Josh Elliott, Chief Scientist at Renaissance Philanthropy joins us to discuss RenPhil’s mission to dramatically increase philanthropic funding for science and technology and encourage people to pursue highly ambitious projects. Renaissance is challenging the traditional models for funding R&D; their approach is based on creating specialized, time-bound "philanthropic funds" led by field experts. In this episode we talk about novel institutional structures like Coordinated Research Programs, helping talented scientists think outside the constraints of traditional research paradigms, and the art of problem selection.
Groups and projects mentioned
Center for Robust Decision Making on Climate and Energy Policy at The University of Chicago
Department of Energy’s Lab-Embedded Entrepreneurship Program (LEEP)
Adam Mastroianni’s “Many Ships” theory of science and research
Parnian: Hello and welcome to the Bottlenecks Institute podcast, where we meet with the leading entrepreneurs, researchers, and policy makers, and explore the barriers to progress and how to solve them. I'm your co-host Parnian Barekatain.
James: And I am your co-host, James Gray. Today we're joined by Josh Elliott. Josh is the Chief Scientist at Renaissance Philanthropy and a former DARPA program director. When I think about the projects that Josh has led and supported, they frequently fall into one of two camps: things that can minimize outsized risk to humanity and things with an outsized chance of providing immense benefits. He's worked on everything from geothermal energy, AI for science, and ways we can better anticipate and avoid the worst outcomes from climate change. Josh, welcome to the podcast.
Joshua: Thanks very much. Really excited to be here. Thank you.
James: To kick things off, we'd love it if you could tell us just a little bit about Renaissance and what you're working on there these days. I know you've got a wide remit.
Joshua: We just launched Renaissance one year ago, we just had our one year birthday. And we've already done some pretty amazing things at a pretty large scale. So the mission of Renaissance is quite simple: increase the amount of philanthropic capital that's going into science and technology by multiple orders of magnitude.
And we do that by really being obsessively impact focused and really looking at what are the highest leverage things that we can do in the world and that we can help to serve the world. Our main way of functioning is to build what we call “philanthropic funds.”
The idea here is kind of simple and kind of obvious, but also a little bit revolutionary. And it’s not something that we've completely invented; if you look back at the history of finance in the 20th century, for-profit finance heavily began to specialize in the 1950s when lots of new companies and startups were emerging. Traditional investors found that they didn't have the expertise to be able to evaluate these different opportunities, and so the finance world started to specialize into venture capital and hedge funds and all these different things.
Philanthropy never did that.We are still based in the same philanthropic model that was started by the Fords and the Rockefellers a hundred years ago. And we think that's inefficient and suboptimal for a number of different reasons. Not for all philanthropists, but certainly for a large and growing class of philanthropists who aren't interested in standing up massive foundations that inevitably create institutional bureaucracy and make it difficult to rotate and pivot and take advantage of opportunities.
What we do is try to build expert-led, time-bound, impact and mission-driven funds that are led by what we call “field leaders” or “field strategists.”
These are experts in their field, but also highly ambitious people that understand how to drive towards impact; how to translate, as we say, “science and technology” into scalable impact.
We describe it as sort of a three part triangle (I guess all triangles have three parts) that's necessary in order to create a fund. We need that amazing field strategist, we need that time-bound thesis that says, “with this much money and this much time I can accomplish X” so it has that detailed theory of change.
But really critically in our model, it also has what I like to call the “theory of scale.” What happens when this fund stops? Because this fund will stop, it is finite. And then what happens next? How does this then scale and change the world?
Whether that's through commercialization, through government adoption or other subsequent programs. And that often means impacting not just the science and technology levers, but also the other kinds of translational levers or translational bottlenecks: things that you have to do in the policy space, in the regulatory space, market shaping. Really holistically looking at a systemic perspective; how are we actually going to create the change we want to see in the world based on, fundamentally, the power of science and technology?
