Building the 21st Century Commons – San Francisco 2025

I would love to start off by asking you uh more about Renaissance philanthropy. I think that it's existed almost a year now. Yeah, we started in May of last year. Great. Uh would love a sense of what what how the goals of the organization have kind of evolved in that rough almost year and where you see yourself heading at the moment really.

Sure. Yeah. So the premise of Renaissance philanthropy is that in the same way that wealthy families supported the Italian Renaissance by backing Michelangelo and Da Vinci, uh today's philanthropists could be supporting a 21st century renaissance that is fueled at least in part by advances in science, technology, and innovation. If you look at the 2,000 families with half a billion dollars or more in wealth, their current level of philanthropy is 1.2%.

So if you assume that they're doing as well as an index fund, there is which does roughly 5%. Uh there is room for them to do much more than they are currently doing. Uh so one question we're interested in in is a why is that? And b what could be done about that? Well, one common challenge that you see is that people have made their money when they're still professionally active.

So they don't have a lot of time to think about their philanthropy. uh or they have no interest in supervising a staff that can help them source, evaluate, select and manage projects and programs. Um and an insight from behavioral science is if you want someone to do something, make it as easy as possible. Um and so one model that we've been exploring is to build the following three legs of a stool. Uh one is a thesis that is uh an uh a a goal uh that uh would get philanthropists excited.

Um and a description of the role that philanthropy can play in making that happen. Uh number one. Number two, um a uh someone who can provide the intellectual leadership for that fund. Uh and then number three, at least one anchor donor. Um and so in the same way that someone who is serving as an LP doesn't have to do due diligence on 200 seedstage startup opportunities, the philanthropist can say I am generally interested in this area and I have confidence that the person who is leading this fund is going to do a good job picking uh projects that are aligned with that goal.

And so if not for one of those funds, what kinds of options do philanthropists on the this the financial scale you're thinking about have? I could imagine like the historic path was go donate a building to an R1. Yeah. So, so they could say, "I would like to have my name on a building, or I would like to increase the endowment of the university that I attend, or um I'm Bill Gates, and I'm going to create an organization that has a staff of 2,000 that is devoted to turning my wealth into advances in in global health and global development." And so, the kinds of options that you've seen traction with at Renfill now, those look very different from building buildings there.

Thanks. So a example would be that uh we are uh running a program at the intersection of AI and fundamental mathematics uh that builds on a uh focused research organization that we helped raise money for uh called lean uh which is in the uh field of automated theorem proving. And so in this case the philanthropist had to agree to the budget and the thesis and we took we are taking responsibility for sourcing evaluating selecting and managing a portfolio of projects that is aimed at using advances in AI to advance fundamental mathematics. So this is works out for the philanthropist because they do not want to have a large staff. Yeah.

And I'm incredibly excited about this particular program. And I'd love to have you briefly mention what a focused research organization is. But first, I also want to say for those of you who have not come across lean, this is getting a lot of traction and a lot of interest. Partly because some mathematicians are very interested in it. But also for those of you who are in software, you should think of this as being able to do for proofs what open- source and git have been able to do for math.

You could take a pro, take a theorem, decompose it into a bunch of checkpoints, have people independently generate proofs for these various checkpoints and then combine them all back together into one unified thing where each individual lema that has been proven could be then used in other systems. I think this is and people are also using it for hardware and software security as well. So in addition to like proving including people on this stage uh you know it it is valuable in other contexts as well. Yeah. But back to focus research organizations.

Yeah. So um this is an idea that two early career scientists had uh Adam Marblestone and and Sam Rodriguez. Um they were both uh uh had been alumni of Ed Bdon's lab who's a professor at MIT. Ed sent me an email said I think you should talk to these folks. Uh I was working for Eric Schmidt at the time.

Um, and so Eric got behind this idea. Uh, so these are, uh, timebound research nonprofits with the CEO. So they have the unity of purpose that you would expect to see with a well-run deep tech startup, but because they're philanthropically funded, they have the ability to uh, work on projects that would be difficult to monetize. So, a a good example since we're talking about public goods is if you went to a VC and said, "Give me $50 million. I'm going to produce a highquality data set and I'm going to put the data set in the public domain."

