Right now, most AI safety talent is concentrated in just a few cities: San Francisco, Berkeley, London, Oxford, and Washington D.C. Some other cities also deserve honorable mentions: Boston, Toronto, New York, Singapore, Cambridge. These cities also have an AI safety (or AI safety friendly) co-working space, such as Constellation, Mox, or LISA. However, if you look at a map of where tech talent is located in North America, we see that there are many cities with large pools of tech talent that don't have AI safety hubs. [1] I think that this is a gap the ecosystem should try to close, as the field rapidly scales and many leading voices state the need for more scaling in the future.
I think it’s worth clarifying what I see the role of an AI safety hub in a city as. Concretely, to me, this means the creation of a co-working space which is specifically dedicated to growing the AI safety ecosystem in that city. This co-working space can:
The first reason there should be more AI safety hubs is that it reduces the barrier to career transitions for mid-career professionals (and therefore increases the high-quality talent pool). When you have an established life in a city, moving is not a trivial task, yet many AI safety jobs require (or at least heavily favor) relocation to an AI safety hub. Mid-career professionals often cannot uproot their life to move to the bay in the way college students or recent graduates can, regardless of how convinced they are that reducing existential risk is important to work on. As a result, they may be pushed away from making a serious career transition, reducing the talent pool available to the field as it scales (both through existing organizations and the founding of new ones). Having an AI safety hub in their city may make this career transition significantly easier, giving potential AI safety employees access to local impactful jobs.
This is similarly true for potential founders. Maybe someone is passionate about reducing AI x-risk, and has credible ideas about how to do it through the foundation of a new organization, but lacks the ability to move to an AI safety hub where they can be surrounded by peers taking their work seriously and enriching their intellectual environment. Local AI safety hubs can incubate these potential founders, creating jobs in the local city (which further helps build the local AI safety community) while reducing many of the logistical barriers to founding a new org.
AI safety hubs also serve as great locations for networking. People considering a career transition would be given a better space to meet and get to know people working in AI safety, and explore opportunities for them to contribute. Events thrown by the hub can serve a similar purpose.
Too much concentration of the AI safety movement can also be suboptimal. If too much of the AI safety movement is co-located, this can lead to excessive groupthink and a lack of pursuit of novel ideas. Already, we can see distinct identities and priorities between the UK and Bay Area AI safety movements; this encourages debate and as a result improves the quality of output and direction on both sides. More perspectives and identities, and therefore more intellectual diversity, can continue to lead to better overall strategy and novel ideation.
The hub can also serve to raise the productivity of existing remote workers in the city. Co-working spaces provide motivation and intellectual stimulation, as well as a professional environment that can enable more focus than working from home. The hub can also host workshops and seminars that boost employees' knowledge and provide them a space to take a step back from day-to-day work to stay up to date with new developments in the field.
I see four main arguments against setting up more AI safety hubs:
Thank you to Harry Waterman, James Lester, and Matt Handzel for feedback and comments on an initial draft of this post.
Seattle, Los Angeles, Dallas, Austin, Denver, Atlanta and Montreal are all examples.