Well, hi everyone. Welcome to this conversation. I'm really thrilled to have Patrick Collison join us. He's founder and CEO of Stripe. And this is a company that I got to know during COVID actually. You might say how did I find out because it wasn't that we were using Stripe as a service but we were introduced and we spent some time talking about biotechnology. He has a keen interest in getting involved in supporting and advancing science in the biomedical fields and we exchanged ideas. He really made me feel old because I actually had to soundwise in responding to his questions and I'm really looking forward to having this discussion because he really is quite an unusual talent both in entrepreneurship but also caring a lot about the scientific enterprise and he and his foundation have done a lot to support this and I really want to get into that. So, welcome Patrick.
Thank you for having me. It's great to be here with a legend of biotech.
Well, legend in my own mind anyway.
So, let's dive in. I want the audience to hear from you a little bit about how you got interested in this area, in this topic. Obviously, it has impact, but lots of things have impact. And then, you know, segue into the Arc Institute, what it does, how it functions, and how it's become a platform over time, and how it's changed.
Sure. Well, in really short terms, my brother and I we started this technology company called Stripe 15 years ago. And Stripe builds what we call economic infrastructure for the internet. And so programmable tools that help businesses move money globally, actually, you know, turn a product into a business and to scale to hundreds of countries. And Stripe started out serving entrepreneurs and developers. So there are lots of big enterprise platforms but we thought that there are lots of interesting things happening at the edges. They sometimes want to do things in some novel way that the status quo was not necessarily adequate for them. And you know, they care about different things like they really care about the barriers to entry and complexity and how easy something is to set up and so forth. So anyway, started in 2010, businesses like DoorDash, Instacart, Shopify, you know, these giants that or at least well today giants that at the time were these nascent little efforts were our first customers and now Stripe serves several million customers. But that initial bias in Stripe towards entrepreneurs and developers is, you know, very much at the heart of the company. And we think a lot about what determines whether somebody pursues an idea, which idea they pursue, what makes an idea succeed or not, like that essence of creation. When I was in college, I thought I wanted to be a scientist. And I studied math and physics and college can be a great experience. In my particular case, I learned that I wasn't good enough to be a world-leading mathematician or a physicist. So that was very valuable and I decided to pursue software and entrepreneurship instead. But I was still very interested in science. And I actually told my girlfriend of the time, now wife, that if Stripe really worked, I'd like to figure out some way to start a new kind of biology or science research organization. And so fast forward in 2021, so 11 years into Stripe, we decided my wife and I to start an organization called the Arc Institute. And the Arc Institute, we can go into any detail you want, but the very short summary is we fund scientists on renewable eight-year terms. So rather than being on the grant treadmill and having to pursue something that you can make a high confidence case for, instead to be able to pursue something somewhat more exploratory. And then secondly at ARC we build infrastructure platforms but you know half the scientists work on these core platforms for things like computation or obviously AI but you know cellular models or better imaging technology like things kind of beyond the scale of what any individual academic lab can easily establish for itself. So those are kind of the two premises, these platforms and these eight-year grants or excuse me programs that enable curiosity-oriented research. And so I'd say across both Stripe and ARC, the central essence is this question of the dynamics of creation and new ideas and how people make them successful and in both cases we want to build platforms that maximize the probabilities there.
Great, great introduction. Maybe give us just a little better sense because the audience may not know as much about ARC as it exists today. How many grantees are there? And roughly what's the size of the ecosystem that's growing around it?
Sure. So ARC is about 300 people today. There are 12 core labs. And ARC operates in partnership with UCSF, Stanford and Berkeley. And so all of the core investigators at ARC also have an appointment at one of those three partner universities. And we think that's very important because to be a vibrant research organization you want to have grad students. You want to be embedded inside the research ecosystem of these other places. And actually around this time last year, we released EVO 2 which was I guess was at the time and still is the largest biology ML model ever trained. It's a DNA language model. And we trained that in partnership with Nvidia. And you know, I was pretty excited to see that because I think it's kind of a validation of this idea that there are some ideas that are valuable to pursue that are kind of hard to do again circumscribed to the narrow boundaries of what's possible on a single R01 or something like that. I think with everything that's happening today, there's some bigger things possible.
And is it kind of like some of the other research institutes, co-located? Is there a lot that people do together or are they all individuals?
It's very much, you know, everyone's very excited these days about remote work and distributed collaboration and so forth. In the ARC context we really believe in having people together and so everyone's in a single building. It's in Palo Alto next to Stanford. And the other thing I guess I should just mention in giving this kind of high level description and I'm curious to get your thoughts on this is one of the motivations for ARC is you can maybe coarsely divide disease obviously into infectious and monogenic and complex and whatever there's lots of different ontologies but we've obviously as a species as a civilization we've made like pretty great progress against the infectious diseases or at least the most problematic of them. We at least have strategies for dealing with the monogenic diseases and screening, you know, whatever. For complex diseases, you can debate a little bit maybe the definition of a cure, but you can at least make the case that we've never cured a complex disease. And obviously the big killers today are neurodegenerative diseases and still CVD and cancer and so forth. You know, all these complex diseases, autoimmune, whatever. And I guess I'm curious to get your take on this like if the say last maybe you can characterize the last 70 years as being one of amazing progress in infectious diseases and strong progress if we're being generous in ameliorating some of these complex conditions but nothing that we can readily say or we can clearly say is a cure flat out. What do you think the next 50 years, well, do you agree that that's a reasonable way to characterize the landscape first off and then second in as much as it is, what do you think the next 20 years look like?