That would be a very short meeting because the VC would might ask, "How am I going to get my money back if you're giving away the intellectual property?" So, we now have uh eight of of these that have been supported through Convergent Research. Uh I'm the chair of Convergent Research. Adam is the CEO. Um we have the first one that is in the UK and Arya which is the UK equivalent of DARPA is working with Convergent Research to fund some more in the UK.

And what I think is interesting is that the meure existence of the model is increasing the level of ambition of the research community because prior to this, very few researchers were asking themselves, "What would I do as the CEO of a $50 million research nonprofit?" Because that seemed like an unl likely outcome. It would be like if I said, "Evan, I'm going to spend the next three months figuring out what I would do if I won the lotto." you would question whether that was a good use of my time. Or maybe the question, what would you do if you could make the president make one phone call?

Yes, exactly. Which for those who have met Tom is a question that he was consistently asking everyone. And it's a great thought experiment, but it's also I think a it has an interesting mixing between how creative are you and how ambitious are your goals and also how well do you understand the executive branch and what the executive branch can do which is meaningfully different now than I think it was a couple of years ago. Yes. Uh so one I have a question for you which is the way in which I am supporting uh public goods is a very normie trad way of supporting public goods which is I'm talking to philanthropists and foundations but this community is coming up with uh you know novel and more decentralized ways of supporting public goods which I fully support but I'm interested in your take about how that's going number one and number two what experiments you think we should be trying So, I might invert those questions a little bit.

Uh, in terms of how it's going, I I've become slightly disconnected because the types of public goods I'm looking at are different. And so, maybe I'll go into some of the the things that I think would be exciting to try. uh for the some people who have uh joined funding the commons more recently. I think actually the very first or maybe the second public funding the commons Tom and I actually had a fireside chat um virtually years ago and I think that there was some conversation about the project hyperserts. I don't think I don't know if anyone from hyperserts is here.

Um but this was the notion of impact certificates to h provide something like equity in nonprofit or public good type efforts. And one of the things that I would be very excited about as a uh as someone who had like me designed doing done some mechanism design in the public good space is to have something where you have on the uh maybe in the near future a declaration that there will be something like an X-P prize that is large enough that it will be profitable for teams not just uh reputationally but also financially and have that prize pledge to reward not just the teams that participated or not just the team that won, but also anyone who supported them in a proportion that was agreed upon as fair a priority. And the reason to have this is that that prize then doesn't look like it's a reward for winning. It looks like it's a buyout of equity for those who have created something of value. And what this does is essentially creates or turns that any pre-agreement into something that looks like equity that you can sell to finance the venture and causally derisk the creation of this thing.

A lot of enterprise teams don't uh a lot of teams don't join enterprises because it they don't have access to the capital that it would take to under do this undertaking. But if you could imagine something like a YC demo day for basic science open research, something like this, you could imagine bringing in enough angel investors that they could provide liquidity to a bunch of scientists who would never consider raising in this way and try and fund their way through to potentially break through science that could be evaluated very subjectively with very little gaming at the end at the outset. So that would be something I'd be excited about. The types of public goods that I'm thinking about now are things that look a lot more like uh open protocols for compute governance of AI or uh rule of law in a world with AI as a public good. And so these things look meaningfully different.

And so the framework that I've started thinking about is what is the relevant quorum of humans that need to adopt this thing for it to be able to have sufficient network effects to take over as a public good in the way that I would say that uh TCP is arguably a public good. No one's uh anyone can use it and there are no restrictions on on it and the more people who use it in fact the more effective it becomes. And I guess there another uh fun call back is that the team that I ran that created funding the commons was called network goods and I would consider those to be public goods or commons where you have actually a network effect that increases the value as more people use it. It's not net negative or even netneutral. So I think there are many things like this particularly exciting things that have rather short timelines for impact in the AI space.

Um I think going into any one of those individually would be potentially too ambitious but as like quick indices having more robust formal verification maybe specificationdriven architectures so that people aren't saying you AI agent you trained with my value system and my utility function don't uh like do what I would do. I think having a system that's rather based on rules and laws and is more of constraints rather than optimizations to get what you want. I think that would be a real tools that do that and architectures that do that would be a really good public good. And then also having systems that are more focused on uh having a protocol where if everyone agreed, yeah, we don't want to do AI like training runs that are over this size or that have these properties, being able to make that a software update rather than a hardware update, especially not a hardware update that would be encumbered by IP. I think that would be really exciting and I've actually started talking to some people about road mapping a focused research organization on what it would take to make that hardware.

That's great. Um well one type of public good that I'm interested in is if you look at the areas where we have made progress um uh in applying either advancing the state art of AI or applying AI and machine learning to some particular problem. It's often the case that we have number one a well-defined problem uh number two a large highquality data set uh a benchmark for evaluating progress and and a leaderboard and this is the common task framework. Yeah. So a canonical example of that is that the reason that we made progress on protein folding and protein structure prediction is the problem was if you know amino acid sequence what is the protein structure?

uh structural biologists had been contributing to the protein datab bank since 1981. So there were 230,000 protein structures that were freely available. Um and uh there was a metric the critical assessment of protein structure prediction. So presumably we don't want to have to wait for 45 years to do this in other scientific domains. So, how could you think about these novel mechanisms uh for uh funding public goods that would accelerate the development of an open data set?

It's a very hard question. Do you have preliminary thoughts? Well, again, I'm sort of doing the normie thing of talking to philanthropists and foundations to to get them interested. I mean I I do think one possibility that you have uh in some areas of science is that you can dramatically lower the cost of generating the data. So you know arguably one of the most important parts of the human genome project was a program run by an NIH employee uh uh Jeffrey Schlloth and it was called the thousand genome.

Uh and since then we've gone whizzing past that goal and we're now rapidly approaching $100. So the big impact of the human genome project was we went from $100 million to $100. Uh so there are opportunities to dramatically lower the cost of generating the data using things like screen pooling and uh and multipplexing. Do you think that there are I guess taking a tiling tree approach you could make generating the data cheaper? You could try to pay to open the data.

You could uh try to make ways or find ways to require less data. [Music] Um it seems like probably most of the avenues are there. Uh to the extent that I imagine people come to funding the commons because they care about new mechanisms to create these sorts of things. Uh I would generally point anyone who has a clever idea about how to do this kind of data set benchmark creation to Tom you have a playbook a set of playbooks on your yes if you go to Renaissance philanthropy you'll see six playbooks uh that we've created and we're working on some more um and you know to your point about hyperserts and uh and impact certificates uh one of the things that drive me crazy is that we have widely utilized mechanisms for making financial commitments that are contingent on failure. Those are loan guarantees.

So that's an if then statement. But the if is if you go bankrupt the and the then is then the government will assume your debt obligations. So we have trillions of dollars in those. But we almost never make financial commitments that are contingent on success. Uh so saying you know if you do X then we will do Y.

Uh where we do they seem to be poorly adopted and generally reinvented as letters of intent or advanced marketing a commitments. Yeah. So ideally what we would have is a marketplace for outcomes that has three types of participants. The entity that's willing to pay uh the teams who believe that they can achieve that goal and the investors who are backing those teams. Yeah.

And I think we're almost out of time. any as as the founder of this community any uh any last words of wisdom? Um I think that I'd probably point to the fact that there are many there is a lot of space for new mechanisms and coordination. I love the quote that civilization does not move forward because people think better or think more deeply about more things. It moves forward because we've dramatically enhanced the number of things people can do without thinking about them.

and mechanisms are the way in which we do that. So, uh thank you for helping highlight the value of these new coordination mechanisms and helping populate them with uh and direct people toward them. And thank all of you for coming and thinking about this and how we can improve our civilizational infrastructure. Yeah, no pressure. [Music] [Applause]