Thanks so much for your kind words on our post, Nick! I really appreciate it.
One of the non-governmental barriers to relocation for international folks is the general non-accessibility of relevant information. Even something as basic as finding an apartment to rent in a foreign city could present a quite high barrier (and certainly a perceived barrier) to relocation.
This is such an incredibly useful resource, Vael! Thank you so much for your hard work on this project.
I really hope this project continues to go strong!
Thank you so much for this extremely helpful suggestion, Linch! I really appreciate it.
A thought: Especially when enabled by technology, people are very capable. In theory, a person can easily offset the negative impact of their greenhouse gas emissions and have a lot of time and resouces left over to pursue positive impact. For example, by donating a fraction of their money to carbon offsetting projects and not having a polluting lifestyle, the median American can easily have a net reducing effect on global greenhouse gas emissions throughout their lifetime. Also, I think the median person in the world can in theory achieve a net reducing effect as well, by devoting a fraction of their time and resources to planting trees (nature’s baseline technology for carbon capture).
So perhaps the right framing isn’t “Should you have children despite climate change?” One alternative framing is: suppose you want to influence the next generation, who will either capably help the world or capably harm the world. Should you do it by parenting and influencing your own children, or by influencing other people’s children?
I think most EAs favor the latter option, and indeed there is a compelling argument in favor of it. Humans are perhaps the only species whose primary mode of phenotypic inheritance is learning knowledge and values from group members: including not only parents, but also a lot of other people. This is why we are so adaptable and capable.
But for EAs who derive a lot of pleasure from parenting and are high-fidelity influencers as parents (i.e., have a high guarantee of influencing their children to have similar values as them), I think parenting can be an excellent use of their time and resources. I think optimal parenting is a domain which is quite neglected by EAs, and hope that this changes moving forward.
That makes sense! Shoes are probably more expensive than malaria nets.
But it might still be a better intervention point than antivenom+improving diagnosis+increasing people’s willingness to go to the hospital.
What about something they can wear on their leg to prevent the snakebite?
Thank you so much for your kind words, Max! I’m extremely grateful.
I completely agree that if (a big if!) we could identify and recruit AI capabilities researchers who could quickly “plug in” to the current AI safety field, and ideally could even contribute novel and promising directions for “finding structure/good questions/useful framing”, that would be extremely effective. Perhaps a maximally effective use of time and resources for many people.
I also completely agree that experiential learning on how to talent-scout and recruit AI capabilities researchers is likely to be also helpful for recruiting for the AI safety field in general. The transfer will be quite high. (And of course, recruiting junior research talent, etc. will be “easy mode” compared to recruiting AI capabilities researchers.)
Thank you so much for your feedback on my post, Peter! I really appreciate it.
It seems like READI is doing some incredible and widely applicable work! I would be extremely excited to collaborate with you, READI, and people working in AI safety on movement-building. Please keep an eye out for a future forum post with some potential ideas on this front! We would love to get your feedback on them as well.
(And thank you very much for letting me know about Vael’s extremely important write-up! It is brilliant, and I think everyone in AI safety should read it.)
I think Elon Musk said it in a documentary about AI risks. (Is this correct?)
Quoted from an EA forum post draft I’m working on:
“Humans are currently the smartest being on the planet. This means that non-human animals are completely at our mercy. Cows, pigs, and chickens live atrocious lives in factory farms, because humans’ goal of eating meat is misaligned with these animals’ well-being. Saber-toothed tigers and mammoths were hunted to extinction, because nearby humans’ goal was misaligned with these animals’ survival.
But what if in the future, we were not the smartest being on the planet? AI experts predict that it’s basically a coin flip whether or not the following scenario happens by year X. The scenario is that researchers at Deepmind, Google, or Facebook accidentally create an AI system that is systematically smarter than humans. If the goal of this superintelligent, difficult-to-control AI system is accidentally misaligned with human survival, humanity will go extinct. And no AI expert has yet convinced the rest of the field that there is a way to align this superintelligent AI system’s goal in a controlled, guaranteed manner.”
