February 16, 2023
What is a "policy entrepreneur"? Can people become policy entrepreneurs if they're not already a political office holder? Aside from literally speaking to the POTUS, what are some ways that policy entrepreneurs can make progress on their goals? Why is it so hard for some people to articulate actionable plans that would accomplish their goals? What is market shaping? Why do some government departments have no budget for R&D?
Tom Kalil is Chief Innovation Officer at Schmidt Futures. In this role, Tom leads initiatives to harness technology for societal challenges, improve science policy, and identify and pursue 21st century moonshots. Prior to Schmidt Futures, Tom served in the White House for two Presidents (Obama and Clinton), helping to design and launch national science and technology initiatives in areas such as nanotechnology, the BRAIN initiative, data science, materials by design, robotics, commercial space, high-speed networks, access to capital for startups, high-skill immigration, STEM education, learning technology, startup ecosystems, and the federal use of incentive prizes. Follow him on Twitter at @tkalil2050.
JOSH: Hello, and welcome to Clearer Thinking with Spencer Greenberg, the podcast about ideas that matter. I'm Josh Castle, the producer of the podcast, and I'm so glad you joined us today. In this episode, Spencer speaks with Tom Kalil about policy entrepreneurship, market shaping, and cultivating agency.
SPENCER: Tom, welcome.
TOM: Happy to be here.
SPENCER: Now I'm really interested in the way that you look at the world because I feel like you have an approach to a totally different sort of ‘entrepreneurship' than I've ever heard anyone talk about before, and you just kind of approach problems really differently than I do. And so yeah, I'm really excited to dig into that with you.
TOM: Looking forward.
SPENCER: So why don't you start by telling us a bit about the work you did for Presidents Clinton and Obama?
TOM: Sure, yeah. So I had the privilege to work for both of them for a total of 16 years. I worked for two different parts of the White House: the National Economic Council and the Office of Science and Technology Policy. And my job was to interact with science agencies, the university research community, startups, large research-intensive companies, and investors to see if there were ideas that were bubbling up from those communities that, at least in my view, were worth the President knowing about and getting behind in some concrete way. And I'm happy to provide some examples. That would be interesting.
SPENCER: Yeah. Why don't you give a couple of examples?
TOM: Yeah. So in the late 1990s, I started to interact with the research community in nanoscale science and engineering — and as many of your listeners may know — if you make something small enough, not only is it small, it begins to have novel and potentially useful properties. And I was able to convince President Clinton that he should launch an initiative in this area. He gave a speech at Caltech, in which he did so and committed to doubling federal funding. And it was great that he did that at Caltech because he could talk about Richard Feynman. There's plenty of room at the bottom of speech which I believe Fineman gave him in the 50s. So that initiative has resulted in a $38 billion investment in nanoscale science and engineering. And important things are coming out of it, like the nano lipid work that was important for the COVID-19 mRNA vaccines. So that's one example. When I worked for Clinton, I was able to get the DARPA prize authority, which they used for a series of competitions around self-driving cars. And then when I came back into government, during the Obama administration, I was able to work with the Senate to get every agency prize authority for up to $50 million. And if you go to challenge.gov you'll see well over 1000 institutes in which agencies have started to use incentive prizes to promote innovation that is related to their agency missions and problems. I was able to work with the neuroscience community to launch something called the BRAIN Initiative, which aspires to do for neuroscience what the human genome project did for genetics. So as you know, the human genome project not only sequenced the first human genome but drove down the cost of that from 100 million to now rapidly approaching $100. And the BRAIN Initiative is investing in tools that will hopefully transform our ability to understand how the brain encodes and processes information. But probably one of the most important things I did was to help a number of other primarily early-career folks learn how to become a policy entrepreneur. So I had a team of 20. And they accomplished amazing things and are continuing to do great things after the Obama administration.
SPENCER: Right. So you mentioned this idea of a policy entrepreneur, which I don't think is one that people will be familiar with. So why don't you tell us what that means to you?
TOM: So in the same way that an entrepreneur in the commercial world will often identify some unmet need in the marketplace, and try to develop a product or service that addresses that need, a policy entrepreneur will identify an area of public policy that could be improved. So it could be something that the government is doing now that it should stop doing or something that it isn't doing that it should be doing. So for example, you see a lot of people who are urging the government to improve pandemic preparedness for future pandemics. And a policy entrepreneur is someone who's capable of taking those ideas, and actually getting something to happen in the world. Because I would say, there's a fair amount of — particularly the US political system where we have checks and balances, and power is often widely diffused — actually getting something to happen in the government is non-trivial. And in the same way that a commercial entrepreneur has to learn how to raise money, hire people, work on product-market fit, and get customers, there are a bunch of skills that a policy entrepreneur has to learn about going from an idea to something happening in the world.
SPENCER: So many people I know avoid politics because they think of it as this sort of unmovable gigantic entity, right? And then there's like, "Oh, yeah. Well, sure. It'd be great if the US government did things better," but they don't even know where to begin. So are usually these people insiders that already are working within the system or is there actually a way to do it from outside?
TOM: Yeah, I think it is definitely possible to do it from the outside. I think, though, that if you are going to do that, you definitely need to team up with someone who is going to be the policy entrepreneur who's working within the system. So I think it is possible to, for example, come up with an idea, and answer the types of questions that policymakers in both the Executive branch and Congress are likely to have built a coalition of people who are supporting that. So, I do believe that there's a lot that you can do on the inside, to help inform and shape policy. But if you are working on the outside, you're gonna have to have one or more partners that you're working with on the inside.
