What is social infrastructure?

I use social infrastructure to refer to a process or institution that connects people in ways they otherwise wouldn't be connected. It's jargon that hopefully makes intuitive sense, but it's important enough of an idea to me that I think it's worth giving a much deeper account of what I mean by it, and what I think of the concept.

This post by Duncan Sabien might explain what's going on here if you're wondering.

TL;DR

• Finding people, at scale, with specific personal characteristics is really difficult, because quantifiable measures (like test scores) are very imperfect proxies.
• In particular, I'm talking about what I call "sparkliness:" roughly, the mix of intelligence, curiosity, and agency. Lots of sparkly people exist, and lots more people would become sparkly given an environment that nurtured it, but there's a surprising lack of infrastructure to connect them.
• Building social infrastructure is how I've seen this problem solved before on a small scale (at summer programs like FABRIC and other small communities).
• Now I want to build social infrastructure that scales, but enables the same kind of connection. (Probably a hard problem considering no one has done it)
• In particular, I want to build social infrastructure for sparkly people who feel adrift/unsure of where they're going, and who would benefit significantly from access to other sparkly people, in a culture that deliberately endorses it.

(See the last section of this post for info on how I want to actually do this.)

I. What is social infrastructure?

I started using this term to denote the institutions, processes, communities, events, etc. that connect our social graphs together. You can think of people as nodes in a network, and relationships as connections between those nodes; social infrastructure is something that systematically supports interactions that create and strengthen those connections.

It's a really wide-ranging concept. Lots of things act as social infrastructure: schools, summer programs, companies, neighborhoods, social media sites and their constituent communities. All of these things support new connections, and allow strengthening of existing ones. Public transit is social infrastructure, if you treat it as such, say by talking to the person next to you when you sit down.

Why is this useful, if it's so broad? Because most social infrastructure is not designed to be such. Most of the institutions that act as social infrastructure are designed for some other purpose, or have no deliberate purpose at all.

Social media is oriented to making money and keeping attention, not facilitating connections — it only does so insofar as it supports the money-making and attention-keeping goals.

Similarly, schools are designed for… some weird combination of moral, cultural, and economic incentives, and facilitating social interactions is maybe a tangential goal.

Modern neighborhoods are mostly a coincidence related to the fact that land is scarce, and cities/cohabitation are economically effective.

These are not designed to facilitate connections — never mind a specific kind of connection (like the kind that produces cool projects or new ideas).

But that doesn't mean social infrastructure that's designed to be such doesn't or can't exist. It's just uncommon. We have things like school clubs or professional societies, events like parties, dinners, etc.; sometimes these events are optimized to act as social infrastructure. But they rarely scale.

What if we build our own institutions, processes, communities, events, etc. — at scale — and deliberately structure them to act as effective social infrastructure?

II. The filtering problem

Humans are complex creatures. Most dimensions of ourselves can't be described numerically, only qualitatively. (Things like personality tests are a neat trick to group correlated traits, but they're severely limited and very difficult to use predictively.)

Hence, with limited data, filtering for people who have characteristic x is pretty hard when x relates to personality or interests or character.

This comes up in all sorts of settings, broadly grouped under the useful term-I-just-made-up of "qualitative filtering problems."

For example, when you're hiring people for your company and you want people that will fit your culture, that's a qualitative filtering problem.

When you are trying to build a new group of friends as a first-year in college, that's a qualitative filtering problem too, just probably more complex because the criteria for passing the filter are not obvious and are tightly linked to your weird personal preferences.

Filtering problems are hard to deal with, and intimately linked to the idea of social infrastructure.

Why? In theory, you could easily solve filtering problems by brute-forcing them given infinite time; for example, you could personally interact ideally (i.e. have a conversation that allows you to judge every relevant aspect of someone accurately) with every qualified job candidate for your company, and vibe-check them all. You could personally meet every member of your undergraduate class, have ideal interactions with them all, and make a friend group based on your top 10 interactions.

But no, the real world isn't anything like this. There isn't anywhere near enough time in the world to do this, never mind social norms that prohibit large-scale attempts to optimize things like friend-making.

So we lean on social infrastructure to solve these problems. To make friends at school, you join clubs, play sports, and maybe get lucky with people in your classes. To find job candidates you go to industry networking events, maybe you advertise on social media. Social infrastructure allows you to solve qualitative filter problems.

III. Ok, what are you filtering for?

Presumably if you want to really optimize your infrastructure it would have one particular trait that it optimized for.

I propose sparkliness, i.e. how you'd describe people who are some combination of smart, curious, interested in making the world better, interested in making themselves better, and in some sense agentic, i.e. actively trying to shape the world around them for the better.

