Kratsios interview

Kratsios: Thank you so much for joining us. I know you don't come often to Washington, so making the trip to be part of this is very special.

Collison: No, thanks for having me, and congratulations on this event.

Kratsios: Wasn't that incredible?

Collison: That was very cool.

Kratsios: That was very exciting. Well, look, as I've gotten to know you over the years, I have continued to admire and be impressed with the incredible efforts that you make to help support the American scientific ecosystem. And I know you actually have a day job, so the fact that you do this is very special, so I want to say thank you.

But I want to begin with something I heard you say on a podcast a little while ago that we had clipped up in the intro to the event, where you essentially cited a study that something like 78 or 79 percent of scientists, when asked if they would continue to pursue the science that they are funded to do — would they do that or would they do something else? And I think they said they would do something else. So why don't you unpack that for us? And what do you think that means?

Collison: So I think this is a really interesting thing for people to understand, and a very useful radicalizing number to know.

A couple of years ago, we ran a survey of top biomedical scientists — actually many people funded through Fast Grants. And there are a lot of discussions and debates in science about what level we should fund science at. Should we fund science more or less? Those are important debates and discussions, of course. But there's another question you can ask, which is: well, what science is funded for any given level? And what are the scientists permitted to do?

So we surveyed all these top researchers at top institutions. We simply asked them: forget questions about how much you're funded. If you could spend your current grant dollars however you want — so you're not getting more money, you can just spend your current grant dollars however you want, rather than in the kind of prescribed, project-based way in which you receive funding today — how much would your research agenda change?

And as you say, 78% of those scientists said they would change their research agenda a lot. I think this is extremely striking. Four out of five scientists would change their research agenda a lot if they had the flexibility to pursue a research agenda of their choice.

And we furthermore asked them, in broad strokes, what would be the character of the change in your research? And 83% — this was five years ago now, but I think it was 83% — said that they would make their research agenda more ambitious if, again, given this unconstrained freedom.

So we have inadvertently gotten ourselves into this ecosystem and into this equilibrium where some of the smartest and most self-sacrificing people in the country — they could all be earning much more in the private sector — are toiling away in their labs, producing these amazing goods for the rest of society, and we have put them in these strictures, these cages, that inhibit them from pursuing the research agenda of their choice.

That's the bad news. But I think the good news is that's also an opportunity for us. There's some low-hanging fruit there where we can really liberate them.

Kratsios: Well, given that, if you could take, I don't know, one or two policy actions to immediately start addressing this issue — that four out of five people aren't working on what they want to work on. That's crazy.

Collison: It does seem crazy. I think—

Kratsios: And not only are they — they're being paid by someone to do it. They've literally gotten a grant to work on it.

Collison: Yes, yes, yes. We, the taxpayer, are funding them to work on something, in some sense, other than what they think would be highest-impact to work on.

Kratsios: Yeah, it's inefficient.

Collison: I mean, look, I think all the mechanism design is tricky, and I don't pretend to be— Maybe at some point we'll get to Arc. Rather than trying to prescribe the answers for everybody else and pretend that I have magic silver bullets, we tried to build something in practice.

But I'd say, generally speaking, funding more young scientists, which I know has also been a focus of yours. There are many analyses and many papers showing that scientists don't tend to change their minds as new discoveries occur so much as future generations supplant and succeed them. There's a famous analysis of the age at which scientists do Nobel Prize–winning work, and even though people tend to receive the Nobel Prizes — this week — later in their careers, in their 50s, 60s, and so forth, the age at which they do the Nobel Prize–winning work tends to be late 20s, early 30s, mid-30s.

And so, generally speaking, skewing more of the research funding towards early-career scientists. A fun or not-so-fun fact on this front: in 1980 specifically — you can divide scientists into two categories. Scientists under 40, let's just arbitrarily call those early career. And scientists 50 or older, let's call those mid- or later-career. In 1980, the NIH funded 1.7x more dollars to these early-career, under-40 scientists than to the 50-and-above. So 1.7x more to the younger scientists. As of 2023, which is the latest year, I believe, for which we have data, the under-40 cohort receives 22% of the dollars that the 50-and-above cohort receives. So there's been this dramatic change over the last 45 years in terms of the composition. So I'd say, first, funding more young scientists.

Secondly, more curiosity-driven work. This is maybe sort of tautological, in that if we want more curiosity-driven work, we should simply enable people to do it by funding people — scientists who we think have amazing potential, iconoclastic ideas, who are likely to produce the future breakthroughs — rather than by trying to judge, in some consensus-oriented way, the minute parameters of any given... I mean, by definition, if you can write a proposal where you know what you're going to discover, it's at least arguably not that scientific. What's the discovery if you can anticipate it up front?