SPENCER: As someone who doesn't know very much about politics, it seems like such a black box of how stuff actually gets decided. And, you hear about things like lobbying, "Oh, you can hire a lobbyist to try to push for something," but I imagine that there are many more ways to work within the system that maybe people on the outside — maybe you could share an example of like how you actually went about getting something to occur on the inside — that actually end up moving a lot of money?
TOM: Yeah. So one way to think about this is that the executive branch which is part of the government that I worked for, I worked for the White House, has five tools that were certainly the ones that I focused on. So one is that the President can work with Congress on legislation. So for example, when I said that I worked on getting every agency the authority to do incentive prizes for up to $50 million. That required working with Congress because, in that instance, we actually wanted a new law that the Congress would pass and the President would sign. The second is the preparation of the President's budget. So when I said we invested $38 billion in nanoscale science and engineering that required increasing the budgets of agencies like NSF, the Department of Energy, NASA, and the NIH. So both the President had to propose it in its budget, and then the Congress had to go along with it. So the budget is the second tool. The third is executive action. So these are things that departments and agencies can do with the legislative authority that that Congress has already provided them with. So a member of my team worked on high-skilled immigration. When it was clear that Congress was not going to pass legislation, he was able to work with the Department of Homeland Security on something called the International Entrepreneur Rule, to create a pathway for immigrant entrepreneurs who wanted to come to the United States and create jobs and economic growth in the US. And that did not require an act of Congress. The fourth is leveraging the President's bully pulpit and his ability to convene. So when President Obama said, "Hey, computer science, and more broadly, computational thinking is really important and very few high schools offer it, I want to change that." There was no single person within the US government who could do that. We had to build a movement around that issue and that required reaching out to companies, governors, superintendents, chief-state-school officers, foundations, and nonprofit organizations. So the President has the ability to invite people to meet at the White House, issue a call to action, and engage in what President Obama used to call an all-hands-on-deck effort. That is there's no single individual organization that's in a position to solve a problem. And K-through-12 Education is like that because in the US, there is no minister of education, who can define the national curriculum, there are 15,000 different school districts. So in that situation, you essentially have to build a movement if you want to make progress. And then finally, the President can recruit people or even new types of people to the government. So after the failed rollout of healthcare.gov, the President created new organizations like the US digital service and recruited people who had skills in software engineering, human-centered design, user research, product management, data science, and cybersecurity. So that the government could get better at creating citizen-facing digital services. So, part of what is involved is not only articulating the goal or the problem that you're trying to solve but what is the specific thing that the government or maybe someone in the private sector needs to do differently than they are currently doing today. So I think that the most important is being able to have a relationship between ends and means in the same way that — you have an idea of how to leverage behavioral science to solve some important problem — you might say, "Well, the very next step is I'm going to do a bunch of A-B tests to figure out what works. And then if that works, then maybe I might hire a CEO and spin it out." Right? So, you have a playbook for getting things done in the world. And policy entrepreneur has to do the same thing. It's just a different set of tools than to be testing and recruiting people.
SPENCER: I find it so interesting because it's so different from how I look at the world. I think the way you look at the world — tell me if I'm wrong about this — through the lens of, you've got this giant machinery already operating, "How can I use this machinery to accomplish something important?" Whereas I think of it as on the margin, "What can I do without having to get permission from someone else without having to make sure someone else behaves differently other than, say, people on my team?" And it just kind of leads to this very different perspective. Do you think that's a fair characterization?
TOM: Yeah, I think people say, "What is the difference between operating the government and what I'm currently doing?" which is serving as an adviser to philanthropists in my particular case, Eric and Wendy Schmidt. And I would say, "The difference is that it can take longer to get things done in the government, that is generally the case, but when you do get the government to do something, you can have an impact on a larger scale." So it's difficult for me to imagine another position that would allow me to design and launch something at the scale of the National Nanotechnology Initiative, for example.
SPENCER: Yeah, and it's really amazing some of the things you've accomplished through this method. I wonder if listeners will think, "Well, sure, you've got the ear of the President. That puts you in this sort of special position to accomplish incredible things." But what can someone do if they don't have to hear the president? How would someone who's new to this even think about approaching these topics?
TOM: Yeah. So it is definitely the case that I was in a position where I could send the President a decision memo and have them check the box that says ‘yes', and not everyone is going to be in that position. But I think that you can also do these things through multiple levels of indirection. So you may not have the individual ability to get a decision memo to the President, for example, but you might be able to find someone who knows someone else who has that type of relationship. So you may not be one degree away from the president, but you might be two or three degrees or the issue that you're interested in, it actually doesn't get me to go all the way to the president, right? So there are lots of decisions that are made at the level of the cabinet and the sub-cabinet, for example. Congressional staff, many of whom are in their early and mid-20s, can have a significant impact on public policy, particularly if the member that they work for is the chair of an important committee or subcommittee within the Congress.
SPENCER: This reminds me of business-to-business sales, when trying to sell to a large organization, it's usually advisable to try to map out who the key decision makers are and think about, "Okay, well, who's going to be using the product that I'm selling? Who's going to be approving it for us? Who's going to be actually allocating the budget? What's the process like to allocate that budget?" and so on. So it's like a stakeholder mapping. And it sounds like you're suggesting something similar. It's like, you want to change something in the government and you kind of think about, "Well, who's actually involved in making that decision? And who can talk to the person making answers?" and so on. Yeah.