And then of course we also need an anchor donor that's committed to that thesis and wants to help and see it grow. And then once we have those three things we launch a fund. We are working on everything from:
revolutionizing astrophysics by bringing down the cost of space telescopes by three orders of magnitude
eliminating childhood lead poisoning around the world
addressing the catastrophic risks of climate change
exploring geologic hydrogen
the science of human consciousness
AI for education
AI for science
AI for government capacity
I could just keep naming “AI for X” if I wanted to; we're pretty much working on all of them because obviously that's a high leverage play in the future.
[00:04:31] Coordinated Research Programs, FROs, and Virtual Institutes
James: One of the things I've always appreciated about Renaissance in particular is that you’re looking at financial mechanisms and institutional structures as a design space and, to your earlier point, realizing that there’s probably a lot of room for different models that just haven't really been explored yet.
You have this framing of “Coordinated Research Programs” as a kind of meta-category above things like focused research organizations. One that was new to me was the concept of “Virtual Institutes.” How mature is that as a category? Is it relatively new, or is it something that has existed for a while?
Joshua: It's actually something that the team designed and created when they were still at Schmidt Futures. So, as you know, much of the team (including Tom, Kumar, Ronit and Parth) spun out of Schmidt Futures when we created Renaissance. I think I'm the only person from the founding team that did not come from Schmidt Futures, in fact. So that was a model that they developed there. And I believe the first one of those was the Learning Engineering Virtual Institute which Kumar designed and built around trying to accelerate the potential for AI in the context of education.
Joshua: So the Virtual Institute model is really somewhere on that continuum of coordinated research programs. In the way it's designed. I might suggest that it's really kind of a continuum of the level of coordination. So we have things that we call “pure philanthropic funds" that have a near-zero amount of coordination where our thesis is: as long as we can get money to these 10 people that can do this amazing work, there will be a huge change in the world. There's not really coordination that's needed.
All the way up to like a typical DARPA-style program where you have intense amounts of coordination. You have a whole bunch of teams that are doing state-of-the-art work, but then you also have systems integrators that work across them. You have integration teams. You have maybe an evaluation team that's creating use cases and using those to continuously evaluate and push the technologies forward.
Virtual institutes are somewhere within that continuum: a little less coordination than maybe a hardcore DARPA program would be, but certainly a lot more than a fund.
And then a focused research organization is kind of another dimension of that. Instead of saying “we need to fund these eight different institutions and then coordinate them in these particular ways in order to create change.” You're saying what we actually need to do is create a new institution and bring the best people in-house. It's kind of the infinite limit of coordination, right? We need them in the same lab, working together, advancing the same goal under one leadership vision in order to do it. I guess FROs are sort of at the far end of the coordination spectrum.
[00:07:23] AI-Informed Funding Decisions
Parnian: Do you think AI could be effective at predicting which people or ideas are worth funding, or helping to make funding decisions? What do you see as the vision for that?
Joshua: Obviously I think that AI is extremely valuable across any application of knowledge work or decision making, including in terms of identifying white spaces for progress and innovation, or evaluating what those best opportunities are.
I think the key is making sure to figure out how to interpret it and that the human users that are using it understand what the limitations of it are and are able to compensate for those limitations. So obviously when it comes to trying to evaluate “big if true,” high risk, high reward things that could have massive impact but might be low probability… Those are fundamentally things that are on the edges of the typical “reality cone" of most people's ideas of what's possible and what's possible in the future.
And so you always have to be a little worried that AI, as effectively the aggregation of human knowledge and opinions, is not causing you to do a reversion to the mean.
We used to have this saying at DARPA that the enemy of innovation was basically consensus, effectively. So a consensus design process was anathema to what we were doing. We want individual people that are really smart and ambitious and have big, crazy ideas. To be largely left unchecked to be able to build and design those ideas into big ambitious programs. Not completely - you help them like, “mold.”
Because what you want is the edges of the idea distribution, and that means you get some stuff at the bottom edge of the idea distribution. But the stuff at the top edge of the idea of distribution is so world changing that you don't care. Whereas if you do what most organizations do, which is designing by consensus, you push everything towards the mean and you end up with a bunch of programs with a very low risk of failure, but also a very low risk of actually doing anything huge.
I worry about that a little bit with AI. But there's no doubt that it's an incredibly valuable tool for understanding the background space of what somebody is proposing and really contextualizing it within the framework of what exists and what's out there.
