I do various things to support the formal alignment research ecosystem and, to a lesser extent, the rationality/EA/int/a communities at large.
Matilda
Karma: 72
Thank you for this.
In this post, we aim to provide a high-level overview of our approach. As you may have seen from the other comments people have been posting on LW, there was a lot to cover in numerous directions, and we want to be thoughtful in how we represent other entities. I appreciate you raising this.
As we said at the top of our post, we want this to be the beginning of a conversation, and there’s a lot more we plan to say. We’ll work our way through comments as we go.
Existential AI safety needs an effective social movement. PauseAI is building it.
Thanks, Ollie. In addition to what Joseph said, we also have a volunteer <> project matching database you can sign up for (https://catalyse.up.railway.app/), which you can set to alert you by email when a project matching your skills/interests come up, or you can check it when you want to and decide which projects you want to get involved with. You can also subscribe to our events calendar (luma.com/pauseai_uk).
Simple summary of AI Safety laws
I considered what I’ve learnt while working in animal advocacy and AI safety. I think these are the 2 most urgent/important cause areas given actuality of/potential for harm/suffering. I ranked orgs based on who I think is doing good work and has sound epistemics, roughly.
Thanks for this, OP!
I hadn’t seen the latest updates to the site since ~launch and have sent my own feedback.
Fellow Forum users—consider taking a few min to look at the site and give their sense of the UX or whatever other feedback feels useful.
You, yes, you could help aisafety.com visitors have a smoother/more impactful experience, adding real value to the time spent by the team on this already at a pretty minimal cost to you :)
Thank you for this post. I think I broadly agree. I feel:
a) some resistance around your framing and valuation of investment in community building in some of its forms, though I appreciate your arguments and don’t really dispute them—it’s more a case of how I would weigh them up against other factors.
Namely, one lesson I take from the FTX affair and other issues that have come up in our movement in recent months/years is that having a more open, inclusive, dynamic community that ‘recruits’ among more diverse socio-economic/cultural backgrounds (even if new entrants largely stay on the periphery of e.g. key decision-making as they build a clear model of how EA looks at the world, get a better sense of what intervening in their fields of interest might look like if they join as students/early career professionals rather than high-ranking experts) could give our movement more resilience by increasing the aggregate balance in what we expect to be normal (in terms of ways of securing capital, degrees of power concentration and risk taking we deem acceptable, etc. - note that I’m not advocating against centralised functioning for certain aspects of the cooperative work we set out to do).
If anything, perhaps the fact that ‘core EA’ seems to not really have seen the FTX debacle coming nor be too troubled by a select few deciding how money gets allocated despite the stakes and the degree of uncertainty and volatility of everything we’re tinkering with suggests to me that we should invest in what I’ll call higher perspectival pluralism, for want of a better descriptor (I’m writing on a very noisy broken down train and my attention is pretty badly challenged by that, in case any of this doesn’t feel too smooth to read). I realise this must seem abhorrent to read to some people—I’m not saying expertise should not play a major role in decision making in EA, which means a lot of decision making shouldn’t be democratic in the sense where everyone’s opinion, however well or poorly informed, counts equally. But I think it’s worth considering that there’s a lot of steering of the movement that exceeds the bounds of technical decisions best made by people who really know their field. Norms around behaviour and risk taking are examples of what I mean by ‘steering of the movement’.
b) curiosity around how you construe the increased ‘moderation’ you find yourself drawn to. It’d be interesting to flesh out what this entails. My sense is that this ‘moderation’ is essentially a response to the ‘naïve optimising’ that has been discussed in some previous comments. It’s a sound response, in my view, and it bears breaking down into something with higher descriptive signal.
The way I would propose to conceive of this ‘moderation’ goes something like this—excessively optimising for an outcome fails to support the resilience of the wider system (whichever system we may be looking at in a given instance) by removing the ‘inefficiencies’ (e.g. redundancies) that provide buffers against shocks and distortions, including unknown unknowns. Decreased resilience is bad. It’s not compatible with sustained robustness—it displaces pressure, often in poorly monitored ways that end up blowing up in our faces unexpectedly. Or worse, the faces of other stakeholders who had been minding their own business all along.