TOM: Exactly. Yeah. And specifically, what is it that you want them to do? So a large percentage of the time, someone would meet with me when I was at the White House, and we would have a conversation like the following. They would say, "Tom, you're never gonna believe it. But it turns out that the thing that I work on is super important." and I would say, "Alright, let's say that, for purposes of today's meeting, I'm prepared to stipulate that, what is it that you want me to do?" and they would say, "Well, you should make it a priority." and then I would say, "Well, yeah, but what would that look like?" And so in a lot of cases, people were not able to articulate what it is that they wanted the administration to do, for the reason that you mentioned at the beginning which is the government to them as a black box. And so I came up with the following thought experiment to help them try to think about this, which is, you have a meeting with the President in the Oval Office. And he says, "Spencer, if you give me a good idea for increasing our use of behavioral science to solve important problems, like depression or anxiety, then I will call anyone on the planet. And it can be a conference call. So there can be more than one person on the line. And if that person is inside the government, like the head of NIH, I can direct them to do something because I'm their boss. And if it's someone outside the government, like the head of a foundation, or the president of the university, then I can challenge them to do something. And all you have to do is who should I call and what should I ask them to do?" And sometimes that would get people to think in a more concrete and less abstract way because the "government is not always a useful level of abstraction." It's really particular departments and agencies, either operating individually or collectively or some sort of external stakeholders in the private sector that the President is seeking to mobilize?
SPENCER: I really like that question, in part because it forces concreteness. And also in part, because if you can't get it, having the President do something for you, it doesn't make progress on your issue in the government, then how do you make progress? Right? It is not an easy question to answer, but I think it's a fruitful question to ask.
TOM: Yeah. Also, because, at least for the types of problems that I've worked on, they have required building coalitions. So they haven't been the type of thing that I could do by myself. And it's very difficult to build a coalition if you can't, at a minimum articulate, number one: who are the members of the coalition? And number two: what are the mutually reinforcing steps that you want them to take? And after you know the answer to that, then you can say, "Why is it in their enlightened self-interest? How can I make it as easy as possible for them to say ‘yes'? Who is the ideal messenger?" Maybe there's one member of this coalition that is willing, but not able, and they are operating under some constraint, and some other member of the coalition can relax whatever constraint they're operating under. So, but at a minimum, you have to be able to say, "This set of actors would need to do this thing differently than they're currently doing today. And it's plausible that if the right person called them that they would be willing and able."
SPENCER: I wonder when you're picking the topics to work on. Do you try to avoid ones that you expect there to be other groups that fight? Because they're politicized? Or because there's gonna be opposition from some group that loses out? Or is that not a major consideration for you?
TOM: I'm definitely thinking about some version of the expected return. In the sense of, if this initiative were successful, what is the potential upside? And then what is the probability of success? There was a famous baseball coach, whose name I'm forgetting, but he used to tell his players to hit him where they ate. And the National Nanotechnology Initiative was a great example of that because there was no one on the other side. In 1999, and 2000, virtually no one in the US government, outside of the science agencies, knew what Nanoscale Science and Engineering were. So there was no opposition. Eventually, there were some non-profit organizations that would raise some concerns about emerging technologies, including nanotechnology, but at the time, there was no organized opposition. So it is possible to make more progress if you're not in an area that is heavily contested. What I worked on the National Economic Council in the 1990s, for President Clinton and worked closely with Vice President Gore. Vice President Gore was trying to get the Congress to pass the 1996 Telecommunications Act, and there was this big fight between different parts of the telecom industry. And that really constrained the range of policy options, because you had to come up with some deal that would reflect the interests of the various parties. And that reduced the range of solutions that could be seriously entertained.
SPENCER: Are there other heuristics you used to think about the probability that something will succeed in government beyond the one we mentioned, is there going to be an opposition to it?
TOM: Yeah, obviously, things that are incredibly expensive are just more difficult to do. So that's something that I think about. I think about whether the executive branch has the capacity to implement. So I think that's something that policymakers don't think enough about. They tend to have a background in law and economics and working on campaigns or working on the hill or working at Think tanks. You don't see people with a — what we would call in the private sector — an operator background, a COO, or someone who reports to a COO. But the longer I was in government, the more I recognized that just signing a bill or having the President sign an executive order was necessary, but not sufficient. And so I gave a lot more thought to the question of, "Okay, let's say that I could get the president to prove this. Are there people within the US government who would have the capacity to implement this? And if they're not, either, maybe I shouldn't do it. Or I should make sure that the government is recruiting people who have that relevant skill set and background."
SPENCER: So anyway, another idea you think about is market-shaping? Can you tell us about what is market-shaping and why you think it's underrated?