Thank you very much for the constructive criticisms, Max! I appreciate your honest response, and agree with many of your points.
I am in the process of preparing a (hopefully) well-thought-out response to your comment.
Thank you so much Jay for your kind words!
If you happen to think of any suggestions, any blind spots of the post, or any constructive criticisms, I’d be extremely excited to hear them! (Either here or in private conversation, whichever you prefer.)
Thanks so much for your comment, Owen! I really appreciate it.
I was under the impression (perhaps incomplete!) that your definition of “phase 2” was “an action whose upside is in its impact,” and “phase 1” was “an action whose upside is in reducing uncertainty about what is the highest impact option for future actions.”
I was suggesting that I think we already know that recruiting people away from AI capabilities research (especially into AI safety) has a substantially high impact, and this impact per unit of time is likely to improve with experience. So pondering without experientially trying it is worse for optimizing its impact, for reducing uncertainty.
The best use of time and resources (in the Phase 2 sense) is probably to recruit AI capabilities researchers into AI safety. Uncertainty is not impossible to deal with, and is extremely likely to improve from experience.
I completely agree with the urgency and the evaluation of the problem.
In case begging and pleading doesn’t work, a complementary method is to create a prestige differential between AI safety research and AI capabilities research (i.e., like that between green-energy research and fossil fuels), with the goal of convincing people to move from the latter to the former. See my post for a grand strategy.
How do we recruit AI capabilities researchers to transition into AI safety research? It seems that “it is relatively easy to persuade people to join AI safety in 1-on-1s.” I think it’s most likely worth brainstorming methods to scale this.
My prior is that one’s degree of EA-alignment is pretty transparent. If there are any grifters, they would probably be found out pretty quickly and we can retract funding/cooperation from that point on.
Also, people who are at a crossroads of either being EA-aligned or non-EA aligned (e.g., people who want to be a productive member of a lively and prestigious community) could be organizationally “captured” and become EA-aligned, if we maintain a high-trust, collaborative group environment.
A general class of problems for effective altruists is the following:
In some domains, there are a finite number of positions through which high-impact good can be done. These positions tend to be prestigious (perhaps rationally, perhaps not). So, there is strong zero-sum competition for these positions. The limiting factor is that effective altruists face steep competition for these positions against other well-intentioned people who are just not perfectly aligned on one or more crucial issues.
One common approach is to really help the effective altruists to break through this competition. But this is hard. Another common approach is to try to convince non-effective altruists who have successfully broken into these positions to be more EA-aligned. But convincing experienced people is often difficult (you can’t teach an old dog new tricks, generally speaking).
A thought I can’t shake is that if we could reduce the competition somehow (expand the pie, target young and high-potential people, and more controversially, to convince non-EA-aligned people to drop out of the race) it would be much more feasible.
So one alternative is to have a preprint server like arXiv (where papers can be posted) that directly serves as a journal, potentially with peer reviews that are also posted. Independent of paper availability to the public, this would also save researchers’ time. (Instead of formatting papers to fit the Elsevier guidelines, they could be doing more research or training new researchers.)
Thanks so much for this extremely important and well-written post, Theo! I really appreciate it.
My main takeaway from this post (among many takeaways!) is that EA outreach and movement-building could be significantly better. I’m not sure yet on the clear next steps, but perhaps outreach could be even more individualized and epistemically humble.
One devil’s-advocate point on your point that “while it may be true that there are certain characteristics which predict that people are more likely to become HEAs, it does not follow that a larger EA community made up of such people would automatically be better than this one.” Despite Goodhart’s Law, I think that there is some definition of HEA such that maximizing the number of HEAs is the best practical strategy for cooperative movement-building. Having a lot of dedicated people in a cooperative group is very important, perhaps the most important factor in determining the success of the group. More complicated goals/guidelines for movement-builders are harder to use, both for individuals and for group coordination.