House itself can’t actually do anything. It’s the departments and agencies that have the capabilities to actually execute on policies. And so there’s a lot of interagency efforts on getting ideas from the departments, agencies, and then coordinating and getting those back down for execution. So there was a lot of that kind of interagency coordination work. There was also a lot of just reading and thinking and characterizing what would make sense in terms of policies, whether on the domestic side or international things.

To take two examples that I worked on the international side. The US and the European Union signed an administrative arrangement on AI collaboration when I was serving, and then we were executing that. That required an understanding of what capabilities and strengths each of the parties could contribute and how that would play out in terms of larger policy goals because it was nice to get collaborative science done, but it was also nice to think through how to harmonize joint efforts across the Atlantic when we have different regulatory constraints. And can we create a playbook on how not just governments can collaborate in terms of data sharing or data collaboration and AI collaboration, but also private parties because it’s really nice if like-minded and democratic countries can work together to push these truly amazing technologies.

And to give another example, the US and India, we worked on some agreements on 6G and 5G wireless joint task forces. One focused more on the research side and one focused more on the industry side. And so again, trying to bring folks together to understand what contributions the US and India can make respectively and how we can drive joint efforts. Because if the second and third-largest markets in the world can really work together, that can be really powerful for spreading more open and secure and interoperable ways of doing wireless.

Lloyd: As you’re working on things, what is the work product? Are you producing memos that then go to the relevant agencies or directives, findings? What are you kind of producing?

Varshney: Yeah, so one kind of big thing that came out of my work—and of course I was only one small piece of this—the president signed an Executive Order on artificial intelligence in October of 2023. And so that’s a fairly long and detailed document. It’s, I think, 111 pages in the White House format. And it covers things ranging from national security to labor to workforce to immigration to fairness. I mean it covers DeepFakes; it covers all kinds of topics. So certain sections of those I contributed to quite significantly. So that would be an example of a work product.

Or coming back to wireless, so 6G wireless. So the US led a group of about 10 countries and to come up with principles of what we believe 6G wireless should be like. And we held a meeting in Washington that brought together academics, folks from industry, civil society folks, and then representatives from these 10 countries. And the final result of that is a joint statement that was issued, I think, maybe seven or eight months ago, which all the countries agreed to these principles. And so I was involved in organizing this fairly large convening, working with the State Department to get things going, et cetera. The document itself plus the convening, plus this coordination, that was the result of my efforts.

Lloyd: Yeah, that’s really cool. So we just did a recent interview with Rashada Alexander. She directs the Science and Technology Policy Fellows program at AAAS. And one of the interesting things she mentioned when we were talking to her was that she estimated that maybe 40 to 50% of each cohort of… These are all scientists, unlike the White House program. That 40 to 50% actually stayed in government service. They left bench science, lab science, and stayed in service to work in policy. But I’m curious if you were tempted to stay in government work in some capacity?

Varshney: Yeah, I mean it was a great experience. I learned a lot and hopefully contributed some things solid to the country and to the world. And yeah, definitely if I’m called back to serve in some way that’s important and critical, I would definitely go back. Depending on where things shake out, that would be really great.

To give you some examples, some former White House fellows have gone on to pretty high positions in government. So for example, Wes Moore, the current governor of Maryland was a White House fellow or Elaine Chao, the former secretary of labor and of transportation. She was a White House fellow. And Colin Powell was a White House fellow as well. So there’s kind of an interesting historical precedent for White House fellows to end up ending pretty high up either in appointed or electoral positions. In fact, even my boss at the White House, Anne Neuberger, she was a White House fellow about 15 years ago.

Lloyd: Oh, okay. Oh, interesting. Well, actually, so that gets to the next question, and we’re sort of moving a little bit back in time, but I am curious about even before you were White House fellow, you had mentioned, you touched on briefly that some of your academic work engages with policymaking. I know one of the things, a piece that you co-authored for Issues was very policy relevant involving artificial intelligence and intellectual property law and institutions. And so I’m curious how you became interested in policymaking and what the journey, for lack of better word, from of studying what sounds like pretty hardcore electrical engineering, computer science to ultimately a White House fellow, but just engaging with policy ideas along the way. What was the interest there? How did you end up getting involved or interested in policy?

Varshney: Yeah, it’s a good question because I’m not particularly self-reflective on my life, nor am I all that strategic. So in many ways it’s more opportunistic. Like when I see an opportunity to make an impact, I try to take it. I mean, I think we all have a duty and a obligation to serve our communities, and it was really the service element that drew me.