TOM: Well, there are lots of problems that the market solves. And so I don't lose any sleep about enterprise software. Because if someone has a good idea for enterprise software, and it has a high CAC-to-LTV ratio, they're not going to have any difficulty raising investment capital from early-stage investors and hiring people, and developing the product. There are other types of problems that the market does not do a good job-solving. And a canonical example of this is vaccines for diseases of the poor. So left to their own devices, drug companies will not work on vaccines for poor people, because they have no money. And it would be very difficult for them to justify to their shareholders that they should spend a huge amount of money on R&D and manufacturing in solving these problems. So economists like Michael Kramer, who recently won the Nobel Prize, came up with a very clever hack for solving this problem, which is a purchase order for a product that doesn't exist yet. And this is known as an advanced market commitment. So as a result of his work, and the work of others, five countries, and Gates Foundation worked with GSK and Pfizer. And they said to those companies, if you develop a vaccine, which is safe and effective, then we commit that we will provide a co-pay for this many million doses at this price per dose. So as opposed to seeking to increase the supply of innovation, by supporting R&D through grants and contracts. Here, you're trying to influence the demand for innovation. Another example of this is the NASA-SpaceX collaboration. When we retired the space shuttle, we had to pay the Russian government over $50 million per astronaut to proceed, anytime we wanted to send an astronaut to the International Space Station. So NASA said to SpaceX, we want a rocket that will go to the International Space Station, deliver and retrieve cargo and ultimately crew. And not only will we buy rides on the rocket, but we will also provide you with intermediate milestone payments toward that goal. And then NASA got access to a capability in the form of the Falcon9 rocket for a fraction of what they would have traditionally paid in a more business-as-usual approach. So there are two things going on here. One is that the sponsor of the market shaping intervention is articulating what they want, what problem are they trying to solve, and what is the outcome, and then they are making a financial commitment that is contingent on success. So they're saying, "If you develop a vaccine, and it's safe and effective, we will buy this many doses at this price per dose, or we will provide you with milestone payments for intermediate progress towards some goal." And what I think is really problematic is that right now, the federal government has a well-established and widely utilized mechanism for making financial commitments that are contingent on failure. Those are loan guarantees. So that's the government saying to an individual firm, if you go bankrupt, the US government will assume your debt obligations, but we're barely scratching the surface on our ability to make financial commitments that are contingent on success. And I would like to continue to work on changing that but before I retire.
SPENCER: Why that asymmetry?
TOM: Well, these are relatively new ideas. So they haven't been around for all that wrong. I think the other thing is that we currently have some things in the budget and appropriations laws that make it more difficult for the government to do this. So a lot of times when the Congress provides an agency with additional resources, they're told that they have to spend that within a two-year period. And the whole purpose of doing one of these advanced market commitments, for example, is that you don't know whether the private sector will ever solve the problem. Or even if they do solve it, when will they solve it? So that's a problem. And then I think we also have a shortage of people who know how to design these types of interventions. Whether it's incentive prizes, milestone payments, advanced market commitments, or a whole suite of these approaches where you're trying to increase demand for innovation, as opposed to increasing the supply.
SPENCER: It seems like they'd have to be designed really carefully to avoid gaming them.
SPENCER: Like, if you set the wrong milestone, then someone's gonna build something that technically satisfies it. But it's not really usable.
TOM: Yeah, we see that with public policy all the time. So when we're trying to hold someone accountable, so we were like, "Okay, we want to improve K-through-12 education. And we're going to evaluate teachers on the basis of the performance on standardized test scores," that can lead to teaching to the test, cheating, or things that are more subtle, like teachers, focusing on those students who are just on the bubble, and are close to being proficient but not quite proficient, as opposed to those that are further behind or gifted and talented students that the teachers are no are going to be proficient no matter what they do. So yeah, you could definitely have a good hearts law phenomena, where you set some goal, and it has some sort of perverse impact. So I think God is definitely in the details in terms of how to do this. Another problem is getting the level of difficulty, right? So on the one hand, you want to stimulate innovation, and set an ambitious goal, but you don't want it to be so ambitious that nobody solves it. Or instead of that you're providing is not sufficiently large to induce the private sector to solve the problem.
SPENCER: I've wondered if this could be a really good approach for climate change, where the government could have extremely large prizes for a series of milestones around technological advancement. I wonder what you think about using it in that way.
TOM: One thing I'm very excited about is a collaboration between the government and the parts of the value chain that are going to be purchasing that commodity. So what do I mean by that? Well, one of the things that we need is a low carbon or carbon-neutral steel, the US government doesn't buy a lot of steel, but the auto industry buys a lot of steel. And so the Biden administration has been encouraging companies to participate in things like the First Movers Coalition. And these are groups of large companies that buy steel, sustainable aviation fuel, carbon-neutral shipping, or carbon-neutral aluminum. So I definitely think that these sorts of demand-pull interventions make a lot of sense. And I think there's some question about what the right way to do it is, and I'm really interested in those people who are looking at this from a value chain perspective because I think they're identifying ways to leverage private sector demand in those cases where it's not really the government, that is the purchaser.
SPENCER: Are there other approaches to market shaping you've explored?
TOM: Well, so there's an idea that is in the global health community that I think should be more widely adopted. And they have the following methodology, which is they will identify a problem, like newborn mortality in Sub-Saharan Africa and other developing countries. And then they'll break that down and they'll say, "Okay, like, why are newborns dying?" And then they'll say, "Well, is there a medical device or some other type of innovation that would make a big difference?" So for example, if many newborns die from jaundice, what is the low-cost, rugged medical device that would solve that problem? And then they don't tell you to build it. They just tell you what it needs to do. They tell you they provide information using a Delphi technique of experts and stakeholders of what are the performance characteristics that such a device would have and what would it need to be priced at in terms of both CAPEX and OPEX for it to be affordable in a developing country context. So this idea of what they call Target Product Profiles is something that I would like to see much more adopted more widely in other problems, not just global health. Because I think a lot of times, the government isn't as explicit as it should be, about what success is. So it just says, "Oh, we want you to do stuff on workforce development, or something like that." but they're not saying, "Hey, this is the goal that we would like people to achieve. And this is how we would know whether or not you've done it or not." But ultimately, what I would like to see is a marketplace for outcomes that has four types of participants. One is the sponsor of the outcome, which might be operation warp speed, the part of the federal government that said to the private sector, "If you develop a COVID vaccine, we'll buy 300 million doses," then there are the teams that believe that they can achieve that goal. So that's the second type of participant. The third is the investors who are backing those teams, which could either be commercial or philanthropic. And the fourth is the subject matter experts who are designing the contracts and these market-shaping mechanisms that are going to allow progress to be made. So I think the other value of these approaches is, it does require people to be more explicit about what problem are they trying to solve and how would they evaluate an innovation that would represent real significant progress. And you would feel good about investing in a set of milestone payments, a prize in advanced market commitment, whatever the mechanism is.