James: The idea of consensus being anathema reminds me… I don't know if anyone's actually doing it, but for a while people were talking about “funding by variance” or “decision by variance”. You can take it two ways. Either look for the things where people are really excited about it, or you could look for projects that were really polarizing, where some people hated it and some people loved it and there’s something interesting about that polarization.
[00:10:19] Incorporating Uncertainty into Decision-making processes
James: Earlier in your career you started the Center for Robust Decision Making on Climate and Energy Policy at The University of Chicago. What are the main ways, either formally or informally, that funders typically incorporate uncertainty into their decision making processes? Are there formal statistical models that play a role in the weighting process, or are there proxies for uncertainty that are frequently used?
Joshua: Oh man, you're bringing me way back here. Way, way back. I was still a postdoc when we started that. For most of my experiences, I would say that the short answer is that they don't and even when they want to they often can't. Because trying to formulate actionable frameworks for being able to estimate and quantify uncertainty and communicate that uncertainty in a way that doesn't just disable the decision space… the default action for uncertainty is to actually just make the decision space impossible.
It's like, “oh, there's all this uncertainty. Now I can't decide because I don't know.” Generally what people want is certainty and they don't want to hear that uncertainty. They want to know what is the best estimate.
What I found, going way back to my UChicago days 15-17 years ago, was that the critical thing was really figuring out what it was that they really needed to know to make decisions. And rather than trying to give them some sort of black box that told them an answer….
For example, when it comes to economic and energy systems modeling around the climate and environment (which is what we were doing then), what everybody thought they wanted was a black box prediction engine that said like, “Boop: Given these things, here's the trajectory of emissions and here's the economic costs over the next 50 years,” or something like that. And that's a massively uncertain thing.
I don't care how good your models or your theory, or your economists, or your scientists are; predicting the future 40 years in advance is just massively uncertain. So if you give them uncertainty bounds on that, like if you actually try to do some sort of, even just parametric uncertainty, let alone like deeper uncertainties, it's gonna be like, “We think the answer is three, but it's somewhere between negative six and positive 15.” And it becomes useless, right?
But that's not what they actually really need to know. What they actually really need to know is: what is the difference in 30 years between if I did this versus if I did that? The uncertainty in that is actually a lot more compact because there are cancellations between those things. So in that respect, if you're saying, like, “this policy is significantly better in terms of the metrics that you care about relative to that policy” the uncertainty in that is relatively small because the difference between those two trajectories will differ a lot less as you're varying different parameters, etc. It still can be a lot of uncertainty, but it significantly reduces it. I found that to be a really, really informative lesson. Think first about what it is people really, really need to know in order to be able to make a really effective decision and then figure out how to communicate that.
In my recent work around catastrophic climate risk and trying to communicate that… we have long time scales; it's low probability, massively high-impact events. It's extremely deep uncertainty. We don't even know what we don't know yet around earth systems tipping points, whether that's AMOC collapse or various risks in the arctic or catastrophic sea level rise.
How are you supposed to give the kinds of insights and wisdom that policy makers and decision makers need to be able to design effective policy on that? And that's a case where you cannot ignore uncertainty, right? You absolutely cannot. So what do you do? It's not, like, “error bars” uncertainty, it’s scenario-based uncertainty. The only way I've ever found successfully to be able to really get those “aha moments” in people that you really need, that really creates that understanding and deep wisdom and knowledge about the topic that they need to then go off and be better at their jobs is to do these intensive events: wargames, scenario events, etc. And communicating that understanding and insight in that way. And those are really great because you can actually build “uncertainty” into that event structure in a way that doesn't just give this error bar that they look at and say, “I don't know,” but helps them understand “Okay, if it's this world, then this is different. If it's that world, then this is different.”
James: I love that approach, and I can imagine that those exercises are probably also very useful for revealing what the unknowns are for the actors themselves, right? It could be that you do this exercise and it turns out, oh, here are all these assumptions that weren't baked into our model at all that came up for the first time during this event. Or it becomes clear that there are people that need to be involved that you didn’t realize ahead of time.