To be clear, my understanding is you are mostly considering embracing more moderate views with regards to EA’s potential to achieve a flawless record, to be a desirable social bubble, etc. - I think my take on this applies to this too (and again, I agree with your general sentiment and am simply trying to clarify what moderation might mean in this context).
In other words, welcoming some degree of inefficiency could actually be a good way to translate the epistemic humility most of us strive for (and the moderation you speak of) into the design and implementation of our initiatives. I’d like to see any approaches to what this could look like from a design/operational research perspective, if anyone has come up with even tentative models for this.
So my sense is that we should:
- be wary of endeavours like FTX that are willing to compromise important aspects of how we’re trying to impact the world for the sake of maximal effectiveness, and—
encourage people to build sustainably robust impact across the board rather than achieving Everything Right Away With Maximal ROI but with serious tunnel vision. In other words, valuing effectiveness means, under conditions of uncertainty, sacrificing some efficiency (as a side note, I suppose this is one of the underlying assumptions for point #a above). Obviously this wouldn’t be the case if we were omniscient, but I don’t think any of us is arguing that, especially post-FTX. So until then, I think we should value slack more highly across the board.
Thank you so much for posting this! I’ve just submitted an application for a civil service position (before reading your post, alas) and while I know they’re all very competitive roles, as someone who didn’t study at a super top university for various reasons, I loved the blind recruitment process and the guidance available online. Fair, clear and reasonable. Your thoughts are really valuable too and align perfectly with what I took away from the online guidance.
Hi Lewis, I’d love to know if there are any particular aspects of research on existing, emerging or desirable strategies to promote the transition of agricultural subsidies away from animal (especially factory) farming and towards plant-based agriculture that you would like to see more of. Context—I’m beginning to design a MSc dissertation (in animal welfare science, ethics and law, UK-based) and think this would be a good area to generate a bit more work on, but it’s obviously moving fast so I’d love your thoughts on any especially worthwhile approaches to this.
On another note, do you think social media advocacy for animals is at risk of reaching a banalisation/fatigue/desensitisation/maximum impact plateau or even decrease anytime soon, or do you see as being something worth continuing investing in fairly aggressively, as many organisations seem to be doing?
Thanks for your good work!
Thanks, Dave. That would be pretty silly of me (I don’t like feeling I’m talking to a wall, and I don’t want you to either), and I’m glad you asked for clarification rather than just concluding that was what I meant to do.
I’m sad that landed this way for you. I’ll try and say this again, and I hope this lands as the appreciative and cooperative reply I aimed to write:
- there were a lot of things to say in this post, and there are several important things we didn’t have space to go into (seriously, that post would have been unbearably long).
- the whole PauseAI and PauseAI US ‘parallel ontogeny’ arc is not central to this overview, especially since PauseAI US are their own entity and we don’t want to muddle that distinction by including them in this already long overview of our own org (it’s already pretty confusing to people that there should be 2 orgs with a related name/brand that aren’t the same org).
- we do want to say more about that whole arc, and we will. If politicians’ or customer service agents’ promises leave you expecting that we won’t honour what we’ve committed to doing, you have my commiserations, and I hope you’ll assume good faith for a few days.
We’re resource-constrained and have other things to attend to; we’re working through comments at the pace at which we can reply clearly and thoughtfully, and we want to make sure everything we say about our ~shared history and present with PauseAI US is accurate. I replied to you over the weekend because I got the sense that there was some urgency on your side in knowing whether we intended to say more about this, and it seemed like a reasonable thing to do to confirm that. I only joined PauseAI’s federation as a staff member a few months ago, so it wouldn’t make sense for me to try and address this area I’m not super familiar with in a comment, but we do want to answer you, as well as other people who’ve asked questions we haven’t yet got round to replying to (substantively or at all).
Please give us a few days to work through comments.