SPENCER: Can you talk a bit more about how investors can be involved? Because I think that's a really interesting angle on this.
TOM: Well, you definitely saw that with SpaceX, right? So this is one of the things that allowed SpaceX to finance the development of the Falcon9, they were able to go to investors and say, "NASA has told us that they're going to provide us with these milestone payments, and by rides on the rocket." Stripe, Meta, Alphabet, McKinsey, and Shopify recently did the same thing with permanent carbon removal. They launched something called frontier climate, which a Stanford economist, Susan Athey, that Schmidt futures has helped to design, but they made a $925 million commitment to purchase permanent carbon removal, this has had a transformational impact on the ability of carbon removal startups to raise capital because prior to this when they would go to VCs, they would spend the first 45 minutes of the conversation trying to explain where the market was. So the VCs would say, "Well, who is going to sell this to?" And that was a difficult question for them to answer in the absence of something like frontier climate. So that's another example. And obviously, for companies like Pfizer, they're able to self-finance. So when Operation Warp Speed said, "Hey. If you develop a COVID vaccine, we'll buy 300 million doses," they were probably able to use retained earnings in order to fund the very large investments that they had to make in manufacturing to scale up manufacturing, but it increased their comfort level that there was going to be a customer on the other end. So in essence, the government or the sponsor of the AMC is saying, "We will bear the demand risk if you bear the performance risk." And I think that's a good allocation of the different types of risks.
SPENCER: My understanding is that there's another model involving investors as sometimes proposed, where the investors can actually collect the prize money. So basically, they're giving the company investment capital to work on this prize or work on this demand order. And then if the company succeeds, the investor can directly get some of that prize money or the revenue from fulfilling the order. I'm curious about what you think about that model.
TOM: Yeah, I think these are all worth exploring. And I think we need just a lot more experimentation. So that we begin to develop some heuristics about when, and under what circumstances you should use these different approaches. So one idea we had during the Obama administration is something that we called the Innovation Toolkit. And what we meant by that was that there are lots of different approaches that the government could use to try to solve problems and foster innovation. We could make government data available in a bulk downloadable machine-readable format. We could use agile and human-centered design approaches to software development. We could use incentive prizes. We could use the DARPA model for supporting high-risk, high-return really transformational R&D. We could try to recruit new different types of people to the government. We could try to encourage the use of randomized controlled trials to figure out whether a given intervention worked. And the reason that I liked the toolkit metaphor is that you don't want to be that person who has a hammer and is looking for a nail to hand. You want to have a hammer and a saw and a screwdriver and a tape measure. So given the nature of the problem that you're trying to solve, you have the right tool or combination of tools. And what I think is important is that over time, we could begin to develop a sense of what tools we should use for a given job. And no one is going to know about all of them. But hopefully, they would have some heuristics about, "Oh, when should you use this particular approach?" I don't know if you ever read the book called A Pattern Language is the architect who passed away recently. The reason he calls it a language is that it's the set of problems, and solution pairs in architecture that go all the way from the knob on the door to the use of natural lighting in a room, all the way up to the role that a Street Cafe plays in promoting serendipitous interaction within a community. And that you have these problems, solution pairs that are accessible to you, as an architect who is worried about the built environment, and what is the equivalent of that for policy entrepreneurs or for what some people call tri-sector athletes, that is people who work across government, industry and civil society.
SPENCER: You mentioned earlier, you're working in philanthropy. Could you tell us a bit about what you're doing there and how it differs from you working for the government?
TOM: Yeah. So I serve as the chief innovation officer for a philanthropic initiative supported by Eric and Wendy Schmidt called Schmidt Futures. Schmidt Futures has a number of priorities, a major focus is identifying talent and supporting exceptional people seeking to make the world a better place at different stages in their careers. So we have a program called Rise, which is about supporting exceptional 15 to 17 year olds, something called the Schmitt Science Fellows Program. These are people who have gotten a Ph.D. in one discipline but want to do a postdoc and another. Innovation Fellows Program, which is our mid-career program. So that's a lot of what we do is finding exceptional people and betting early on their ideas. And then we are also making really interesting investments in science and technology that I'm happy to talk about.
SPENCER: Yeah, I'd love to go into that. But before we do, I'm curious about this focus on talent. Can you tell us a little bit about why you think talent is so important?
TOM: Yeah, I think that it's just a way in which philanthropic organization, I think, can have an outsized impact. So, a lot of times — and we do some of this, too — we're identifying a specific problem that we're trying to solve. But I think there's something very powerful about asking ethical, smart, intrinsically motivated people. Well, what problem are you trying to solve? And how can we help you solve your problem, as opposed to saying, "We work on problem A, in Region B, using the theory of chain C?" And unless you're directly aligned with that, then we're not interested in talking to you. And, ultimately, I do think it is the case that if we can bind and support the right people, not only individually, but as part of a network — I know a lot of our programs are cohorts of people, not just supporting people one at a time — that that can have a really outsized impact.