Joshua: Oh yeah, absolutely.
[00:16:01] Career Paths for Scientists & Institutional Models
Parnian: There are many people focusing on identifying ideas and talent, but less around the surrounding ecosystem. And I think that building and designing the ecosystem is equally important for ensuring that ambitious people and gifted founders are able to flourish. What do you think is a bottleneck for designing that innovation ecosystem?
Joshua: The first thing that jumps out at me is institutional innovation, and I think there's real progress being made on this front. Most people, most scientists (particularly in academia), still think that their only pathway to doing science is to get that postdoc and then that second postdoc and then get an assistant professor job somewhere, right?
And if they know of any other pathway, it's like, go start a startup and get venture funding, right? And that's it. When I was an academic that was the only pathway I knew, and I think at the time those were pretty much the only ones that existed for the most part.
I got really lucky and got offered a job at DARPA and discovered, hey, wow, there is actually another pathway. It's these ARPA-like things. But now there are all these fantastic new models that are more emerging; from the stuff we're doing at Renaissance, to Convergent Research’s focused research organizations, to this new model that we're really trying to help people push and build, which we call the “BBN model.” It's based on one of the most famous performers from the early DARPA days, the one that built the ARPANET. It's this company called BBN that spun out of MIT. It was actually started by these three acoustics professors in MIT that kept getting asked to design big, beautiful auditoriums and they just couldn't hire postdocs fast enough to do it, so they were like, “we should actually just start a company to do this.”
Then they had to have computers to do these designs, and so they ended up randomly hiring J.C.R. Licklider, one of the most famous early computer scientists of the 20th century. He and a bunch of other people basically transformed it into this amazing organization that was partner-led (so no investors, no pressure to be profit maximizing). It was ultimately a bunch of scientists that wanted to do great science, but wanted to see that science had impact. So it was also combined with a lot of fantastic applied engineering skills and just had these really adaptable teams that could be applied to projects.
ARPAs and DARPAs, whether that's the ARPAs in the US or ARIA in the UK are the perfect funders for organizations like this because we want to do translational R&D, right? We want state-of-the-art basic science, but we also want applied engineering that can translate those things into TRL four or five prototypes that are ready to actually scale, commercialize, etc. And so this kind of class of organization is just incredibly valuable for that, for filling that translational R&D gap.
Unfortunately, it's become a lot less common in the US. BBN eventually grew too big and got bought by Raytheon, and it's now like a bureaucratic hellscape- evolving in the same way that most institutions do over time, obviously.
So one thing we're trying to do, across lots and lots of different spaces, is work with groups of people to help them start new BBNs. To say, “what is the amazing superpower that you have in terms of combining your state of the art knowledge and science with applied translational engineering to create a sort of contract services model that can really drive technology forward.”
And this is based on, I'm sure you guys are familiar with Eric Gilliam's blog FreakTakes, but he's the one that developed this. We were lucky enough to be able to hire him. He was previously developing all of these theories with Stuart Buck at his organization and he came and joined us to try and implement a lot of those theories that he had built with Stuart. We're just really, really excited about this thesis.
Going back to my more abstract point, there are so many different ways that you can have impact in the world and it doesn't have to be getting a job at DARPA where you can leverage massive funding opportunities. It can just be figuring out what is that amazing value niche that I can do, and then going out and doing it and scaling it slowly in this kind of model.
[00:20:28] Improving Program Design Over Time
James: I find a lot of the translational R&D stuff that y’all are focusing on to be really promising, and I think about some of the other program models in the same ilk - the DOE has their Lab-Embedded Entrepreneurship Program (LEEP) and I-Corps, you've got LabStart, you've got Activate, there are a handful of programs that are working on translational R&D and technology commercialization in some exciting ways. That said, as an overall part of the ecosystem it's still quite small. Is there a lot of shared learning between these programs around what actually works and what doesn't? Are we getting better at running these types of translational R&D programs for smart and talented people? Or is it just so infrequent that we're kind of rebuilding a lot of it from the ground up every time there's a new one of these programs that gets stood up?
