Make ALERT happen
The idea of a rapid emergency response team, like ALERT [^1]has been floating around. Subsequent people have at times partially taken up the mantle, but because of the vicissitudes of life, they have each given up.
I think the project is one of the more promising ideas to come out of the EA brain trust in recent years. I don't think the project is limited on theorizing (see e.g., a past detailed proposal [^2]), but on execution:
- Gather a team of reservists -- done by previous projects, easy to pick up
- Immediate next step: Sort through those who could join a pop-up effort immediately, and those who are interested in theory. Have the more committed reservists talk with bosses to be able to drop their day job during an emergency.
- Step after that: Look for funding so that reservists don't have to internalize the costs completely. Organize reservists so that they are some available at any timezone.
- Gather a team of forecasters and alert signals -- easy to do in collaboration with, for instance, Samotsvety, or with the deep pool of talent around the forecasting community
- Immediate next step: Set up an MVP, any MVP, of a method which could call a rapid response team into action.
- Step after that: Improve alarm raising so that it catches more emergencies in time, and throws up fewer false positives.
- Run training exercises
- Immediate next step: Design a training exercise that has the team of reservists deal with something which, to them, was an "unknown unknown". This could be a completely made-up exercise, but my preference would be for this to be a regional event of some significance, or an attempt to solve a valuable but non-urgent problem urgently.
- Step after that: Using the forecasting/alert signal team, choose the next non-catastrophic event as it pops up, and react and act in real-time.
Personally, I am interested in this project because it doesn't require me to trust the concepts [^3] or for us to be able to forecast AI precisely [^4]; it doesn't require future catastrophes from AI to correspond to the Yudkowskian framework, or to any preconceived notion of AI. Indeed, in generality, having a reserve team that can act during emergencies doesn't require future threats to humanity to come in the shape of AI; the more surprising the future turns out to be, the better this intervention looks related to alternatives. It is a bet on variance. I have been looking for a project to commit to, and this seems like a good fit.
- The ask: Cash money to get this started; to do the steps in "immediate next steps". I will also put money and time of my own, financed by my lucrative consulting [^5]. But this is easier if a few people each pitch $20K to $50K.
- The ambitious ask: A sizable budget to carry out the "step after that" items, and to not be bottlenecked on money, ever. Previous projects aimed to get this and failed. I think this is a red-herring, because it requires getting the trust of large funders, who tend to move slow and be slow to trust.
- The institutional structure: I will set up an institutional structure if I have to, but prefer not to, at least at first.
[^1]: https://forum.effectivealtruism.org/posts/sgcxDwyD2KL6BHH2C/case-for-emergency-response-teams
[^2]: https://docs.google.com/document/d/1IStSw8MIU1z68NjAmMHBM3gvzaBRrj6W_Wuw6hY4ICE/edit
[^3]: https://nunosempere.com/blog/2023/01/23/my-highly-personal-skepticism-braindump-on-existential-risk/
[^4]: https://nunosempere.com/blog/2023/11/07/hurdles-forecasting-ai/