SPENCER: So this suggests a sort of market inefficiency, if you will, in resources and help to go to really talented people, right? Because one worldview might say, "Well, there are super talented people, they are gonna get a lot of resources and are going to meet other really talented people and be in the right networks." But this suggests that, "Oh, it may be actually there under resource, a lot of really talented people, and maybe they aren't meeting the people they need to make their ideas have legs," and so on.
TOM: Yeah, I think, particularly at a very early stage when they have the idea, but they don't have a lot of evidence to conclusively demonstrate that their idea is going to work. So I think Eric and Wendy are willing to bet early on their ideas, which I think can make a big difference. I'll give you an example. So one of our Innovation Fellows is out of marble stone. And he had been a researcher at the church lab, at Biden's lab at MIT, who's a neuroscientist at DeepMind. And he and one of his colleagues, Sam Rodriguez, had this really interesting idea, which is that there's a class of scientific and technical problems, that is not a good fit for startups, because there's no path to profitability on a timeframe that is attractive to VCs, or the whole purpose of the project is to produce a public good, like produce a giant data set and put that data set into the public domain. If you went to a VC with that idea, that would be a very short meeting, because they would say, "Dude, how am I going to get my money back?" but for reasons that are a little more subtle, some of these projects are also very difficult to do in a university setting. And that's because if you are an individual graduate student or postdoc, you are appropriately worried about your own career, the way you advance your career is by demonstrating your unique intellectual contribution to an idea that you might do through a sole author paper or first author paper. And so it's very difficult to have a situation where 20 people in an academic environment are all rowing in the same direction. So they had seen a number of projects, where it was clear what the next thing to do was to help scale something up in neuroscience. But they didn't see a path to getting that done. And I encouraged them to write something up on that idea. And then Adam said something very important to me, he said, "I would be willing to leave DeepMind and spend the next 10 years of my life trying to get this off the ground." And so that was a pretty strong signal to me that he was serious about the idea. So Schmidt Futures gave him support under the Innovation Fellows Program. He and I interviewed well over 100 scientists, engineers, and entrepreneurs, and asked them, "If this funding mechanism existed, is there a key bottleneck in the field of science and technology that you're familiar with, that this would help address?" And when we had a critical mass of ideas, we discussed this with Eric and Wendy and they ultimately decided to support two of these focused-research organizations, and a list of other ideas that are now bubbling up from the research community now that people know that the model actually exists and is getting some momentum. So that's why I think there's a real value in being in the world, identifying people who have an idea that I call important and true, and seeing if there are ways for philanthropists and others to support them. I think evidence about the importance of this was...did you happen to read the survey that the past grants team did have the biomedical researchers that they supported?
SPENCER: I did, yeah, it's kind of shocking to explain that.
TOM: Sure. So these were researchers from some of the top research universities in the world. These were people at Berkeley, Stanford, MIT, and places like that. Patrick Collison, Eric Schmidt, and a number of other philanthropists provided funding for something called post-grants at the beginning of the pandemic. And the theory was very straightforward, which is, "Hey, we're in the middle of a pandemic, we should make decisions quickly, and try to make them in ideally 48 to 72 hours as opposed to months." And at some point, they did a survey. And they said, "Imagine that you had the same amount of money that you currently have. But if you had total discretion about how to spend it, would you change your research program, A. Not at all, B. A little bit, or C. A lot?" And I believe that the percentage who said a lot was 78%, which is shockingly high. And as a former policymaker, what comes to mind for me is, number one: what would they be working on, if they did have more discretion? And if you were to come up with a taxonomy of the reasons why they can't work on the thing that they like to do, what are the principal reasons? So I think that is worth a lot more exploration. So for example, one of the things that you hear pretty consistently is that the peer review system is good for screening out really bad ideas. But it also screens out ideas that are high-risk, but high-return, because NIH can only find, say 10 to 20% of the proposals that they get. So it just takes one person on the committee to say, "Well, I don't think they have enough preliminary data to demonstrate that they can do what they say they want to do." And so now researchers joke that they have to have done the experiment before they write the grant. So that's clearly one reason why some of these proposals are not getting funded. But I'm sure there are others as well.
SPENCER: I heard some statistics recently, I'm gonna forget the exact details. But it's basically one of these federal agencies bragging about how they fund high-risk research, and also 80% of research projects succeed. And it's like, "Well, okay, I think something's gone wrong here."
TOM: Yeah, I think one issue is, within the private sector, particularly within VC, there is an understanding of power laws that you're counting on a very small number of your investments to return the fund. It's more difficult to have a very high level of failure within the government because we have a two-party system. And therefore, whichever party is not controlling the Executive branch is going to criticize the administration for allowing these failures to occur. So I think that is one problem, but we have to come up with mechanisms that allow us to support this.
SPENCER: One thing that I think is really interesting about the ideas you're talking about is that they can inspire people to sort of dream of things that they normally wouldn't even consider, like, with FROs. A scientist would never normally even think about the possibility of having a large amount of money to spend five years just trying to solve some scientific problem with a team of scientists, in the same way. Did you want to just comment on that, like this idea of opening up new possibilities?