Intellectually, policymaking is interesting because we’re in a deliberative democracy where there’s a lot of different viewpoints and a lot of different desires, and trying to bring those together in ways that actually improve people’s lives is a challenge.

And if you look at some of my ancestors, they’ve had a bit of an influence as well. My great-great-grandfather, he was actually the second Indian to study at MIT back in 1904. So he took the boat over to San Francisco and then took the train via St. Louis and saw the World’s Fair and then ended up in Boston. So he studied glass making, and then he went back and started the first glassmaking factory in India and really established that industry. And the interesting thing is that first factory was part of this what’s called the Swadeshi movement, kind of self-sustainable industry, and he worked with some of the freedom fighters and then really helped push towards Indian independence. He was an example of taking science and technology and really applying it to industry as part of a freedom movement.

Another example is my mother, she passed away two years ago, but she was quite involved in volunteer work in our local community in Syracuse. She was also a public high school teacher in the city of Syracuse. And so she influenced me in this desire for public service as well. Those were kind some of the examples of public service or driving impact towards scientific and political goals and societal goals, that were in the background for me as well.

And then also intellectually, policymaking is interesting because we’re in a deliberative democracy where there’s a lot of different viewpoints and a lot of different desires, and trying to bring those together in ways that actually improve people’s lives is a challenge. And it’s in many ways, very similar to research challenges as well, except it’s not a closed deductive system where one can prove mathematical theorems. As you know, I’m a big fan of Claude Shannon who was the master of mathematizing. Here it’s all open, everything has to come in and out. And so it’s a different intellectual challenge as well.

Lloyd: I had forgotten, I guess, that Shannon was sort of a protégé or at very least studied under Vannevar Bush, who was the architect of the postwar scientific enterprise in the United States. But Shannon did not go in that direction. He stayed on the research side and really focused on problems in computer science, in engineering systems. But it seems like you’ve, unlike Shannon, have found a way to integrate these things a little bit more in your research in ways that the research and the policy are connected in this of service-oriented way.

Varshney: Yeah, it’s a little embarrassing to compare oneself to Shannon or to Bush. I’m nowhere in their leagues, but-

Lloyd: (laughs) You can let me do the comparison.

Varshney: (laughs) Yeah, I mean the endless frontier that Bush imagined as compared actually to the closed frontier that Shannon imagined in a way because he established fundamental limits, they’re actually quite kind of different and different ways of thinking. But yeah, I’ve been lucky enough to work in different epistemic cultures as it were, to walk around and do different things. And because I’m not particularly strategic, it works out, so it’s enjoyable to do all of these different things.

Lloyd: Well, for not being particularly strategic. I mean, you’ve chosen sort of a focus of study, certainly in artificial intelligence that has a tremendous amount of policy relevance. So it seems like you could kind of choose whichever direction you wanted to go in, which a little bit gets to the last question here. What are some of the big questions that animate your interest in science policy?

It seems to me that there’s things in-between—between curiosity-driven and mission-driven science that are still worth doing and often fall through the cracks.

Varshney: I think the question of innovation and how to carry it out I think is an overarching question in science policy. And the way we’ve organized our system following the Bush blueprint is that we have these mission-driven agencies that carry out particular kinds of scientific research, whether it’s the Department of Defense or the Department of Energy. On the other hand, we have the more curiosity-driven approaches to scientific research and its efforts like the National Science Foundation. This is the key example. But it seems to me that there’s things in-between—between curiosity-driven and mission-driven science—that are still worth doing and often fall through the cracks. There’s been some theoretical work in science policy that describes Baconian science and Newtonian science, and then Jeffersonian science is kind of in between.

Where Jeffersonian science is curiosity-driven, but has a direct impact in making a mission successful. And so that Jeffersonian science, I think personally is something I strive to do in my own scholarly work, but also I think is important going forward in terms of science policy because there are certain kind of gaps that are less filled. And it seems to me that it would be helpful to have more analytic capability in government to be able to fill those gaps and create settings where scientists can explore their curiosity and be the best scientist they can be, and yet drive towards missions that help society the most. And I think further in society, there are problems that often remain unstudied by those of us in academia, in part because we’re not aware of them. As I was mentioning, we’re a broad society with different concerns and not all of them end up flowing into the scientific research system. Not that I’m claiming science as a solution to everything, but science can actually say something about a lot of problems that remain unstudied. I think that collectively is something that is super important for science.