Joshua: I hope we’re getting better. A huge part of what we want to continue to build at Renaissance – through the Playbooks that we've been developing that are like the pattern language of innovation, through a lot of the metascience that we want to do across the work that we are doing… we're very close partners with the folks at Convergent Research, for example, and really thinking about what are the sort of metascientific lessons that we can learn from the existing FROs that are starting to now mature towards their later stages and things like that.
I think in a lot of cases, at least for this newest generation of innovative institutions, I think it probably is a little early to draw any critical conclusions (none of the actual FROs of Convergent have actually finished yet), to say if they are successful or not.
But I absolutely think that we're learning. And I think that through the stuff that Convergent puts out in terms of playbooks, the playbooks we're trying to do, even the fantastic Convergent Gap Map that I know you guys are really excited about. I think that is another awesome example: this was Adam and them just realizing like, “Oh, you know what? One thing that's happened to us over the last four years is that we've just come across all of these crazy bottlenecks to science that people are identifying. What if we actually organized these in one place and made them accessible publicly to the community and allowed people to add to it and build on it?”
They just released the first version and it's already incredibly illuminating and useful in terms of helping to focus people on opportunities for actually addressing big, important, high impact problems.
So I definitely think that there are a lot of learnings and improvement. But I always tend to somewhat be… what's the right way to say it? I wouldn't consider myself an optimist when it comes to human's abilities to learn and grow and improve, but I think we're doing the best that we can. I'll say that.
James: I'm such a fan of the whole FRO movement, and I think oral histories are a very underrated form of knowledge capture; when done well, it can be a really great way of capturing some of the cultural and informal knowledge that accumulated along the way. I don't know if any of the FROs are doing something like this, but it'd be very cool to see a series of oral histories as the first wave of FROs are coming to maturity around what went right and what went wrong. Maybe I'll have to talk to Eric about that at some point.
Joshua: I was gonna say, you should talk to Eric about that. We also have a program that we're just scoping with an amazing former colleague of mine from DARPA about basically “AI-powered social science.” Revolutionizing social science with AI by enabling the power of qualitative social science, of interviewing, of storytelling. Enabling that to actually scale to the point where we can actually start to have deep and complex and causal understandings of systems. The kind of things that you get out of qualitative interviews, but that you can't actually scale because you can only do like eight of them or whatever in two years with an army of grad students. I fully agree.
[00:24:23] Bottlenecks holding back different fields
Parnian: Compute was the bottleneck for AI progress until roughly 2015, and more recently we had a huge surge in AI due to compute. Do you see a similar analogy in other fields where there's a specific bottleneck that’s holding back progress from accelerating?
Joshua: My sort of taxonomy for bottlenecks (which is extrapolated from this taxonomy for different sorts of impacts and ARPA programs that I created one time) is that in some areas there are specific bottlenecks that can help a field accelerate.
I would say that the biggest and most obvious example there in so many different fields is data, especially AI-ready data. There are probably a hundred different important fields and applications that if we could create the right data set, we could have an AlphaFold moment in that field. That's what's holding it back, right? The reason that AlphaFold happened was not because of some magic stuff that happened at DeepMind. I mean, there was some magic stuff that happened at DeepMind, but the biggest thing is that for three decades scientists had been collecting this incredible well-developed protein folding data bank. And that dataset is what enabled the explosion in our understanding of protein folding, or at least our ability to predict it. And there's probably a hundred other fields out there that with the right dataset we could have that similar kind of accelerating explosion via AI. So that's the biggest, most obvious example.
The other thing that I always like to point out is if you go back through DARPA programs, one of the most high impact things, technologies that you see, are things that I call “tools programs” (or platform programs or whatever).
It's one of the hardest things to actually communicate what the impact is in advance because you're not solving some specific problem or 10X-ing some specific domain. What you're doing is creating a tool that will allow lots of other people to 10X their own different domains, and you can't necessarily predict how people will use that tool in advance. My favorite example of this is CRISPR which struggled to even get funded in the basic research because people were like, “I don't know what this is. This is crazy. This is weird.”
Once it finally came out and people realized it was a tool, it wasn't a singular advance, it was a tool that could be used for thousands of other breakthroughs. It revolutionized so many different potential fields. I think that in many different cases there are specific bottlenecks and breakthroughs and data is just the easiest one.You can see it across so many different areas.