And then lastly — and this is speculative; people should try it, and this is where I think metascience can play a role — I suspect longer grants. It differs across NIH and NSF, and obviously there are different grant mechanisms and so forth, but generally speaking there's some data from Pierre Azoulay, for example, showing that HHMI grants — the Howard Hughes Medical Institute — which tend to be longer than standard NIH or NSF grants... kind of identical scientists who receive an HHMI grant, which is longer, tend to produce larger and more consequential discoveries than scientists who don't have that same project time horizon.

So I'd say: younger, more curiosity-driven, and longer time horizons.

Kratsios: Wow, it seems like you're summarizing this document called Science: A New Golden Age.

No, I agree with you there. I think one area that—

Collison: Well, let me just say on that — and none of this is prepared or scripted or anything — I'm immensely encouraged, and in some sense surprised, in a very positive sense, that we're here having this discussion, and discussing specifically some of these ideas, in Washington in 2026.

A scientist, originally a quantum physicist, Michael Nielsen, and I started doing some work on some of these questions, trying to do some of this meta-analysis of science, back in 2018. We wrote an essay about it, and we spent a lot of time thinking about it, and it was a very depressing and discouraging effort overall, I would say. Because, one, the data was discouraging, but then, two, nobody else seemed remotely interested in or engaged with these questions, and we thought they were incredibly important. The scientists are working super hard, and so it feels like the least we could do for them is to try to figure out what the optimal support structures around them would look like.

And fast-forward, here we are with these as central objects of discussion and debate. For any given issue, I don't know where we'll land, or even, on many questions, where we should land. But I think we should be centrally focused on these questions, and I commend and appreciate you orienting people in that direction. So thank you.

Kratsios: No, thank you.

Before I got into the weeds on a lot of this science policy stuff, I had spent much of my career in Silicon Valley, and an observation that the team I was working with always had was this idea of an unfortunate technological stagnation that had broadly taken over the U.S. and the world. Planes were flying slower, and the amount of stuff you could grow on an acre of land was slowing down—

Collison: TFP on its long slow decline.

Kratsios: Yes, yeah. The cost of producing a drug continued to increase, the time it took — all these indicators that things just weren't going well. And then you're like, I guess, what do we do about it?

Do you believe that we're still in this technological stagnation period? And do you think superintelligence is something that will get us out of it? Or how do you think about technological stagnation at this intersection with science?

Collison: Well, predictions are hard, especially about the future. But my best guess is we're in the early stages of emergence from the Great Stagnation. That's my best guess.

Global GDP growth rates have not yet accelerated enormously, but we are starting to see some acceleration in the U.S. We're even seeing some suggestion that labor productivity is starting to grow, and TFP, and some of these basic macro measures.

Those are macroeconomic data. A very interesting time series that I saw recently is the fraction of scientific papers with a single author. This is interesting because you might imagine, if science becomes extremely bureaucratized and institutionalized and beset by these giant, sprawling, multi-institutional teams, you're going to see more and more authors on a typical scientific paper. And that, in fact, is the trend that we've observed for several decades. As of about two years ago, that long-term trend line has started to reverse. We're seeing more single-author papers. And again, if you imagine an environment that is hospitable to the high-potential iconoclast, that would seem like something we should see some amount of. So I just think it's interesting that that long-term trend line has reversed. It's obvious that AI and computation — there's immense potential there.

Just a couple of other things I'll briefly mention that I think give us maybe some hope. The second is there's a new generation of people who've been very successful in industry — we just celebrated some of them here — who are now starting to turn their focus to science and discovery and basic research and so forth. I think that's quite encouraging, and I think those resources, but also those new ideas, will be helpful.

Three, as we were just discussing, people are here in D.C. thinking about some of these possible reforms and ways we could pursue things more effectively.

And then lastly — and this is obviously double-edged — China. China is a geopolitical competitor and a significant force to be reckoned with. But it's also a country that is itself taking scientific progress really seriously. For example, as many people here will be familiar with, their biotech sector and their nascent pharma sector are doing extremely well. There are a lot of new drugs in early stages — phase one trials and so forth — coming out of China. And that's forcing us here to ask some hard questions about our mechanisms, the FDA processes, how expensive things are — Eroom's Law, as you mentioned, the exponentially increasing cost of clinical trials and so forth. And I think that's hopefully — and again, we're seeing some early signs — going to prod and focus our minds to change and improve things somewhat.

So I'd say across those four fronts, in a way that's consistent with the macroeconomic data, I think we should recognize the early wiggles of hope.

Kratsios: I like the optimism.

You're very well known for having started Arc. And I think one of the core tenets of Science: A New Golden Age is this challenge to the community that we have to be open-minded about new institutions of discovery. And you did that long before the White House was talking about it. How did you get to the point where you said, look, I've got to do my own thing? What was it that motivated you to do that? What was it about the way the existing institutions were operating that made you think, I have to do this?

Collison: Well, there are a couple of things. Every villain has their origin story. And a big part of the Arc origin story was a tweet that I saw from the NIH many years ago. They'd launched a new website or something for how to prepare for submitting your grant application. I don't remember exactly — I'd have to go back and look — but as I recall, it described an end-to-end, two-and-a-half-year cycle. And the idea that you would spend a majority of an undergraduate college education just going through the review process for a single grant seemed so excessive to me, and so oppressive and so burdensome, that I thought, geez, someone really has to do something here. So that was part of it.

