<aside> ▪️ Sections
</aside>
<aside> 💡 Our Recommendation
Lean towards the side of exclusivity. Most results your group produces will come from a small handful of students. After running experiments at many universities, groups that leaned towards smaller groups of higher context members fared better than larger groups with low context ones.
</aside>
Organization naturally (and early on) have to decide whether to be more open or exclusive
A key consideration your group will have to make is how exclusive versus inclusive to be. At first, it might be difficult to worry about exclusivity as you struggle to even find other students at your university who are interested in forming an AI safety group.
Eventually, you will have a group, create applications and be tasked with determining who to let in. Everyone? Only the best of the best? This is not an easy decision.
Everyone single person you accept will soak up energy, attention, and resources from the club. They will also take up space and energy in the room during AI safety discussions. It’s important to contemplate whether every individual accepted would be a good use of resources and will have a net-positive presence when it comes to AI safety discussions. For example, someone who’s only interested in AI ethics but not at all interested in AI safety could reduce the quality of conversations had at your group.
The level of exclusivity could also serve as a draw for strong talent and give students confidence that they are in the right room and working on solving the right problem. The opportunity cost of dedicating your time to anything while at university is incredibly high. There are typically hundreds, if not thousands, of groups and communities competing for the finite attention of the brightest students. University students are generally pretty risk adverse and worried about ensuring they will be employed after graduating. If your group doesn’t inspire confidence that you can help with that pursuit, you will likely lose the talent you are looking to reach to other, more established groups.
Growing the group so quickly that the average member and discussion quality become too low, could result in high quality people no longer wanting to join or interact with the group. This is a huge problem because the goal of each AI safety university group should be to attract the strongest talent at your university interested in AI safety. If you end up in this situation, you likely lose the mandate to be “the AI safety group” on your campus and could fail to produce talent to join the AI safety space after graduating, which should be another key objective of yours.
Overexpansion can be avoided by:
There may be many people at your university who are interested in AI safety, and you could fail to capture valuable talent by being too exclusive.
One goal of your group should be to help develop students into future leaders in the AI safety space and it is unrealistic to expect all students to show up at your organization as 100% perfect candidates that check every box. Your organization should be a vehicle of growth that helps students get more involved, and we can’t have great AI safety leaders if no one ever takes a chance on them.
The skills and qualities that are required to be impactful in the AI safety space are diverse and likely to grow and change over time. This space needs brilliant minds who can solve hard technical research problems, individuals with strong people and organizational skills who can be great operators, and great negotiators and collaborators who can make important policy and governance progress, to name a few.
Conversations could likely benefit from diverse perspectives. Many AI safety problems are multidisciplinary and require input from people with exposure to different fields and ideas. If done correctly, bringing students studying different subjects who all deeply care about the risk posed by AI could help elevate conversations and prevent groupthink.