I’ve been one of two main organisers for the EA London Tech group for roughly the past year, and at our most recent meetup someone asked me to write up what I’d learned about community building. So, here is what I learned about running an EA group in 2023.

TL;DR

• The bar for me was quite low - it might be for you too.
• It’s hard to get good data about your community without a lot of investment.
• You can apply the scout mindset to the whole of your community, including what it means to be a part of that community and what direction it should go in.
• Be prepared for failure and bad things, and know beforehand who to turn to in these eventualities.
• Community building is potentially an extremely rewarding part of the whole EA experience.

Context

The group aims to cater for folks who are interested in both EA and technology, and there is no requirement to be working in the latter; in fact, we try to appeal to a fairly broad audience. We aim to have monthly meetups, but I think we only managed 11 in the end. These meetups have primarily been socials, but there have also been invited speakers and a hackathon. Attendance has ranged from 15 - 40, and feedback has been broadly positive (I feel like we generally get 4 stars out of 5), including tacit feedback from people continuing to attend our events. Both myself, and my co-organiser Sam are doing this purely voluntarily and have to fit it into our (busy!) schedules. EA UK provide a small amount of financial sponsorship for each meetup, that we usually spend on vegan pizza for attendees.

I’m a software engineer and before this year I had no experience with any kind of community building, nor any skills that seemed especially relevant. I had attended several of the meetups in 2022, and became involved with organising when the idea of holding a hackathon was floated, as I had been attempting to compile a list of EA-related open source software projects. I had somewhat intended to only help with this event as a one-off, but one of the previous organisers wanted to talk a step back, and given the relative success of the hackathon I thought it would be a worthwhile thing for me to continue doing.

Few specific skills are required to have some success as a community builder

Before starting as a community builder, I hadn’t seriously considered it as an option - I’m generally awkward and introverted and have a primarily technical skillset, with little generalist or ops experience. I’ve found that simply having some amount of agency has taken me surprisingly far - a lot of the time someone just needs to do the thing, even if the thing isn’t necessarily hard to do.^[1]

As mentioned above, I am generally an extremely busy person, as is Sam, and I would estimate we each spend 2 hours a month on the group outside of attending the actual meetups, so organising doesn’t have to be a huge time burden. If you’ve thrown a successful birthday party before, then I’d posit you have the necessary skills to get started as a community builder.

It’s difficult to get high-quality feedback or quantify impact

I’ve found it hard to get meaningful feedback, which in turn makes it difficult to understand the impact and effectiveness of the group. There are broadly four ways we’ve tried to get feedback this year, all with their own issues:

• Spontaneously asking for feedback during conversations I have at meetups. This isn’t ideal because I’m not giving the people I ask any time to think of a measured response, and I often feel like they are reluctant to offer critical feedback in such a direct way (especially if they don’t know me well personally, and therefore don’t know how receptive I might be to critical feedback). With this being said, I have had some valuable candid feedback during these conversations, for which I am extremely grateful.
• Surveying people during/after events using an online tool and a QR code etc. When we have tried this, we haven’t felt there were enough responses to feel like we had a representative sample and therefore struggled to draw conclusions. In future perhaps there are ways we might incentivise answering the question(s).
• Soliciting feedback from people who are stakeholders, eg staff from EA UK or Newspeak House (we ran most of our events there last year). This is usually valuable, but ultimately these people aren’t our target audience, and it is apparent that they often have a different perspective to community members. My intuition is that I would rather have our actions be informed directly by community members, but I also recognise that the broader and more experienced view provided by stakeholders is likely more valuable and closer to the truth than the median community member’s.
• Tacit quantitative feedback via metrics such as attendance. The problem here is it’s hard to draw causal links between correlates. If attendance is lower one week, is it because people are disinterested in the speaker’s topic, because the event is held at a new venue too far from where they live, because there was another clashing event they chose to go to instead, or any permutation therein? While not ideal, I do think this data has nudged us in the direction of our comparative advantage, which I will discuss below.

Relatedly, it’s difficult to measure the impact that running the group has, for a number of reasons:

• It’s hard to evaluate counterfactuals.
• It’s difficult to know which part of the funnel is most valuable. At the mouth, we frequently have first-time attendees who have had no prior exposure to EA before. Giving them a good first experience in the community and pointing in the direction of relevant resources about cause areas, careers and effective giving could be hypothetically extremely valuable, but this is very “hits based” and we have no systematic way to follow up with them. At the other end, we could run a purely AI safety focussed group for people who already had a good level of technical knowledge, perhaps something akin to ARENA but on an ongoing basis, and we’d provide accountability and coordinate mentorship etc. The impact here would be a lot clearer, perhaps we could help ~5 people/year enter the field 6-12 months sooner than they otherwise would’ve.^[2] It seems plausible to me that with the “wide net” approach, we could likely get a larger number of people than this involved in EA on an ongoing basis, but how much impact would they need to realistically have to make this a better option than the narrow focus?
• A significant part of the value of our group currently seems to be in preventing value drift, which seems valuable in the abstract but is also difficult to quantify.