The other one that I think about and talk about a lot is about high-throughput and automated experimentation. We are getting to the point now in many fields (for one thing, to generate that data that we need to revolutionize those fields) where we need to think about how can we rethink the experimental protocol?
How can we take that lab, bench-scale… whether it's in material science, in chemicals, in bio (obviously). How can we rethink our experimental platforms in order to generate the kinds of high throughput experiments that are needed to advance a given field? And when you connect that high throughput experimentation with the power of data and AI, you can easily see how exponential advances are possible in so many different areas.
[00:28:15] Progress & Bottlenecks in Automated Labs
James: It's not a field that I track super closely, but for probably the better part of a decade I’ve heard about the idea of the self-driving lab or the automated lab. My rough sense was that none of them had really gotten so far along. Is there renewed energy in the field now because of everything that's happening with AI, or have there been any material advances that makes automated experimentation more likely to happen now than previously?
Joshua: I think definitely AI and the accelerating advances in robotics, absolutely. I ran a program on cloud labs for synbio for a long time when I was at DARPA. Because of that experience, I can't say I'm terribly optimistic about self-driving labs in the synbio space. That's not the lab's fault, that's biology's fault. Biology is this, like, infinite dimensional, insanely complex problem where one change in a million different experimental parameters can lead to fundamental changes in your results and make everything so insanely hard to reproduce.
So it's hard, is what I'm saying. It's hard. I'm sure we'll eventually figure it out, but it's hard for me to be optimistic based on my experiments in that space.
[00:29:24] Advances in Materials Science
Joshua: In other spaces, like in material science and stuff, the coolest stuff that I've seen is from people that are actually trying to like, you know, rethink what it is we really need to measure in an age of highly predictive models and data and AI. Most of the way we develop new superconductors is we try to predict what a new superconducting material might be. We create a large amount of that substance and then we test it for superconducting properties. But there are proxy measures that can be built into models that can improve those models’ predictions of the ultimate superconductivity. And if those proxy measures can be measured by generating a much more easily developed… for example, like a sprayed on laminar sort of multi-stage ceramic version of something, right?
Then you can think about creating these experimental configurations. Where you have a bunch of spray paint cans of different substances (not literal spray paint cans, but you know what I'm saying). And you're creating a hundred different configurations of different lattice structures, different ceramic structures in one plate, measuring them rapidly, throwing them away, doing it again.
And you're going through thousands or more of different material properties a day, and you're not measuring the superconductivity because you don't have the right substance or material, but you're able to measure some key parameter that goes into a model that then significantly improves your predictive ability for what? So then you're reducing the number of actual like, big chunks of superconducting ceramic that you have to make by three or four orders of magnitude. Rethinking it as a design space problem and rethinking it in the era of data and AI and thinking about how we optimize the targeting of the optimal places in the design space.
I think that's where you can start to really think. It's all about thinking about our systems that we have in place, thinking about them systemically and thinking about how we redesign them systemically in an age of AI, of robotics, of all of these different things that we have that we have our toolkits. Rather than saying like, “Okay, we have a lab bench here and we have a grad student sitting at it doing this thing. How do we try to mimic that with a robot?” In my opinion that is not the right approach to lab automation, not necessarily.
[00:31:59] Problem Selection, Cause Prioritization, and Expanding Ambition
James: One of the core premises behind Bottlenecks Institute is helping groups get better at problem selection and cost prioritization, and we think problem legibility is a big part of that. Do you think humanity is getting better over time at problem selection as a discipline? What are the institutions or people that you think are especially good at problem selection?
Joshua: I won't speak for humanity, but I will speak for the scientists that I work mostly with. I guess I'd have to say no, as a baseline, I don't think we're getting any better at problem selection. I think that most scientists, we push them through this pipeline of, you know, undergraduate and grad students and postdocs and we train them to be in the details, in the weeds, bottom-up thinkers. Like, here's the next advance I need to do. In order to get this paper, I need to show this. I need this data to be able to make this thing, to be able to write this new paper.