Super. So, as a warning to the audience, unlike many of our other discussions, this is going to be a back and forth of questions. So, if you're wondering if Patrick is going off script asking me a question and worse yet, I'm actually answering it off script. That's what I should have said in the introduction. That's what we're doing. So look, I've been involved in this kind of innovation in biotech and health, human health for now 39 years. And one thing I could take away from it if I stopped doing it today would be it's remarkable how much we've done by knowing so little. And I'd say we now know a lot more than we did before. And we also know that we know even less of what there is to know than we thought in the beginning. So it's a little bit like a mountain the peak of which you can't see and the clouds as far as you can see it and once you go up the mountain you realize actually it's a much much bigger mountain and so you're nowhere in a way. It's remarkable how much practical things we've done despite that. That's human ingenuity and essentially we found ways to create micro understanding on little things that either at the biochemical level or at the cell level. And we've done all amazing things to it with lots of unintended consequences and side effects and etc. So you know for example we'd like to believe somebody came up with the lock and key analogy and since then we go around saying lock and key lock and key lock key. The reality is our keys open lots of locks and doors and many other keys can open the same doors and so it's a bit of a mess and I think that's reality. So despite that rather than waiting for complete understanding the way physicists would have probably preferred to do it people have just said you know what people are dying people need to be treated let's just do better and better and this whole enterprise with hundreds of billions of dollars of research has been built on that compromised way of advancing with relatively little knowledge. Now to your point, what is likely to change? Good news is I won't be doing this in 20 years. I'm not very much of a longevity aficionado and so I can make lots of predictions that I don't have to be accountable for. And the prediction I would make is that the biggest impact we'll have is in creating models of human biology, physiology, the complex systems that we are bigger and bigger ones. And I don't mean by that just kind of number of parameters but I mean bigger and bigger at scale and begin to use that to make more predictable and more precise interventions. I think that is clearly the promise that we have. So what began at the beginning 25 years ago with human genome sequencing that was supposed to cure all immediately is now beginning to be this first layer of the stack of things you need. But and I think this EVO 2 model is a great thing. We were with Jensen Huang yesterday last night when they had this affair with this event with Lilly and he talked about EVO 2 among the things that they felt very proud of. So I think that the AI-driven revolution in science is going to be interesting and to that end I'll just mention
You think it's real. There's lots of naysayers and skeptics, but
I think there are lots of skeptics and naysayers about the role AI will have in lots of other fields. Fortunately, I don't know anything about those fields. But if there are actually skeptics about what the role AI can have in the scientific enterprise, I don't think they have enough data to support the argument and we do. We at Flagship have worked in this area for about seven years. So long before there were AlphaFolds and lots of the more recent generative AI and so by trial and error we've kind of seen what a learning system can do and I'll tell you kind of at a high level because we haven't had a chance to talk about this my kind of conclusion and my premise that we put out last year is this notion of poly intelligence which is if you look out 50 years I think what's going to happen is that humans will coexist with two other forms of intelligence that we will not understand one is artificial intelligence and the other is nature's intelligence and all the forms of nature's intelligences whether it's a virus that can take down humanity or a tree that every 50 years blooms or I mean these things are things that we have nowhere near the sophistication complexity of engineer today but we will begin to create models for them. So humans will not need to understand to emulate using artificial intelligence machine intelligence and nature's intelligence and those two as they reveal each other will be the new challenge for humans. How do you conduct that process? That's what I think will have profound effect on better medicines, better cures, better vaccines, prophylactics. It really is going to be a very interesting time.
I don't want to totally usurp your interview and your stage here, but if I've asked so that's a very optimistic take on what's to come. If I was to then jump to the kind of countervailing pessimistic sentiment and mood, the latest figures are but something like 40% of new drug candidates are being developed in China and there are many prognosticators with apocalyptic tones talking about the biotech sector in the US and whether it is viable today whether it will remain viable. Yeah, you could also say that the rise of China is in fact a very good thing that will grow the sector as a whole. Just what are your views?
