Tldr: I think that if you want to have an outsized impact as a community builder, it's worth your time learning how to develop a product that's so good that imitation products are made or that's so good people tell their friends about it. To learn more about how to do this, you could skip to the last section of this post, check out Peter McIntyre's resources, watch 10 minutes of this video (starting where I linked), or read a book from this list of books.
Community building projects are like products
Community building can look like:
- Running a reading group at a university
- Running an event, e.g. EA global or a speaker event
- Running a global scholarship
- Writing a book
- Making a youtube video
To me it seems that most community building projects involve building some kind of "product," in the sense that the products have a particular target population (e.g. "world's smartest high-schoolers” or "members of the EA community" or "altruistic college students") and are more effective if they more effectively fill the needs/wants of the target audience.
Interestingly, it seems like community building projects can actually sometimes be two different products at the same time; for example, if the point of running a global scholarship is to find people who are talented enough to work for ARC, then you also need to make sure you are fulfilling the hiring needs of ARC (e.g. perhaps by talking to ARC, you learn that you should target theoretical CS PhD students). This is analogous to Facebook being a product that’s built for both the general population and for people who want to purchase advertisements directed at the general population. Though it seems that “broader” community building projects look less like two products generally (e.g. for The Precipice or translating EA content to other languages, it doesn’t seem like there is a clear ARC analogue).
Why product development models are useful
According to standard product development advice, you most effectively learn about the needs of your target audience via high-quality user-testing (though I think you shouldn’t take the advice at face value and instead figure out your own gears-level model). For example, this post talks about how to do high-quality user-testing when trying to write a persuasive article. High-quality user testing is important since it helps you figure out how to genuinely fill people’s needs, and the more that your product actually fills a need, the more that people will use it. If it is a really great product, they will tell their friends about it, causing it to spread like wildfire.
Two examples that come to mind of "things that spread like wildfire" include 1) the idea of "fellowships" for college groups, originally conceived by Harvard EA and 2) the creation of really nice office spaces, originally conceived/developed by the Lightcone Infrastructure Team. Though I believe only the latter product was created via typical product development processes.
Because the idea of fellowships was decent, albeit not the outcome of product development processes, there are now likely dozens of groups that run them. And from what I can tell, in large part because the Lightcone Infrastructure Team did a fantastic job of creating a longtermist office space in Berkeley, several EA groups were inspired to do the same for their groups (EA MIT, Columbia EA, Harvard EA, Stanford EA to name the ones I know about, though I am told there are close to 10 examples of office spaces being made inspired in some part by Lightcone's).
Another piece of product development wisdom is that since there are often key uncertainties that affect the way you develop your product, or your sense of its effectiveness, you should resolve key uncertainties ASAP. Because of this, you should expect that to make a product everyone loves, you will have to go through at least several rounds of iteration where you rethink major parts of your product. To resolve key uncertainties, you could develop your entire product and deploy it to resolve the uncertainty (slow feedback loop, not recommended), or you could develop some experiment/prototype that resolves the uncertainty faster and ask a handful of people to participate (fast feedback loop, recommended).
For example, you might not know whether CS majors will want to come to an event about X, and so you decide to go door to door in your dorm hall until you find a CS major, and you offer them $20 for them to verbalize their thoughts as they read the email you are about to send about event X. Additionally, just like with any other kind of experiment, you aim to imitate the conditions you care about as closely as possible, so you send them the email and have them check their inbox at their desk rather than just handing them your laptop.
Example EA infrastructure improvements inspired by product development models
I feel like almost every time I engage with community builders, I can see how their goals would be better achieved if they had a better understanding of product development (which is why I am writing this post!). I can also often see existing parts of EA infrastructure that feel like they could be improved, and I give two examples below.
Caveat: I think the organizers of both of these events deserve a lot of praise for organizing the events at all and I suggest example improvements mostly so that I can give examples that the EA forum readership can relate to, not because I want to criticize the organizers of these events (who were likely trying their best and dealing with time constraints or other factors I’m not aware of).
As an example, the SERI conference in 2021 seemed to be tailoring two different audiences: students and academics. It's as if it was trying to both get academics to care about x-risk as well as serve the needs of aspiring EA students. If it had decided to focus on only one of these audiences, it could've optimized for filling that audience's needs, and made design and strategy decisions more effectively by asking "how does this achieve our goal of [onboarding students to x-risk careers]/[persuading academics]?" The alternative was making decisions by thinking about a more unnatural middle ground, which probably led to both groups feeling that the conference was more "meh" than "wow!"
If the target audience were just academics, maybe the organizing team would've spent time figuring out how to look more persuasive to academics or maybe they would've held events that allowed academics to brainstorm how to contribute to x-risk reduction (probably a pretty different type of brainstorming than students would engage in). This might have required the team disallowing students to attend the conference (a short-term cost), but it would've made it more likely that academics would return to their home institutions and tell their friends about the conference (a long-term benefit, which I suspect would dominate).
Another example is EA global -- it seems as though the main benefit of EA global is the 1-1s rather than the talks, yet talks are still put on. It seems pretty possible to me that at the next EAG, instead of running talks, effort should be dedicated toward making more 1-1s happen or making them more valuable (I hear that the organizing team is in fact moving in this direction, which I’m glad to hear!).