Speaking genuinely about sparkliness is hard. Think of it like how the Supreme Court talks about obscenity: it's impossible to create systematic, objective, empirical definitions of it, but I know it when I see it. It's qualitative.

But it's meaningful — being in places that are optimized for sparkliness (I'll talk more about that later) is awesome. The concept isn't a cop-out for "people I like," either: agency, desire to build and improve the world, desire to learn, etc. are all objectively observable traits — they're just difficult to observe.

Sparkliness is in some sense an ideal. Most people aren't as sparkly as they could be. Especially when we have an education system that incentivizes following rules, not changing them; memorizing formulas, not inventing new ones; formulaic accomplishment, not creativity. When learning is a task, not a joy, you lose something fundamental. Schools — at least, most of them — kill sparkliness. So do lots of other, deeper-seated things: dead-end jobs, poverty, lack of opportunity, discrimination. But some of it can even just come down to bad luck: you haven't found the friends that you can sparkle with in your local environment.

The implication is that there are lots of people who could be sparkly but aren't yet, for any number of reasons. People who feel trapped in school despite wanting to learn, or who are interested in lots of subjects but don't know what to do with themselves. People who are really smart, but have been beaten down and don't have the sense of self-efficacy to advocate for themselves. People who don't have the space in their lives to build their own projects because they go to school in the day and work to support their family at night. People who feel like they don't have friends to share their interests with.

I think surrounding those people (proto-sparkly people?) with the resources they need to thrive and self-actualize would have an outsized impact on the world. Partly because agency rubs off — once you realize you can just do things, the whole world opens up. By putting people like this in a situation where they're exposed do all the cool stuff they can be doing, and people who can help them do it, you could change their whole life path.

And while it's really hard to try and fix those systemic issues — education, poverty, discrimination — that kill sparkliness, there are a lot of low-hanging options to help enable it flower in spite of those factors. We can build social infrastructure that targets these people, supports their growth, and reinvests the returns.

IV. A case study in effective infrastructure.

Ignoring the obvious pretense involved in making this claim, I was proto-sparkly (💀) until at least Sophomore year of high school.

In middle school I was smart and wanted to do stuff, but didn't know what stuff I wanted to do, so I ended up doing what a charitable person would call "exploring" and what a cynical person would call "flailing aimlessly". I explored a lot, learned to program, tried to understand political philosophy and ethics, got into student government, played sports, figured out how to produce music, read novels, etc. This continued into high school.

I remember thinking about similar ideas to these, sitting in the school library, lamenting the overwhelming lack of sparkliness in my life — and thinking about ways to try and change it but hardly acting on them. I didn't know where to go.

This was kind of judgmental, and cringe, but also completely understandable. Even if there were sparkly people around me — there were, but at the time that seemed like an if to me — I didn't know how to find them, other than just by brute-forcing interactions, which was exhausting and difficult and unlikely to succeed.

I tried building things — my philosophy association was an attempt to at least partially address the qualitative filter problem, and it was at least somewhat successful.

But the thing that really helped me understand what I was looking for — and start self-actualizing — was discovering LessWrong, and then finding ESPR/PAIR, and applying, and getting accepted to PAIR, and attending.

It was a long process, with a significant amount of luck and variance — but it worked out, for me at least! PAIR was extremely effective social infrastructure. I became close, very quickly, with other attendees. We built stuff together. We stayed up till 5am talking. We hung out in the common room and learned category theory together.

This was, in a large part, I suspect, because the people who run the camp put a lot of effort and experience into understanding and emulating the cultural factors that produce social infrastructure. The introductory session had a focus on group norms, things like "just leave conversations if they're not interesting to you" and "just ask for things you want or need, people might say no but that's fine." We were free to spend our time however we wished — i.e. it was fine to skip classes if we felt like other things were more valuable uses of our time. (We rarely did.)

It was also because the people there were all sparkly themselves — the camp had an effective selection mechanism. For PAIR, this was a multi-stage application process — they were able to select a small number of people from a large pool of interested applicants, applicants who were already self selecting based on interest in the camp's topics.

Using PAIR as a case study on effective social infrastructure, this is what I've come up with: effective social infrastructure has (1) effective selection mechanisms for deciding who to connect and (2) strong supportive cultural aspects that enable effective connection between the people who it does connect.

The problem is that PAIR is limited in scope. It only scales to the capacity of the people who run the camps; they run ~4 camps/year; at ~30 attendees per camp, and probably ~2000 applicants each year (maybe more, maybe less, I'm giving a well-educated guess), meaning they're serving less than 10% of the people that are interested in the camps. That's without including the number of returning attendees.