And in our existing institutional pipelines we don't ever tell people like, “okay, your job for the next three months is to forget everything anyone's ever taught you and to think from the top-down or the forward-back and say ‘what is the change that you wanna see in the world? What is the impact that you think that you can have in five years and 10 years and 20 years? What is the big, big problem you wanna solve? Or the big opportunity that you want to create?’
Okay, now let's work backward from that. What will it take? What do you need to build, what problems do you need to solve? What bench scale studies do you need to do to get there?
We never tell people that. And this is the greatest thing about DARPA and ARPA, right? When I first started at DARPA, it was basically like… there was no training, there was no anything. Day one, they just drop you in your office and they're like, okay, go and figure out how to change the world. Right? It's just learning by doing and you sort of figure out and fight and it's sort of like do-or-die kind of, you figure this out.
But that's the mindset you have to get to, exactly that mindset. You don't start with this little technology or that one. And so that's exactly why I originally designed the Brains program and built that with Ben Reinhardt at Speculative Technologies and now our version of it at RenPhil, which we call the Big If True Science Accelerator (BiTS).
It's all about giving awesome young scientists the space to say, okay, your job is no longer solving that problem or producing that data set, or doing that thing. Your job for the next three months is to think about what is the revolutionary thing that is going to change the world in five years and ten years. And now let's work backwards and figure out how to build the program that can actually do that. And it’s just my favorite thing in the world to do. I've probably done, like, three or four of these programs at this point. I've mentored 20 or so young scientists working on everything from education to synthetic fuels, to new ways of 3D printing materials to next generation cryptography, you name it.
And a whole bunch of things - by the way, that I know absolutely nothing about, just to be clear, right. But you tell them that, they start out and they're always like, here's my big idea: I want to revolutionize this thing and it's gonna be like this. And here's the thing we're gonna do and blah, blah, blah.
And I'm like, oh, okay, that's great but I absolutely don't give a shit about that. Park that for the next two months, we can come back and talk about it. What is the actual big thing? And you just see people's brains go like 🤯in real time. And it's really fun.
James: That reminds me, we spoke with Jason Crawford recently and a big part of what they're doing at Roots of Progress is building these industrial literacy syllabi for high schools. It would be great if this mindset shift you're describing was something that was somehow instilled or practiced at an earlier level so people could get used to that mode of thinking by default.
[00:36:46] The mindset shift needed to plan truly ambitious research projects
Parnian: As someone who went into academia for a while and just got completely disappointed, I also think one of the problems is that people are overly focused on their individual projects. They don’t work together on bigger problems that we need to solve, they’re just thinking about their own PhD projects so they can graduate. That's like one of the bottlenecks I've encountered personally.
Joshua: The original idea for Brains and BiTS came to me when ARPA-H was first standing up. Arati and Wade (who were at Actuate at the time) invited a bunch of us in to do these virtual mentorship sessions with incoming potential ARPA-H program managers.
Wade would always start out, and the first question he’d ask them is like, “Alright, what's your big idea?” And like, literally down to a person, they would say, “My big idea is that I wanna do exactly what I'm currently doing in my lab, only bigger!”
And he'd be like, okay, okay, that's great. But we'd work with them over the course of a few hours and we'd explain to them like, your job is not to figure out what you could do with a lot of money in one lab. Your job is to figure out what could you do if you had the eight best labs in the world all working together in a coordinated way on one problem? And again, you see their minds explode, they're like, ohhhh, okay. Generating those mindset shifts, that's when I discovered that it's actually not that hard. People want to think that way, they've just never been given the license to think that way. No one's ever told them you are allowed to think that way. And because in academia it's like, oh, here's the project; we need to get this $1 million NSF grant so that we can fund these graduate students so that we can write the next paper so that we can get the next grant so that we can write the next paper.
[00:38:11] Second-Order Impacts of Cuts to US Research Funding; Climate Science in the Era of Tipping Points; "Small Ships"
Parnian: You’re speaking right to my heart, this completely resonates. There are a lot of US government grants running out or like getting reduced - what do you think are the second order consequences of that for academia and for private funding? Do you think there is a potential for new types of organizations like Renaissance to step in?