TOM: Yeah. So because of the positions that I've been in, I've developed this very strong sense of agency. And the way I describe this is, I could be in a situation where someone gives me a brilliant idea. I spent some time evaluating that idea. I send a decision memo to the president, and he checks the box that says ‘Yes', he mentioned that in the State of the Union address. And then in a relatively short period of time, there could be a ballroom in a conference hotel, which will have people working on that idea. And so if that happens enough times, you begin to develop a very strong sense of agency. That is, there are more things in the world that are at least potentially changeable because they're the result of human action or inaction. And so a lot of times, what I find is, when I'm talking to someone, and they have a kernel of an idea, there's something that they view as fixed, and I view as potentially changeable. And so one of the things I try to do is to see if I can make my sense of agency contagious, or for it to be shared, as opposed to just a feeling that I have, but also a feeling that they have. And so an example of that is the reason that I believe that the concept of focused-research organizations, even though it's just getting going, could be so generative, is that researchers were not asking themselves, what would I do if I were the CEO of a $50 million research non-profit, where I could do more ambitious projects, projects that require a high level of team science, projects that are really important for the field as a whole, but might not generate a huge number of applications that might generate a huge number of publications. There's some other reason why it would be unlikely to get done using a status quo funding mechanism or institution. And then, Schmitt Futures can credibly say, "If your idea is good enough, this is something that we might be willing to support." And that just opens up a much broader range of possibilities for a scientist, an engineer, or an entrepreneur to consider. And I'm also interested in doing this at the intersection of AI and science, and happy to talk about that as well.
SPENCER: Sure, yeah. Well, where do you see the link there? And what gets you excited about AI in science?
TOM: Well, obviously, the AlphaFold 2 result was pretty impressive. And so the question that I've been asking researchers is, number one: Do we understand the preconditions for AlphaFold 2? And number two: Are those preconditions, something that we could more intentionally try to create in other fields?
SPENCER: Could you just take a moment to explain AlphaFold?
TOM: Sure. So there were three things that needed to come together. Number one is you had this very well-defined problem, how do you go from amino acid sequence to three-dimensional protein structure? Number two, you had a very large public dataset called the protein data bank that the research community had been contributing to for decades. And number three, you had a benchmark for evaluating the performance of different algorithms for protein structure prediction. So then DeepMind had a team of 18 really smart people with access to amazing computational resources. And they were able to build on those three building blocks to generate this breakthrough in our ability to do protein structure prediction. And they have now put into the public domain hundreds of millions of predictions of protein structure. So that field is now sort of supercharged, in terms of its ability to make progress, and you're seeing lots of advances that build on that. So what I'm interested in is, there is something about the protein-folding problem that is sort of soul generous, that's just going to be very difficult to do in other fields. Or if we thought the right people were thinking about it in the right way, are there similar opportunities in chemistry, material science, condensed matter physics, and engineering? So here's one example. I believe that when researchers are thinking about the relationship between machine learning and science, they might ask themselves, "What data can I already get access to?" Or number two, if I wrote a grant to the NSF or the NIH to support my lab, what is the maximum size of the data set that I could create? They're not thinking about what data set would transform the field. Because if the cost associated with that has too many zeros after it, they're gonna say, "Well, no one is ever going to give me, an individual researcher, all that money to create a dataset of that size." So again, researchers don't spend a lot of time answering your question, if they think that the expected return from answering that question is really low. So I think what a philanthropist could do is say, "No, no. We're really interested in the answer to the question, what data set would be transformational for a given field, And if the case for doing it because it is a public good, it's strong enough, this is something that we might be willing to support," And philanthropists could also support innovation tools that lower the cost of collecting the data or investing in things that are called self-driving labs, where you're not creating just a static data set. But you're using active learning so that the machine learning is not only analyzing the data, it's informing what experiment to do next. So I think there's some really exciting opportunity at the intersection between machine learning and scientific and technological progress. I think in many cases, large public datasets are a bottleneck to doing that. And I think researchers may not be motivated to describe the ideal data set that they would like because they don't see a path to getting that funded. So I think there may be some areas where by posing the question, by having a credible commitment to fund the best ideas, we could get researchers to answer questions, where there's a really actionable idea on the other end of that, but they're just currently not motivated to do so. So it makes me wonder what other questions are like that, that are generative. If someone did have a good idea, it's actionable and potentially high-impact, and they're just not motivated or incentivized to answer that question because they don't think it would lead to anything. In the same way, at the beginning of our conversation, you said, "Well, people may not get motivated to engage in public policy, because even if they did have an idea, they just don't see a pathway for how they could go from that idea to Congress passing a law or the President signing the executive order or an agency reforming some key process."
SPENCER: Tom, before we wrap up, how about we do a quick fire round? We'll ask you a bunch of questions, you give your quick thoughts on them.
TOM: Sounds good.
SPENCER: All right. Great. So first question, you've done so many different projects over your career, what's one that comes to mind that you're really proud of?
TOM: The team of people that I recruited to OSTP, and not only what they did when they were at OSTP, but what they're doing now. So, expanding access to computer science, improving high-skill immigration, and reducing or eliminating the waiting list for an organ transplant. So not only the progress that they made when they were working for President Obama, but the fact that they're out there crushing it, because of skills that they learned around things like coalition building, influence with authority, policy, entrepreneurship, getting people excited about their ideas.
SPENCER: Does that stand for Office of Science and Technology Policy?
SPENCER: Great. So because you've been involved in so many projects, I'm interested in the themes that you see that have kind of cut across many different things you've done and what are some of those themes?
