A small group of AGI existential safety field-builders and I are starting research exploring a potential initiative about informing the public and/or important stakeholders about the risks of misaligned AI and the difficulties of aligning it.

We are aware that a public communication initiative like this carries risks (including of harming the AGI x-safety community’s reputation, of sparking animosity and misunderstandings between communities, or drawing attention to ways to misuse or irresponsibly develop scaleable ML architectures). We are still in the stage of evaluating whether/how this initiative would be good to pursue. 

We are posting this on the forum to avoid the scenario where someone else starts a project about this at the same time and we end up doing duplicate work. 

How you can get involved:

  • If you are currently undertaking work similar to this or are interested in doing so, message me your email address along with a bit of context about yourself/what you are doing. 
  • We are drafting a longer post to share our current considerations and open questions. Message me if you would like to review the draft.
  • We are looking for one or two individuals who are excited to facilitate a research space for visiting researchers. The space will run in Oxford (one week in Sep ’22) and in Prague (9-16 Oct ’22) with accommodation and meals provided for. If you take on the role as facilitator, you will receive a monthly income of $2-3K gross for 3 months and actually get to spend most of that time on your own research in the area (of finding ways to clarify unresolved risks of transformative AI to/with other stakeholders). If you are interested, please message me and briefly describe your research background (as relevant to testing approaches for effective intergroup communication, conflict-resolution and/or consensus-building).

32

5 comments, sorted by Click to highlight new comments since: Today at 2:45 AM
New Comment

I’m confused why this got downvoted without any comments; maybe someone thought this was duplicative of an older post/project? Regardless, assuming it isn’t duplicative I think it sounds like a good idea, and even if it is duplicative it might still be good to have another project attempt (so long as this one integrates lessons from the old one).

Great idea to look into this!

It sounds a lot like what we have been doing at the Existential Risk Observatory (posts from us, website). We're more than willing to give you input insofar that helps, and perhaps also to coordinate. In general, we think this is a really positive action and the space is wide open. So far, we have good results. We also think there is ample space for other institutes to do this.

Let's further coordinate by email, you can reach us at info@existentialriskobservatory.org. Looking forward to learn from each other!

I'm the author of more than 100 high-quality entries in the AI Safety Arguments Competition and the writer of this post. I effectively majored in US-China affairs, specifically on the axis of AI and tech policy. I live and breathe this stuff, in a provable way, and I will DM you my email address because I intend to contribute significantly.  

But I want to clarify that mass public outreach, by default, comes with profoundly complicated and unpredictable risks, and it's certainly not the kind of thing that someone can intuitively conclude is a good or bad idea.

This is exactly the kind of domain where very smart and insightful people trip up, in ways that no amount of intelligence could give someone a fair chance of getting right without deliberate guidance from someone with arcane, technical experience in the area. In a best case scenario, it is reinventing the wheel; in the worst case scenario, it gets the entire concept of AI safety burned

A couple weeks ago, I wrote a really helpful flowchart designed to introduce newcomers to international AI policy, so at the very least they don't waste a ton of time and energy reinventing the wheel: https://www.lesswrong.com/posts/brFGvPqo8sKpb9mZf/the-basics-of-agi-policy-flowchart

Ultimately, the best strategy is meeting policy folk and talking to them, nothing on the internet can really substitute for that.