I think both of these possible EA infrastructure improvements are the sort that you are more likely to notice if you have product development models, since e.g. you might be aiming to figure out how to make something that 4-5 key people really love, rather than something 100s of people like.
Some product development models
Now that you're inspired to learn about product development, I figured I would share my models (note that I haven't even read a book about this topic and basically all of my knowledge comes from hanging around the Lightcone Infrastructure Team and thinking about it for myself). Additionally, you could watch 10 minutes of this video or read a book from this list of books.
- Figure out what your “customers” need (analogous to FB advertisers or ARC from previous example)
- Roughly, customers = EA orgs, EA professional, or the beneficiaries of your cause area (these needs are hard to understand though since it sort of requires figuring out how to solve a global problem, e.g. "space governance")
- If you don’t figure out what kind of people you need to attract, you won’t have an impact as a community builder
- For example, last I heard Redwood Research was in need of more senior engineers. This fact + the fact that EA is generally very young and that this might continue to be an AI safety bottleneck makes me tempted to run a community building project that aims to get more senior engineers to care about AI safety
- Goal factor your product
- EA infrastructure products will achieve some high-level goals like "get people to learn about AI safety" or "get people thinking about their careers"
- Usually, some of the goals your product achieves aren't that useful compared to the other goals
- Usually, there is a better way to achieve the most important goal then the way you are currently thinking, which is the point of goal-factoring
- Example in footnote
- High quality user-testing is crucial; high quantity (of users) is helpful, but secondary
- You should first aim to make a product a small number of people absolutely love, rather than someone that many people like because 1) that usually generalizes better to a product everyone loves and 2) you get a lot of free product recommendations and people being more likely to persuade their friends to use your thing or attend your program
- People have a lot in common, such that learning a bunch about how one person responds to a product can tell you a lot about how other people will
- The first sample contains the most information since before the first sample you have no idea what reasonable answers even look like
- Intuition pump: Think of your favorite to-do list software or some other productivity software that you think is fantastic. Have you wanted to tell your friends about it because you figured it would obviously help their productivity too? Imagine if EA infrastructure products were that good!
- Here is an example of how to apply extensive user-testing to make an article more persuasive
- Resolve key uncertainties ASAP so that you know when to return to the drawing board
- There are often key uncertainties that affect the way you develop your product or its effectiveness or both
- Example: will running a longtermist internship help onboard aspiring students to longtermist research agendas?
- Because of this, you should expect that to make a product everyone loves, you will have to go through at least several rounds of iteration where you rethink major parts of your product
- You should resolve these uncertainties ASAP and iterate fast, so that you can get to the best version of your product as soon as possible
- Example: email EA professionals asking whether they would take or actively want interns, what skills, knowledge, and competence someone would need to work with them, and whether they think an internship is the best way for someone to get involved in their field
- Alternative: Ask someone you know who works in the Berkeley longtermist offices whether you can visit to talk to professionals and then ask the professionals these questions in person
- There are often key uncertainties that affect the way you develop your product or its effectiveness or both
Go forth and create EA infrastructure!
Okay, that’s all I have to say. Please feel free to add more product development models or tips in the comments, or contest any of the points I made :)
Let’s say you think that creating a syllabus for every cause area was a high-impact community building product. Goal factoring this might look like the following chain of reasoning: “What does this product achieve?” “Well, one of the main benefits is that it makes it easier for some EA group members to learn about EA topics.” “Is there a better way to get people to learn about EA topics?” “Yes, if you can just get them to be motivated to learn about EA topics on their own, even if it is slightly harder since there is no compilation of resources, then that would be better since it turns out motivation is a bigger bottleneck than the degree to which resources are scattered or not -- even if resources were compiled, basically nothing good will happen without creating interest/motivation. And historically, there would probably be a high correlation between being willing to look into things on your own, without needing things compiled, and being high-impact “ ''How can we create motivation?” “Maybe we could run a 2-day retreat where promising intro fellows talk to smart professionals from fancy places, and they tell them the world is on track to end?”
"How much does this flonarb weigh?" - "Idk, 100 pounds?" - "Hah! No, it was 10 micrograms. What about this other flonarb?" - "10 micrograms?" -- "Close, 12 micrograms"
Strongly upvoted because I think product is an important, underrated framework for movement building (though I only skimmed).
(I think it's particularly true for building web products, and if you're running other services like 1:1s you might find fields like service design or sales more useful.)
The product development literature has informed a lot of the processes and frameworks I'm using to build Non-trivial Pursuits.
My take on the best two books to get up to speed on modern product development:
If you want to dive deeper, here's my list of resources to learn product management.
Thank you Peter! Definitely taking a look at the books and resources. Also, I now link your comment in the tldr of the post :)
Regarding figure out your customer, I saw this recent post that makes me rethink my weighing on customer personas and shift more towards "Jobs to be done". It taught me that personas are better collapsed into JtbD -
I agree, especially with the point of iterating your "product". I would take it further and add that in the beginning, the team should be trying to optimize learning. They should to open to changing the product and even throw away that v0.1 website/format. They should keep the learning and rebuild. Also consider that learning includes learning what doesn't work, so keep the lesson, archive the product and treat it as a "one time we did that experiment".