Also, they're limited by publicity, and applications. I'm guessing that nowhere near the amount of people who would enjoy PAIR/associated camps know about them, and there's also the fact that they might not not have time/mental space to complete the application to the best of their ability.

So for the people that get in it's awesome. But there are lots of bottlenecks, leaving the other >90% of people who didn't get lucky or didn't quite make the cut to fend for themselves.

I don't want to just build for the people who get in. I want to build for the people who don't even know that these things exist.

V. Selection mechanisms

There are four kinds of selection mechanisms that social infrastructure has access to, as far as I can tell:

1. Random or random-ish selection: The people that the infrastructure connects are random or random-ish.things like neighborhoods, randomly assigned public schools, and other geographically-determined social infrastructure fall into this category.
2. Algorithmic selection: Mostly this has to do with online infrastructure like social media, where infrastructure is developed as a result of algorithmic factors. Twitter subcultures, for example, develop based on correlated recommendations. Subreddits acquire members in a large part due to similar factors.
3. Self-selection: When groups or processes are open to anyone, but have a specific focus that they serve — say, an auto worker's union or a club soccer team — members self-select into the group from some larger pool of people exposed to the opportunity to join.
4. Manual selection: AKA admissions. You hand-review applications, and allow/bar entry to some portion of candidates.

A lot of social infrastructure is a mix of these things — e.g. PAIR was a mix of self-selection (people who are interested enough to apply) and manual selection (people who the people who run the camp think would benefit/contribute to the camp). School clubs are a mix of random selection (who ends up at the school) and self-selection (who's interested in the club). (These are of course useful heuristic categories that don't account for every case.)

At any sort of scale, self-selecting processes and/or algorithmically-selected processes (maybe with some additional manual selection on top, e.g. community moderation) are necessary, due to labor and time constraints.

There are a lot of obstacles that get in the way of efficient self-selection; some of the important ones include (1) people knowing the community exists — they can't select themselves into it without knowing it exists — and (2) having accurate information abut the community available, such that the people you want to self-select for the community actually select into it and aren't turned off by aesthetic factors or stigma from people who don't know about it.

Algorithmic selection, on the other hand, is really difficult to do for yourself effectively, for various reasons (such as that existing social media optimizes for attention, not intellectual interest, or that you need to know what to look for/who to follow in order to effectively train the algorithm to give you what you want, which is difficult unless you already have solved the problem at least partially).

But this is what we're working with, if we want to build for everyone. We can try and maximize publicity and maximize accurate information to ride on self-selection; we can use SEO tactics and build online presences to ride on algorithmic selection. At least being aware of the selection mechanisms, giving them a name, allows us to be attentive to them.

VI. So what are we building?

That's the hard part. Here's what I'm doing right now:

• Building semi-open communities. I'm helping build a discord group around AI and Cognition, specifically targeted at young(ish) people. It's in a weird state between public and private where links are on the open internet but only if you know where to look — we (well, not me, the person who founded it) sent word out through various means, including to last year's applicant pool for a different rationalist-adjacent summer program — but right now this is planting the seeds. (If this sounds interesting, reach out!!)
• Building a coherent network for FABRIC Alumni. This is gated by the admissions process (sorry) but it's useful in a lot of ways; I'll (a) find other people to build with (b) get experience with building effective infrastructure/culture and (c) hopefully help build a coherent, well-connected social graph that can eventually get connected into other existing social graphs to create a mega-graph that we can then plug into open-invite communities and… listen, it's all part of my master plan.

Here's the long-term plan:

1. Build an online community with sparkly people. Currently working on this.
2. Work with cool people from that community and start chapters of a federated club in their local areas. (sorta like NHS; an exiting proof-of-concept federation/community is Hack Club)
3. Leverage local connections, existing networks, etc. to promote/recruit for that club in various cities. (Part of the question is "what do we do with the people once we found them"; some ideas include competitions, essay contests, speaker events, participant-run lectures/workshops on interesting subjects, and facilitated 1-on-1s, networking events, group skill-up sessions, etc.)
4. Scale. Get charitable funding, scale up, maybe hire people to actually support the org.
5. Become "how things are." This is the end goal: make it feel really normal to have easy access to sparkly people and resources as a high school, middle school, college student. Just like how school sports, band, Model UN, etc. is a "thing people do in high school," the goal is for whatever this organization becomes to be a fact of life. This is how we know we have succeeded.

This is an extremely high-level view of what I think is the strongest path towards building large-scale, open social infrastructure that actually supports sparkliness. I'll get into details, and expected failure modes (e.g. the organization becoming infected by college apps optimization, or getting too big too fast, or not getting any money, or losing culture) into some post in the future.

Until then, thanks for reading! Reach out, I'd love to talk about this with you!