Another thing on AI in particular that I think is related to this point is that for national competitiveness, we need to drive not just AI innovation, but also diffusion of AI across all industrial and societal sectors. And a lot of my work is actually in that direction these days. I’m involved in some startups. So one is focused on music co-creativity. It’s called Kocree. And we’re building AI technology that is inherently human engaging, human interpretable, human understandable, and further it’s directly using very small amounts of data rather than stealing and scraping everything on the internet, which has all kinds of, not just security, but also intellectual property issues. We’re drawing on this information lattice learning technology that we developed to allow people to decompose music and recompose it into new songs, which also seems to improve people’s wellbeing, but also help drive cultural wealth growth, especially among historically exploited communities. So that’s an example of using AI in a way that’s different than the dominant narrative of these large-scale models that take in everything and are hard to control.

Critical infrastructure is a place where AI can have huge impacts, and we can really build smart and secure infrastructure that allows us to be much more efficient and yet is safe.

Another startup that I’m involved in—it’s called Ensaras—it’s focused on the intersection of AI and wastewater treatment. Very different than music, but it feels to me that critical infrastructure is a place where AI can have huge impacts, and we can really build smart and secure infrastructure that allows us to be much more efficient and yet is safe. We don’t want to cause a cholera epidemic if we misclassify wastewater or significant environmental damage. And so what we’re doing is we’re building kind of this approach to critical infrastructure that brings both of those together. And it seems to me that it’s a perfect sweet spot because it’s the here and now. It’s relevant to us very critically. It’s not a long-term threat. It’s something that we need to worry about right now. And yet it’s very visceral. I mean, it’s something that we engage in every day, and it’s core to government function, providing critical infrastructure. Those are two examples of startups I’m involved in right now.

Likewise with my academic research, we’ve been pushing on some ideas in the foundations of AI to understand how emergent capabilities happen or how scaling and different resources translate into capabilities. Because when one’s doing policymaking, one can sometimes be doing it a little bit blindly because you can regulate resources like compute and data, but not capabilities, and yet that’s what you care about. And so if one has an information theoretic mapping between how resources lead to capabilities, that would be super useful scientifically, but also for policymaking.

Lloyd: Yeah, that’s really fascinating. It seems like one of the things you’re trying to do is sort of open up what’s often described as the black box. AI has these inputs and then the outputs occur, and no one’s quite sure what occurs in the middle in that black box, but it does seem like what happens there is really important potentially for policymaking, but even for intellectual property. Just knowing what the system is doing to existing songs in order to create “new ones”. But opening it up seems like really important work.

Varshney: Yeah, I mean, white box AI, I feel, is the future for nearly all applications, whether it’s critical infrastructure where the safety comes from the openness or for creativity, like I was describing with music and further even for policymaking. In a sense, these alternative paths for AI that are distinct from large language models and related technologies can be a good possibility. Which is not to say that those LLMs and related technologies don’t have a role. I actually, back in 2019, when I was on leave at Salesforce, I helped develop some of the early billion parameter plus language models and worked on open source releasing those. So yeah, so I was involved in those technologies myself quite a bit. I definitely see their place in a variety of applications, but I think these white box AI techniques can be really great as well.

There’s a need for more scientists and engineers and technologists to serve in government because I think we can provide different perspectives, and our training and our life experiences as whole people can be really valuable.

Lloyd: Well, this has been really fantastic. Thank you so much for spending some time with us. I really learned a lot, and I have learned a great deal about being a White House fellow and also your work, which is fascinating.

Varshney: Thanks, Jay. I would encourage listeners to look into the White House Fellowship and other ways of coming into government because I think it is really a great program, and I think there’s a need for more scientists and engineers and technologists to serve in government because I think we can provide different perspectives, and our training and our life experiences as whole people can be really valuable.

Lloyd: If you’d like to learn more about the White House Fellowship or find links to Lav Varshney’s work, check out our show notes. Please subscribe to The Ongoing Transformation wherever you get your podcasts, and write to us at [email protected]. Thanks to our audio engineer, Shannon Lynch, and producer Kimberly Quach. I’m Jason Lloyd, managing editor of Issues. Tune in on October 8th for an interview with Jennifer Jacquet about octopus farming.

Your participation enriches the conversation

Respond to the ideas raised in this essay by writing to [email protected]. And read what others are saying in our lively Forum section.

Cite this Article

Varshney, Lav and Jason Lloyd. “Lav Varshney Connects AI Research, Executive Policy, and Public Service.” Issues in Science and Technology (September 24, 2024).