<aside> ▪️ Sections
</aside>
<aside> 💡 Our Recommendation:
Get your core organizer demographics correct from the outset. Make sure you have key demographics represented early on.
</aside>
This means that if there’s any dimension on which your group is lacking at its inception (epistemic norms, diversity, ML knowledge, undergrad vs. grad), it will be hard to correct for this later on. Therefore, it’s worth investing energy in making sure your initial group captures the things you care about.
On the one hand, this can be used to your advantage because if you can build the right kind of group from the beginning, there’s a good chance the demographics of the group will be relatively sticky as the group groups and long after you are part of the organization.
But conversely (and more probably), if you have a major demographic shortcoming, it will be harder to change this later on. This happens because we are constantly, sometimes consciously but usually unconsciously, trying to determine whether we belong in a space or not.
If you are a grad student who cares a lot about AI safety but your school’s AI safety group doesn’t have any other grad students, it will take a much higher bar for you to feel comfortable enough to join. If you have AI safety research experience and have read many research papers, and you show up to your university’s AI safety group’s event and no one seems to know what they are talking about, you will feel very out of place and likely not join. As much as AI safety university groups are about research, sharing ideas, and learning, they’re also inherently social environments. People want to fit in, and if they feel out of place, they will spend their time elsewhere.
We do not want AI safety groups to appeal to everyone. If the majority of students at your university felt they were a good fit for your organization, you likely would not be doing your job correctly. We want you to specifically seek out intelligent, kind, value-aligned members who care about AI safety and are willing to dedicate their careers to solving this issue.
We’re hoping to mitigate situations where there are members who meet all these requirements but don’t feel like they are represented by your group. Members of your organization should have common traits and a vibe that makes you all feel cohesive, but as your group expands, it should cater to a few different demographics or archetypes.
If, for example, you don’t have a lot of members who identify as female, you’re potentially missing out on 50% of the group of members that could have counterfactually joined the club.
Many people reading this likely already have AI safety organizations and might be worried about some of the demographics of their group, given the makeup of the founding group. The good news is that this can be changed with a concentrated amount of effort.
An anecdote from Neav’s experience building up TAMID at Northeastern was that the first leadership team he served on had nine men and two women. Predictably, the club as a whole had a major gender ratio balance problem. The leadership team during recruitment was very conscious of trying to make women at recruitment events feel comfortable and did more personal outreach to women to encourage them to apply.
As a result, there were many women in the applicant pool, and the gender balance of the next accepted group greatly improved. After this effort, there were even more women to help with recruitment, and within one or two more semesters, the gender problem was fixed. This narrative is a bit of a simplification of what happened and how much work and intention were put into fixing this issue. But the key takeaway here are that imbalances in your club’s demographics, such as diversity, knowledge, norms, etc., can be fixed through intention, effort, and time. However, it’s usually a better use of time to try and get this right from the beginning.