I was just feeling sorry for the scientists involved. My wife is herself also a scientist. You mentioned at the beginning that I've had some kind of tangential involvement in science; she's directly, squarely involved in it. I got to witness many of the challenges vicariously through her. She worked on CRISPR, for example — specifically, she came up with the first RNA CRISPR that worked in human cells. And it's interesting: if you go and trace back the lineages, the vast majority of CRISPR work was not funded by the NIH, because it sounded too weird, right? There was a bunch of DARPA funding and some other agencies, but it was not principally the NIH. Anyway, so there was Silvana's experience.

And then a bit of the sense that it's easy to toss pebbles and to expound from the sidelines — a lot of people say all sorts of things, right? So I felt there's a question of, well, what happens when you apply some of these ideas in practice?

And if you look back at U.S. science over the last 150 years, so much of what we have enthusiastically adopted, which has become institutionalized, was first pioneered at a specific institution. We just saw the video about the Rad Lab at MIT. There was obviously the early ARPA community. There's the Broad Institute at MIT and some of the early genomics work. Even organizations like MIT and Caltech — these were quite pioneering and novel institutions when they were founded, not that long ago. Relative to Harvard, MIT looked pretty weird. So if it is in fact possible to do things differently or better, we thought we had to try it in practice. And so that was the hope with Arc.

And just the last thing I'll say on it: the specific goal of Arc is to try to cure a complex human disease. You can divide human diseases into three categories. Infectious diseases — we've done pretty well, as a civilization, at curing those. There are monogenic diseases, where a single mutation causes the problem. And then there are complex diseases, where it's a combination of environmental or lifestyle risk factors with multiple genetic risk factors: most cardiovascular disease, most neurodegenerative disease, most autoimmune disease, most cancers. We've basically never cured a single complex disease, and incredibly brilliant people have worked incredibly hard. So the hope with Arc — we haven't done it, of course — is to find a cure for a complex disease. We're working on Alzheimer's, and we'll see what happens.

Kratsios: The last thing I want to cover for a couple of minutes, because I know we need to wrap: you have been a pioneer of these Fast Grants. I think many of you heard the intro, but you very famously stood up and said, look, we've got to move faster in getting scientists money during COVID. This is literally life and death for humanity; we have to move the ball forward. And I think you opened the aperture for people to think a little more seriously about that.

Intermixed with that is a concept that you have advocated for a lot and thought about: broadly, the concept of metascience. This has come up a lot, and I found it very fascinating. One of the core tenets of the Kyoto vision document that we got 17 countries to sign up to last week in Japan was the idea that we would, as countries, endorse the concept of metascience. And to me, out of all the things we asked them to do — including using the word "superintelligence" — I thought it was the easiest slam dunk. Of course: we're stewards of taxpayer money. We should of course be analyzing how our grants are given, evaluating them, and giving more money to the types of processes that are working and less to the ones that aren't.

But interestingly, the European Union: not interested. They said, "Metascience — we're not really sure how the community is going to accept that. We can't sign on." Obviously we didn't give them a lot of time to think about it, but that was their initial gut reaction.

You've observed metascience evolve in conversations about science policy. How do you think about that? Why isn't it just so obviously what people should be doing?

Collison: Well, as someone who grew up in Europe: European institutions of late have rarely been accused of an excess of innovative thinking.

But look, on the metascience stuff, I think it's easy to forget how many different viewpoints have been held by scientists themselves over the last century. It seems so self-evidently obvious to us that the federal government should be a major funder of science. But scientists themselves were very split on this question for a long time, because they thought it might have pernicious side effects — politicizing science, or forcing people towards a kind of consensus, and so forth. And we chose to do it. I think, on balance, it was the right thing to do. Maybe some of those concerns were valid, but there are probably compensatory things we can do. But if you just take the scientists themselves seriously, there's been a very robust debate for more than a century on some of these questions.

If you go back to, say — the NIH really grew to something on track to its current scale under James Shannon in the '60s. And if you read what James Shannon thought the NIH should be and how he thought it should work, and you compare that to today, what the NIH is today is extremely different. Again, that's not to say that what he wanted was right or that what we have today is right. But there's an obvious question of, well, who was right, and which parts of this were right?

So I think there's a very significant metascience debate embedded within science itself. And to fail to reckon with that, to fail to take it seriously, or to fail to ask what we should learn from the complex adaptive system that science itself is, would be a grievous error. So I consider myself a very eager, enthusiastic supporter of metascience, and I know that many countries did embrace this concept in Kyoto, and over time I very much hope that more do.

Kratsios: Amazing. I know we've run over, but we could go on forever. Thank you so much. Really appreciate it.

Collison: Thank you.

Kratsios: Thank you.