TOM: Yeah, so one is I'm very interested in general-purpose technologies. So this is the concept that economists have, which is that there are some technologies, electricity, interchangeable parts, the transistor, and the internet, that doesn't just result in one product but result in transforming our economy and driving higher levels of productivity and economic growth and job creation. So that's one area that I'm very interested in. Another is that I am very interested in approaches to solving problems. And one of the things I've discovered is that communities will come up with particular approaches, in the same way that the global health community said, "Oh. If there's a market failure, someone should issue a purchase order for a product that doesn't exist." And sometimes that idea will not make it outside of that community, it'll just be stuck in that community, and there will not be broader awareness of it. So I think there are a lot of opportunities to look in a comparative way at how different disciplines, sectors, and communities of practice solve problems because I think some of those are generalizable. So I know that's kind of wonky. But that's an idea that I'm really excited about.
SPENCER: Why is it a problem that some parts of the federal government have no research and innovation capacity?
TOM: I'll give you one concrete example. The research budget of the Department of Labor is $0. It literally has no research budget. So when I would interact with DARPA, for example, which has a research budget of slightly less than $4 billion. There were people who said, "Oh. What if we could go from New York to LA in 11 minutes, and 20 seconds?" Dan Wotton Dorf, who I worked with, had this idea, "Oh, what if we could develop new vaccines in months rather than years?" He was the DARPA program manager who funded Moderna to work on infectious diseases. So I'm very glad that he did that. The Department of Labor doesn't have the resources or the mandate or the culture to say, "What if x were much better than it currently is today?" And they are responsible for working on problems that I think are important, right? So what's happening to the real wages of non-college-educated workers, they're going down. What's the impact of some of our largest workforce development programs, in some areas, they are negative. So that is, people who participate in the workforce development programs are net worse off than people who did not participate in the program. So both the problems that they're working on are important. And the current solutions that we're supporting are not all that effective. And so I think that giving more agencies a research and innovation capacity, might allow them to say, "How would we leverage science and technology to develop workforce development programs that increase the wages for non-college educated workers by $20,000, or more in under six months?" or something like that. So the thought experiment that I'm interested in is if you woke up and you were the chief scientist for an agency that currently had no research capacity and all of a sudden it did, and you were the chief scientist, what goals would you set? And what are examples of projects that you would support in order to achieve those goals? And I think that that absence has resulted in these large and important gaps in our research and innovation portfolio because there's no patron for research in that area.
SPENCER: Alright, last question for you. So what is a tour of duty in the federal government? Why do you think that people should consider that?
TOM: Our mental model is that there are two types of people within the US government: there are civil servants who might join the government in their 20s and stay until they retire, and then there are people like me who are political appointees, who, as we say, serve at the pleasure of the President, and are generally aligned with one of the two political parties. Actually, there is a third way. And that is people who decide, "I would like to serve at some point during the course of my career, but I'm not going to do it full time. And I'm not necessarily going to be a political appointee." So this is the reason why DARPA is so effective, is they're able to get top researchers from universities like Berkeley and Stanford and MIT and Carnegie Mellon to come in and serve as a program manager for three to four years and then go back to doing whatever they're doing before, like being a computer science professor at Berkeley or CMU, for example. And that means that the DARPA program managers are operating at the cutting edge, which, if you're going to serve in the government for 40 years, it is more difficult for you to do. So, that's an example. But it's also the way the US digital service works. So the people who work at US digital service are not planning on being there forever. But these are people who have worked at Google or Amazon or Stripe or Microsoft. And they say, "I'm in a position financially, where I do not have to worry about where my next meal is coming from. I care about my country, I care about my government, and I want it to work better and cost less. And therefore I'm willing to go into government and try to leverage my skills in software engineering, human-centered design, product management, or data science and solve some really important problems." So we did this on an ad hoc basis after the failed rollout of health.gov. And the President said, "Why would we wait until there's a disaster like healthcare.gov?" Before we do something like this, why don't we have these people involved at the beginning of the projects, as opposed to waiting until we have a crisis? And that really was the impetus for the creation of the US digital service. So, the policy is made by those people who show up. And I love the motto of the civic-tech community, which is, "No one is coming. It's up to us." So I have a lot of respect for people who not only identify something in the government that can be improved but don't just whine about it. They say, "Okay, over the course of my career, there is some nonzero amount of time that I'm willing to spend working in the government to make it better and more efficient and more responsive and better able to deliver on the challenges that we face whether it's accelerating the transition to a low carbon economy, or improving our ability to respond to future pandemics."
SPENCER: Tom, thank you so much for coming on.
TOM: Thank you. I really enjoyed the conversation. Great questions.
JOSH: A listener asks: Similar to how Guided Track offers a domain specific language for form building, research, and app development, what other categories of apps do you think would be an opportunity for end user facing domain specific languages?
SPENCER: I think we could have some pretty cool domain specific languages for visualization. And I think some people have done work in this direction, but I think that could be cool, like trying to visualize data or visualize results. I also wonder whether there could be value in a domain specific programming language for machine learning. I haven't seen that explored too much. And I don't know that it would add much value because it's possible. You know, developing in Python is perfectly fine. But I do wonder if people start thinking in that way and saying, "Well, if you really were [inaudible] specifically what kind of elements would you have?" Maybe there's something interesting there.
Click here to return to the list of all episodes.
Sign up to receive one helpful idea and one brand-new podcast episode each week!
Subscribe via RSS or through one of these platforms: