A Public Goods Wish List

Mar 1, 2024

One time I told my mom I was interested in public goods funding. She said, “Oh, cool – what kinds of public goods?”

“Goods that lots of people want, but where there’s no way to stop people from using them, so it’s hard to get anyone to pay for them.”

“Like what?”

“Like a park.”

“Don’t we already have parks?”

“Yeah, but not as many as we should. And there are other public goods.”

“Which ones?”

“You know – the public ones.”

Another time, I had a conversation with my boyfriend at the time that went like this.

“I want to solve coordination problems,” I said.

“Which coordination problems?” he asked.

“I want to solve them in the general case. Like with, you know, mechanisms.”

“Name a specific coordination problem.”

“Iterated prisoner’s dilemma.”

“No, like, name a real one in the world.”

“Well, there’s that scenario with the polluted fish pond… hang on, there’s another on the tip of my tongue…”

And then the ghost of Elinor Ostrom swooped down to start clubbing me over the head, but I waved my smoldering sage from the common-pool garden and said, “Begone! I can explain.”

Why Isn’t It Obvious Which Public Goods Are Missing?

One problem with figuring out which public goods are undersupplied (and oversupplied) is that our existing public-goods institutions aren’t a reliable signal. When it comes to government-funded goods, the only real feedback we get about which results are true improvements on an uncoordinated world is public sentiment after they’re funded. This sentiment is reflected to some degree in elections, but elections are laggy feedback loops and suffer from design flaws, like tyranny of the majority, that make them imperfect arbiters of popular will. There’s also a principal-agent problem, in that the actual work of undertaking projects is delegated to bureaucrats, who aren’t particularly incentivized to get those projects done under cost or ahead of time, so even a good that would have been worth its cost at market price can fail to be worth the cost that’s actually paid for it.

With a private good that isn’t yet supplied, you can sort of estimate how much people might want it – investors can speculate, or the supplier can offer preorders, or you can look at the demand for the good in a foreign market where it is supplied, or look at the demand for the closest substitute. But since there is no really effective, incentive-compatible system of supplying public goods yet, there isn’t much of a reference class to compare potential goods to, and there are no investors to speculate because there’s nothing to speculate on. The closest thing you can do is look at cases where the government or a philanthropist has supplied some public good and conduct a rough vibecheck of how much the good gets used and how much people like it. But you still don’t know how much they’d pay for it.

(Incidentally, I suspect this might relate to the reason dominant assurance contracts fail more often than not: there’s not a great theory of which public goods we should expect to be undersupplied.)

One pleasant development is that a handful of non-governmental public-goods funding programs have been spun up in the last decade. I’d group Gitcoin Grants, Optimism’s RetroPGF, Kickstarter, and Astral Codex Ten’s impact certificate grants ground in this category (I’m excluding anything that’s primarily for altruistic charity with no coordination element; those have existed forever). Looking at where these programs end up allocating money gives us some more signal about the demand for various public goods.

But I also suspect that current PGF programs are still occurring within a small enough network that individual reputational and altruistic incentives come into play, tainting the signal. The Ethereum PGF community, for instance, is not only small enough that people know each other and thus care about what the rest of the community thinks of them, but also selects for people with altruistic personalities. It’s not clear how much of the success of its fundraising is really from preferences being aggregated and externalities internalized, and how much is just the same vanilla alms that would have existed anyway. By analogy, people manage to raise large pools of cash on Kickstarter, where contributors receive a product for their money, but they also manage to do it on GoFundMe, where contributors are just donating out of the goodness of their hearts or to help out their communities, and so “we collectively raised a large pool of cash” doesn’t tell the world whether you’re a Kickstarter or a GoFundMe. Bizarrely, I worry that PGF people are too nice for current PGF schemes to really give good information about what works.

So until we can round up a test cohort of public goods enthusiasts who are nevertheless antisocial grouches, time to do the second-best thing: make like an effective altruist and speculate from first principles.

Some Categories I’d Bet On

Each of these could be a post in itself, and probably will be as I research them in more depth (see tweet).

A quick note: throughout this post, when I say public goods, I’m specifically not talking about the social safety net or about altruism. As a refresher, the definition of a public good is a good where 1. your use of it doesn’t affect (or barely affects) my ability to use it, i.e. it’s non-rivalrous, and 2. it’s infeasible to prevent people from using it, i.e. it’s non-excludable.

So if you and I and a hundred other people all raise money to send medical supplies to earthquake victims, or start a foundation to provide free daycare to families who can’t afford it – those are admirable endeavors, and they might even be highly cost-effective, but they aren’t public goods, because medical supplies and daycare are both excludable, and medical supplies are also rivalrous while daycare is somewhat rivalrous.

(I will play a little fast and loose with criterion #2, non-excludability, but I think there are good reasons for this, which I’ll talk about more in the concluding section.)

Basic research and R&D

The nice thing about basic science is that it’s probably the hardest thing to Goodhart: a well-defined target lends itself well to precise measurement, making it hard for the measurement to get gamed away from the target. For the same reasons that the hard sciences lend themselves well to the scientific method – that is, it’s obvious which experiments imply which conclusions, unlike in fields like psychology – they also lend themselves well to having faithful measures and not being especially messy. Also, “replicate this other finding” is very well-specified. For that reason, scientific research works nicely as a target for mechanisms like challenge prizes.

Insofar as a field is legible enough to have a tech tree, showing which discoveries are downstream of which others, you can even build abstractions on top of that tech tree. For instance, you could draw up some function for funding and rewarding research teams based on which other innovations their work would unlock. (There are efforts to do this in open-source software funding as well, by looking at which packages import which other packages, and then propagating the credit along that path. Or at least, I could have sworn I heard someone give a talk on this at DevConnect or ZuConnect 2023, but cannot for the life of me find it again – tell me if you know what I’m talking about!)

You’d think that science’s Goodhart-resistance would also make it easier for the government to fund, because it’s easier to vote on exactly what should be on the research agenda. In practice, though, people in the field seem to sense a dead zone somewhere between basic research and commercial applications, where applied research or very early-stage commercialization efforts can’t get funded. Here’s a slide from a Juan Benet talk, for instance:

I’m not sure I fully understand the case for the existence of the dead zone – isn’t there lots of applied research in academia? And preseed funding exists, right? If anything, I’d expect the dip to be at the basic research phase, rather than at the development phase. But regardless, it feels like inefficiency is inherent to a system where the people or companies who benefit from research have to go through a large, complicated third party, the government, to allocate funding to the appropriate researchers. It might also be that research is best thought of as a public good whose consumers are engineers, in which case it’s more appropriate for companies to coordinate within industries to fund research useful to all of them, rather than for the whole industry to try to lobby for a piece of the tax-dollar pie on behalf of those researchers.

I wonder if this explains why metascience is such a common specialization of progress-studies researchers. There’s definitely a correlation between wonk and nerd and science enjoyer, and scientific-technological innovation is a powerful lever of growth. But I personally don’t feel particularly strongly about growth or even just poking around with microscopes and telescopes for their own sake, and yet even I like scientific research as a funding target because it’s a uniquely tractable public good, making it a potentially useful test case for funding mechanisms in general.

I also suspect the system of patents and intellectual property that currently governs drug discovery is suboptimal, and there’s room for an improved system that doesn’t require as much monopoly, but I don’t know enough about patent economics yet to say. (Little enough, in fact, that it’s possible that I’m even wrong about the direction, and pharma companies aren’t making enough money, as Alex Tabarrok argues on Marginal Revolution.)

Art: music, writing, visual, film, theater…

One purpose of art is to influence culture, and to give people something to talk about. This introduces an incentive not to gate art. Often, it gets gated anyway, because somebody has to pay for its production somehow. But even then, you can only charge so much if you’re trying to become a cultural phenomenon; if too many people are priced out, then the art doesn’t really do its job.

In the past, artists were often sponsored by patrons: all you had to do to make a living was to turn one eccentric plutocrat into a superfan. The transition to institutional funding of the arts made the art more populist to some extent – fewer portraits of eccentric plutocrats, more Diego Rivera murals. In that sense, it reflected more people’s tastes better. But on the other hand, art that’s too populist starts to suck. For one thing, attaching art to a a big important institution’s reputation incentivizes bland, risk-averse art. And even aside from that, art that optimizes for appealing a little bit to as many people as possible tends toward mediocrity: the Marvel movies feel like a self-inflicted repugnant conclusion. I’d hypothesize that the best art is adored by some particular group of people, liked by many, unintelligible to some, and absolutely loathed by a few. There’s a missing middle between “thing most people think is an improvement on a blank canvas or two hours of silence” and “thing one rich person or powerful cabal loves.” It’d be cool if regular people could coordinate to fund art they actually like, as opposed to the art that bureaucrats vote for; like Patreon, but more incentive-compatible.

Also, art, like friendship, gets tainted by money. There’s an inherent transaction cost: the fact that there is a transaction, at a particular price, is information that messes with the signaling value. Art says simultaneously “here’s my take” and “drink Pepsi”, and sometimes it says “here’s my take” in a quiet voice and “drink Pepsi” in a loud voice. Obviously that’s bad. But sometimes it says “drink Pepsi” in a medium quiet voice, and “here’s my take” in a medium loud voice, and that starts to jam your perception, like when you’re trying to have a conversation while talk radio is on in the background. It’s not just annoying, like it would be to hear bad talk radio in an empty room; rather, it actively messes with the thing you’re trying to perceive. We’ve become resigned to everyone having to compromise and sell out a little, some retaining their dignity more than others, but no one truly being able to say they’re above an ad read. I don’t think this is just the natural order of things, though. There was a brief heyday around the 19th and 20th centuries when you could mass-distribute art but also charge for it, and that period did yield a lot of great art, which signifies a glimmer of hope.

Some people can afford to weave logoless baskets, either because they were born into wealth or because they have a high-paying job that doesn’t eat up all their time (or had one and saved up). Unfortunately, having the skills to get a high-paying job, or being born into wealth, are both orthogonal to being good at art. In a more efficient world, we’d allocate the resource of “spare time to create art” to the people who were best at it. Also, if only rich people make art, then art about things like material hardship or poverty or anything that rich people don’t encounter much will be undersupplied.

I’d tentatively also lump in-depth journalism into the art category, although it perhaps overlaps with research as well.


If you think about it, it’s kind of surprising how well open-source software works in the status quo, compared to what a naive economic model might predict. We’ve got a giant Jenga game stacked on top of the proverbial random person in Nebraska, and things are mostly fine! But possibly this implies that if the baseline is so high, then we could get incredibly good and performant OSS if we were able to allocate funds more extensively and efficiently toward it. (Note that at the time of writing, I’m only a few chapters into The Cathedral and the Bazaar, and haven’t gotten to Working in Public yet, so probably much more information on this lurks!)

Open-source software is a public good not only because anyone can run it, but also because it’s arbitrarily remixable and customizable. Letting anyone make a modified copy of some code is the incremental mutation process that lets code improve by evolution, which is usually more powerful than central planning.

But just the ability to improve code isn’t enough – you also need to provide an incentive to improve it. There’s a school of thought that says good software is born from passion, and anything worthwhile will eventually get written by a couple of hackers on nights and weekends who can’t help but scratch the itch of a juicy problem or a tool they need themselves. There’s some truth to this – tons of open-source libraries are written and maintained as passion projects. But notice I said libraries, not applications. Passion can fuel inspired fits of programming, but it’s very difficult for it to fuel software engineering, because improving and maintaining a product has both a programming dimension and a time dimension. The time is really difficult to get for free, because ultimately you need both dependability and full-time work, which are very hard to guarantee without paying anybody a salary. And in fact, a few open-source products do exist, but not nearly at the same rate as open-source libraries; there’s no good way currently for those kinds of products to capture anywhere near the value they create, so I suspect they’re way undersupplied. (More on this in There Are No Muggles.)

This is especially true for products that have complex, multifaceted functions (rather than just a single utility, like a password manager), or that entail networked platforms, or that interact with frequently-updated external components.

The other thing a networked application typically has that a library or a utility doesn’t is a remote database. One model for providing this as a public good is to collectively appoint and fund a group to administer a centralized database (although then you need to make sure there’s a good enough system of governance to make sure the administrators’ interests don’t diverge from the users’). Another model is to use a decentralized database (or in some cases, a blockchain). Either way, there’s an additional public good lurking here – on top of the source code, a network is a valuable good in itself. For instance, even when Twitter open-sourced a large chunk of its secret sauce, it wasn’t easy to build a competitor because part of the value of Twitter (or any social network) is that everyone and all their content and state is there.


Also in the category of “it’s actually kind of wild that this works at all” is review platforms. No one’s getting paid, either in money or in influence, to write Google or Yelp reviews or to rate their Uber or Lyft drivers or to fill out email surveys. The way those systems manage to work are a) using very opinionated design nudges (make it really easy to click a star rating and hard to find the X button; keep surfacing popups and sending emails until users give in) or b) relying on people’s innate passion for saying their opinions about stuff. But especially as platforms become more monopolized, it becomes dangerous to rely on a single centralized reputation system. Getting banned from a platform like Airbnb, Uber, or Twitter can impose non-negligible limitations on your everyday life, and there’s no due process. And as the Twitter takeover showed, there’s no guarantee against arbitrary bans. Or in a case like China’s social credit system, there’s a serious risk of political targeting. Also, corporate reputation systems aren’t interoperable, so each new startup has to build its own, which slows everything down and limits reputation systems to individual use cases.

Decarbonization, geoengineering, environmental cleanup

This is the kind of project whose natural timescale is similar to the electoral timescale – in other words, it’s relatively large and slow, just like the government – so it’s more plausible for central governments to fund than some of the others.

But for one thing, anything in the realm of climate change is itself a coordination problem among governments. It was hard enough to make one Paris Agreement, let alone an ongoing regime of climate cooperation.

And for another, the U.S. government seems to have been capable of engineering projects on this scale from the 1920s through the 1960s, but not since. I’m not sure about other governments; maybe they’re better at it, although I haven’t heard much evidence that they are. It might just be that gridlock and bureaucracy are powerful enough that public-goods engineering projects can’t be accomplished at anywhere near maximum efficiency by modern governments, and there’s lots of room for alternative public-goods funding systems to do better.

Rapid development of cities and towns

When I began writing this piece, I had paragraphs and paragraphs of notes on all the different ways cities were undersupplying public goods. But over the course of writing it, I came to believe that basically none of them were actually public-goods problems. Some were issues of public bads, like loud construction or hideous facades or toxic fumes, and backfiring efforts to prevent them, like overgrown zoning laws. Some were issues of rent-seeking, the sort Georgism tries to address. Some public goods were already funded by most municipal governments in developed countries: road maintenance, sidewalks, parks.

Because cities are so ripe with externalities – people are packed so densely that every little thing affects every little other thing, like subways enabling mass commuting enabling more businesses, or infectious disease proliferating, or gentrification breaking down the cultural cohesion of a neighborhood – there are various goods, like cafes, libraries, museums, and mass transit, that are strictly speaking private or club goods, but that generate positive externalities on top. But even in these cases, the positive externalities tend to either be correctly supplied in the first place through government subsidies, or undersupplied for non-funding-related reasons like zoning.

I will mention three potential exceptions, though.

One, startup cities: if you want to start a city quickly, you have to build infrastructure from scratch before you have taxpayers. The ability to bootstrap cities quickly would itself be a public good insofar as it would allow innovation in the space of municipal governance, the way that the existence of a free market is a public good because competition begets innovation begets knowledge.

Two, underdeveloped cities: ones that sprang up faster than infrastructure could keep pace. Empirically, these cities lack lots of the nice public infrastructure that older, more developed cities have. Maybe there’s a chicken-and-egg problem, where a city will be poor until it kickstarts its economy by funding public safety and transit, but it won’t have the money to fund those things if it’s poor. But this is just speculation and I’d have to read more about urban development to know for sure.

Three, dysfunctional cities: ones where the government is corrupt or incompetent. There’s probably a lot of overlap with underdeveloped cities here – a phase of being run by the mob is a common growing pain for municipal governments. But there are also some otherwise rich or developed cities where the government gets stuck in a bad state, mired in bureaucracy or corruption, and then nothing people want gets done. So someone else has to do it.

What patterns show up?

Intangibility and inexcludability

One property you might notice about this list is that most of the goods on it are intangible: either they’re literal information, like research and reputation, or they can be represented as data and distributed arbitrarily, like art and software. Many of these are public goods in a nebulous sense: they’re definitely non-rivalrous, because information can be reproduced so cheaply, but you can sort of try to gate them, like with paywalls, meaning they’re partially excludable. But trying to squish them into the club-good box doesn’t work so well either, for two reasons.

Reason one: excludability is hard and getting harder. Information wants to be free – there are few papers you can’t find on SciHub, or books you can’t find on Library Genesis, or movies you can’t find on the Pirate Bay. Data gets hacked, secrets get spilled. Protecting information requires preventing every single possible accessor from leaking it, whereas doing the leak just takes one person (as long as there exists a sufficiently uncensorable way to continue to propagate or host the information). And unlike pre-internet, the hosting and sharing of small amounts of media is too cheap to meter.

Reason two: in many cases exclusion is socially suboptimal, even if it’s possible, because so many quasi-excludable goods are also network goods or have positive externalities even on people who don’t use the good. For example, a social platform is a network good because it’s no fun if you’re the only one there; a TV show is a network good insofar as it’s more fun to talk about it with the whole country than just to watch it yourself; security is a positive externality of open-source software because more eyes on the code means more bug-spotting; a boost to the economy is a positive externality of a public transit system. (For this category of club goods with public-good-like externalities, I propose the term “clublic goods.”)

Interoperability and remixability

The other nice thing about intangible goods is that they generally take less overhead to produce and distribute, which makes it easier to use them to test funding mechanisms. Once you figure out how to fund your webcomics, then you can figure out how to fund your carbon dioxide removal plants.

Broadly, public infrastructure helps people build stuff, and when stuff gets built, that in turn benefits lots more people. On the level of technical protocols, this looks like decoupling different layers and allowing them to interoperate (for instance, the way your internet service provider is totally separate from the companies that build webapps). Interoperability lets producers specialize, and it also lets consumers choose from competitors on each layer, not just across the whole stack. With intangible goods, you can see this effect develop really quickly; when people talk about an “ecosystem,” often they’re talking about a few interoperable layers playing off each other.

When it comes to knowledge and culture, the equivalent of interoperability is remixability. One notable entry on my list of mysteries I have yet to unravel is how Musical.ly, the precursor to TikTok where users filmed lip-sync videos, didn’t get immediately blown out of the water by copyright law. Whatever they did, it must have worked out fine, because they didn’t die and instead ended up landing agreements with the big labels later on.

Why’s that important enough to make the list? Well, if you think about the successful startups of the 2010s, Musical.ly’s business model would probably have been the hardest to guess from scratch. A platform to stay at people’s houses instead of hotels, a way to store your files online, a way to chat with your friends using photos, sure, all these sound pretty believable a priori. But an app to… socialize by making silly amateur music video clips? My theory is that whatever legal maneuver they pulled off literally unearthed a whole system of oral tradition from a long stagnation. The song is more or less the original unit of memetics – that, and the myth – and I even think there’s a lot more latent cultural potential lurking behind the copyright morass we dragged into the extremely-cheap-reproduction internet era from the distributing-physical-records era.

Look, I love a good internet transport layer. I love a railroad system, a quadratically funded research accelerator, a proletariat-owned Starlink, whatever. Nevertheless, in practice, the benchmark that lives land-value-tax-free in my head is this: I’ll consider public goods funding halfway to being solved when we have a fully free and open-source Spotify equivalent. By this I mean: 1. the platform itself is open-source, 2. it has most well-known songs, and 3. you can obtain the rights to use, remix, or sample 90% of the songs on it in under five minutes for under five dollars upfront.

Let’s get that done, and then it will seem trivial to make the trains run on time.