Just because something is hard to quantify, doesn’t mean you shouldn’t try and quantify it, and in the new year Sam and I will hopefully do some BOTECs to evaluate our impact. Our struggles with feedback and impact this year could also reflect that we’re still an immature group in some ways, with a somewhat fluid identity, no codified theory of change, etc. This is something I anticipate changing next year, as we have already taken several steps towards “growing up” - see below.

Explore, then exploit applies to groups as much as individuals

As mentioned, the group is still “young” and while it has a healthy attendance, it’s never felt like we had completely found product-market fit. Earlier in the year EA UK (very kindly) suggested to us that perhaps we ought to try to come up with a more concrete vision for the group. When prompted to sit down and think about it, we realised there were multiple potential pathways to impact, and many different ways the group could look in a year’s time. Given what I discussed above about struggling to directly elicit feedback from the group about what they would find most useful, we decided to vary the format/venue/topic of the meetups in order to better gauge how we might serve the community.

We’re planning to continue these experiments for a few more iterations next year, but so far we had one event that was particularly successful and one that seemed particularly unsuccessful, and others in between. The particularly successful one was valuable in that it seemed to offer strong evidence for what our comparative advantage is (AI safety events with speakers in central London, possibly that cater for folks without much ML/technical knowledge). The unsuccessful one was particularly valuable to me in that I learned several things that I can use to improve the quality of future events:

• Avoid venues with significant travel time from central London.
• If the planned event format requires a certain attendance threshold in order to be effective, have a plan B.
• Consider advertising more widely than solely through EA channels.

I anticipate doing further exploration in the first half of 2024, and then switching into exploitation mode.

Know who to escalate to

Despite there being many highlights for me in running the group this year, the event that was by far the most memorable was when I had a report that a male attendee had acted inappropriately towards a female attendee who was attending the meetup for the first time. I felt absolutely mortified that this had happened on my watch, and felt like there might have been something I could’ve done to prevent it. I realised that I was wildly, wildly out of my depth, and perhaps should have had some kind of relevant skills or training before taking on organising. I had some amount of responsibility for the wellbeing of attendees, and it was a mistake to not have realised that before I was in this situation.

Fortunately, I had people who I could escalate to, who did have the right skills to handle this, and I think ultimately the incident was handled well. I’m being somewhat deliberately vague here to protect people’s privacy - concretely the CEA community health team should be your first port of call for incidents like these.

Community building is an excellent way to stay motivated about EA

This point is probably more about community participation than community building per se, but it’s important to me so I want to include it. My experience interacting with people at these events is very different to the experience I have interacting with EA on the forum or on social media. There is so much discussion of drama, identity, criticism and whether EA is a totalising ideology online (and indeed, specifically within EA circles) that I often find it exhausting to consume, let alone participate in. Attending these meetups and getting to have conversations with people who, I suspect, compromise the silent majority of EAs helps put things back in perspective; there are people who give 10% to GiveWell’s top recommendations and forget about EA most of the rest of the time. There are people who aren’t particularly interested in EA, but are extremely concerned about AI safety and they want to help. There are simply many people who agree with 60-80% of the shared EA views and then don’t worry about the rest, and they are perfectly happy with this. These are people who are doing a lot of good in the world but, crucially, are not over-thinking it. Many times when I read criticism of EA online, I think to myself “if only this person would come to one of our meetups with an open mind, then I think there’s a good chance they’d re-evaluate”. These people have helped me maintain and strengthen my commitment to EA, and have frequently been a joy to be around.

Conclusion

My first foray into community building this year has had its ups and its downs, but I learned a lot, and I think what I did helped to make the world a slightly better place. It has been a great way for me to learn new skills, grow my network, and evaluate my comparative advantage.

In my opinion (and that of other community builders that I’ve spoken to), the London EA scene has room to absorb more groups that are run by volunteers. I wholeheartedly believe that if I can do it, you can do it too. If you’d be interested in starting a group, then please get in touch, I’d be happy to share more of my experiences with you and point you in the direction of further resources. Similarly, if you have attended one of the meetups this year and have any kind of feedback, especially ways in which we could improve, then I’d love to hear it.

1. ^{^}

   Plausibly the group self-selects for people similar to me, who might not feel they have the relevant skills for organising events. Organising a different group might require a higher bar, but I would be surprised if it were significantly higher.

2. ^{^}

   These are made up numbers I thought of in the course of typing that sentence, this could be a dramatic overestimate.