Well, I put out this letter this week in which I mentioned this as part of the good times and the bad times we're living in at the same time. I think competition from China is both a good thing and a threat. It's a good thing that it's coming at a time when we seem to be losing our way in relying on the scientific method and arguing based on facts instead of opinions. All of which this kind of science-based policymaking is how we got here in a positive sense of the word and there seems to be an erosion in that anywhere you look you'll find whether it's the same 40% could be applied to the proposed funding cuts at NIH whether it's cutting visas for people like you and for me to come to this country to join universities to join private enterprise all of that suggests that there's somehow an undoing of the reliance on the scientific method and its outputs as a way to create more prosperous, more secure society. In a funny way, the rise of the Chinese investment in the very same method and output is a good reminder that we imagine if we stopped doing it and nobody was doing it, in which case we'd kind of go, well, you know, we'll just take our time and we might change our mind later. That we don't have the luxury of and that's the alarming part. I think that the not so good part is that it is the case that it's not a level playing field. It hasn't been in many industries and it isn't in our industry. And I do think that there is something to be said about looking carefully at the precondition for collaboration to me is fair competition. In other words, I wouldn't collaborate with people that are then going to not compete fairly. So where there are opportunities to improve the competitive landscape, whether it's IP related, whether it's how clinical trials are done, whether it's access to data that is available but only to local companies, all that stuff I think should be factored in. But overall, I think we can't as a country but welcome competition because this country made it based on its ability to compete. But I think we need to ensure that we're not kind of changing some of our principles at the same time as we're seeing the rise of this thing. So let's come back to what you've been doing this year. It really is fascinating that your day job and this other day job I suppose because I'm sure you need to get some sleep is so impactful. You know, I'm thinking here this institute that you're building, how have you thought about it in the longer term if you have obviously you have an institute where the sun sets you might think about an institute where the sun rises by that I don't mean in China I mean in east coast so you could imagine this as a model for a way to think about scientific enterprise but maybe say what are your early observations of the impact of the eight-year cycles and these platforms and how could that be institutionally grown?
Well, a couple of thoughts. So even before the sun rises in the east coast, it rises in Ireland. So maybe that's the logical place for expansion. But yeah, ARC is now just over four years old. And mostly I feel it's premature to be extracting any major definitive lessons from that which has happened so far. We're definitely cautiously encouraged and something that's quite striking is that four years ago or say five when we were just starting to think about this it really felt like the shape of the research enterprise in the US, the formula for it at least at the basic science level but maybe also at the translational level was pretty well established and it didn't feel like there was that much de novo structural thinking about well how should we just change the overall edifice here and both in good ways and in bad ways it feels to me that there's a lot more beginner's mind today and a lot more openness to new ways of doing things and that's partly because of AI it's partly because of disruptions and some of the issues that have faced universities it's because the rise of China we're just discussing but there's lots of things happening but in general the aperture feels like people are more open-minded. To your point about simulation, from a kind of scientific goal standpoint we're very focused on two things. So one is we, to this complex disease point, want to test our ability to marshal our different capabilities against complex diseases generally and Alzheimer's in particular Alzheimer's both being a huge disease burden in its own right but also the archetypal complex disease. And again, I think it's premature to render any verdict, but I'd say four years in, I'm cautiously optimistic about the progress being made there, though, obviously a lot of hopes have been dashed on the rocks of Alzheimer. So, who knows? But then the other thing to your AI point is we're working on a virtual cell. And again, two years ago when we started to work on this, we weren't at least aware of anybody else pursuing this. Now, obviously, there were a couple of efforts, which by the way is great to see. Like, it'll just first and foremost be an amazing day for humanity if and when somebody really gets this to work, although it's probably not a single binary thing. But our intuition is very much the same as yours where, you know, if we ask ourselves, well, Nixon declared the war on cancer in the early '70s. You know, we're here in 2026 and cancer is not cured. You know, why not? Obviously, you can answer that question at lots of different abstraction levels. But one way we think about it as well, a typical biology experiment costs, if you round to orders of magnitude, six months and $100,000 just like a typical experiment. And so if you think about it serially how many biology experiments we've gotten to do over the 55 years or whatever since the war on cancer was declared it's on the order of a hundred sequential experiments and that's obviously not a lot and when you compare that to the computational world where you can run maybe a million experiments in a couple of hours or something. It's just such a different regime and so obviously it's an open question whether we can build something with sufficient predictive accuracy and sufficient ability to generalize and so forth and we'll see but we want to build a perturbational model of gene expression at the level of individual cells on the theory that that'll provide sufficient insight for guiding hypothesis that then you can make your actual wet lab work an awful lot more efficient. Hopefully that works. Whether or not it does, I think you're exactly right that sort of figuring out these computational accelerants for some of the most binding bottlenecks is just going to be a very big deal. And I guess speaking of bottlenecks, something I also was kind of curious to hear your thoughts on is there's a really interesting book by actually I'm curious if anybody here has read it by James Lefanu, The Rise and Fall of Modern Medicine. Okay, I'm seeing at least one head nod. And he kind of characterizes mid-century medical and biomedical research as one of in good and bad ways very freewheeling experimentation. And it's good in that lots of amazing discoveries ensued and then it's bad in that lots of people were also injured perhaps with experiments that were not very well thought through. And then we kind of transitioned to a very rational era and it works but it's slow and expensive and so forth and in that rational era one of the things that is slowest and most expensive is clinical trials. And so Flagship is in some way a machine for generating approved drugs by proceeding through all the different steps that go into that and I'm curious how much you think about or care about clinical trials and I guess it's easy to think about the really scientifically compelling stuff about how we can do better target ID or how we can do better validation or whatever but then there's the maybe less glamorous stuff about okay how we do trial recruitment more efficiently or whether we can merge stages or I don't know so how much does the less glamorous more bureaucratic stuff matter and how much do you care about trials in particular