Joshua: I think it's gonna be all of the above. Absolutely, yes, I hope there is going to be a lot of really interesting and vibrant energy and people developing new models and new types of institutions for change. I've been a little disheartened by the fact that - I don't know if it's possible to pull the numbers on these - I'm pretty sure in the last four months there have been more single-member consultancy LLCs started than at any time previously in human history because of all the people that have left government and they were like, ah, I guess I better start a consulting gig.
Like obviously that's fine as a first step. We should all have our consulting LLC or whatever, but I've also been working with some amazing climate scientists, who are in the process of getting - what's the technical word for it? Shitcanned? No, no. These are some of the top climate scientists in America, right? Like the leading climate scientists. And I was like, so what do you think you're gonna do next? And they're like, I don't know. We were thinking about maybe starting a podcast and no offense to you guys at all, but like I was like: you guys could probably do something more impactful than start a podcast. And I would like to help you figure that out. Like you're literally some of the country's preeminent climate scientists. Let's see if we can't work on something a little bit better than a podcast.You should still do the podcast! Just to be clear. Podcasts are rad, but like, let's also think about something else.
So I'm working with them on one of these BBN model organizations, right? I think this is incredibly scalable. I would love to see, as the US climate science sector basically just gets hollowed out from the inside, this incredible, burgeoning dynamic group of new institutions that are able to do things much more rapidly, able to do things much more cheaply, right? Able to pivot and react.
One of the things that emerged from this conversation is that one of my fundamental beliefs is that we are in a non-linear era of climate change. We are no longer in that nice, slow-and-steady era of climate change. We are in the era of tipping points and feedback loops and cascading and systemic risks within the human system. And the seven year IPCC cycle was not designed for this era of climate change. It's still very important; reaching that scientific consensus is what helps to move the slow ball of multilateral negotiations forward, and that's still very important.
But we also need to be able to have a climate science that's much more adaptable to the realities that we're currently facing at much faster time cycles. And this might be a way to do that, right? Rather than having our three institutions of climate science in the United States that each have a hundred of the best climate scientists employed there, what if we had 50 different institutions?
I don't know if you guys know this “we need more small ships.” Have you guys seen this blog Experimental History? Adam Mastroianni started this trend called “Science Houses.” His thesis was basically that we have too many “big ships.” And big ships are great, big ships are useful, but they turn really slowly. They're not the adaptable ones - we need more small ships too. And so this might be an opportunity. Granted, I don't love that we're literally, like, firing cannonballs at all our big ships, but it might be an opportunity for us to like build some of those small ships out of the wreckage of the boards and… yeah, that metaphor probably doesn't need to be taken any further than that, so I'll just leave it there.
James: I think yes, by hook or by crook, this trend is happening and hopefully some interesting projects will spin out of it. Largely in part to the things that you're doing at Renaissance. That's all the time we've got today, thanks so much for coming on. This was a really fun conversation.
[00:43:00] Getting involved & Supporting Renaissance Philanthropy
James: Any parting advice or suggestions for ways that people can get involved or support y’all in your work?
Joshua: Reach out to Renaissance. Reach out to me, to Kumar, to Tom. We have one of those super secretive email structures that makes it really impossible to figure out how to contact people where all of our emails are firstname@renphil.org. So we're really, really difficult to get ahold of because we use this super secretive code for our email addresses. Reach out, we'd love to hear from you. We'd love to brainstorm, whether you're a scientist who wants to do something big and amazing and has a big idea and doesn't know how to make it work, or you are a philanthropist who wants to have more translational impact in the world and wants to partner with us on creating some big change.
Or you're somebody who thinks you're doing one of these cool new, innovative institution models. So this is one of the greatest things as we've been going out and exploring the world, is finding people who are actually already doing this in the world but just didn't know that there was a word for it and a community for it and that there were models and templates and resources and organizing those folks together. So feel free to reach out and engage, we always love meeting new, exciting, innovative people that want to make change in the world. We need a lot of change and we need it quickly. We need a lot of progress on climate, on poverty, on disasters, on health. Let's make some progress.
