[Epistemic status: Shallow dive into research questions, backed by some years of on-and-off thinking about this kind of plan.]

Introduction

There is some chance that civilization will cease to function before we hit an intelligence explosion. If it does, it would be good to preserve existing alignment research for future generations who might rebuild advanced technology, and ideally have safe havens ready for current and future researchers to spend their lives adding to that pool of knowledge.

This might delay capabilities research by many decades, centuries, or longer while allowing basic theoretical alignment research to continue, and so be a potential Yudkowskian positive model violation for which we should prepare.

Setting this infrastructure up is a massively scalable intervention, and one that should likely be tackled by people who are not already on the researcher career path. It would have been good to get started some years ago given recent events, but now is the second best time to plant a tree.[1]

Preserving alignment knowledge through a global catastrophe

What data do we want to store?

Thankfully, the EleutherAI people are working on a dataset of all alignment research[2]. It's still a WIP[3] and contributions to the scripts to collect it are welcome, so if you're a programmer looking for a shovel ready way to help with this then consider submitting a PR[4].

How do we want to store it?

My shallow dive into this uncovered these options:

  • We could print it out on paper
    • Lifetime: 500+ years in good conditions (might depend significantly on paper and ink quality, more research needed). Vacuum sealing it with low humidity seems like it would help significantly.
    • Pros: Totally human readable.
  • Microsoft's Project Silica is the longest lasting option I could find
    • Lifetime: 10000+ years
    • Cons: Would require high levels of technology to read it back. I'm not seeing an option to buy the machines required to write new archives and expect them to be very advanced/expensive, so this would be limited to storing pre-collapse research.
  • CDs could be a minimalist option
    • Lifetime: Maybe 50 years  if stored in good conditions
    • Pros: Good ability for researchers to explore the information on computers while those last)
    • Cons: It's very plausible that a severe GCR[5] would set us back far enough that we'd not regain CD reading technology before they decayed so they aren't a full solution.
  • The Arctic World Archive seems worth including in the portfolio
    • Lifetime: 1000+ years
    • Pros: It's a pretty straightforward case of turning money into archives
    • Cons: Not very accessible in the meantime
  • The DOTS system (a highly stable tape-based storage medium) might be a strong candidate, if it is buyable.[6]
    • Lifetime: 200-2000+ years
    • Pros: Human readable or digital archives. 

Each has advantages, so some combination of them might be ideal.

Where do we store it?

Having many redundant backups seem advisable, preferably protected by communities which can last centuries or in locations which will not be disturbed for a very long time. Producing "alignment backup kits" to send out and offering microgrants to people all around the world to place them in secure locations would achieve this. We'd likely want basic (just pre-collapse work) and advanced (capable of adding archives for a long time post-collapse) options.

If you'd like to take on the challenge of preparing these kits, storing an archive, or coordinating things, please join the Alignment After A GCR Discord (AAAG). I'm happy to collaborate and give some seed funding. If you want to help collect and improve the archive files, #accelerating-alignment on EAI is the place to go.

Continuing alignment research after a global catastrophe

It is obviously best if as many people survive the GCR as possible, and supporting the work of organizations like the Alliance to Feed the Earth in Disasters seems extremely valuable. However, a targeted intervention to focus on allowing alignment researchers to continue their work in the wake of a disaster might be an especially cost-effective way to improve the long-term future of humanity.

Evacuation plans

A list of which researchers to prioritize would need to be drawn up.[7] They would need instructions on how to get to the haven, ideally someone with reliable transport to take them there. In case of moments of extreme risk, they would be encouraged to preemptively (and hopefully temporarily) move to the haven.

Designing havens

The locations would need to be be bought, funded, and partially populated before the GCR.[8] I have some ideas about which other subcultures might be good to draw from, with the Authentic Relating community top of the list.[9]

The havens would need to be well-stocked to weather the initial crisis and recover after. They should be located in places where farming or fishing could produce a surplus in the long term to allow some of the people living there to spend much of their time making research progress. Being relatively far from centers of population seems beneficial, but close enough to major hubs that transport is practical. There are many considerations, and talking to ALLFED to get their models of how to survive GCRs seems like an obvious first step to plan this.

Avoiding the failure mode of allowing so many people to join that the whole group goes under would be both challenging and necessary. Clear rules would have to be agreed on for who could join.

The culture would need to be set up to be conducive to supporting research in the long term while being mostly self-sufficient, this would be an interesting challenge in designing community. People with the skills to produce food and other necessities would need to be part of the team.

Call to action

Even more than archiving, this needs some people to make it their primary project in order for it to happen. That could include you! I would be happy to provide advice, mentorship, connections, and some seed funding to a founder or team who wants to take this project on.[10] Message me here or @A_donor on the Discord.

This project could also benefit from volunteers for various roles. If you or someone you know would like to help by

  • Searching for locations
  • Potentially moving to a haven early and helping set up
  • Researching questions
  • Putting us in contact with people who might make this work (e.g. people with experience in self-sufficient community building)
  • Doing other tasks to increase the chances that we recover from GCRs with a strong base of alignment theory

Please join the Discord and introduce yourself, specifically indicating that you'd like to help with havens so I know to add you to those channels.

I can fund the very early stages of both projects, but in order to scale it to something really valuable we would need major funders on board. If you are or have access to a major funder and want to offer advice or encouragement to apply that would increase the chances that this goes somewhere.

It's quite likely that I won't post public updates about the havens part of this project even if it's going relatively well, as having lots of attention on it seems net-negative, so don't be surprised if you don't hear anything more.

  1. ^

    "The best time to plant a tree is twenty years ago. The second best time is now." - Quote

  2. ^

    They want to use it to train language models to help with alignment research, but it aims to contain exactly what we'd want.

  3. ^

    Work In Progress

  4. ^

    Pull Request - A way of suggesting changes to a repository using version control, usually used in programming.

  5. ^

    Global Catastrophic Risk - An event which causes massive global disruption, such as a severe pandemic or nuclear war.

  6. ^

    The website is unclear on whether it's immediately available.

  7. ^

    If you're a researcher and want to be on the list, feel free to contact me with your location and I'll keep track of everyone's requests. We might possibly use Alignment EigenKarma as an unbiased metric to prioritize if that exists in time.

  8. ^

    Unless anyone knows of good places which might be joinable already, if you do please message me!

  9. ^

    They are compatible with Rationalist/EA culture, more likely than most to be able to create stable communities, and some of them like the idea of building strong community for the benefit of all of humanity.

  10. ^

    I have a reasonably strong track record as a Mentor/Manager/Mysterious Old Wizard/Funder package deal. If you're enthusiastic and bright don't worry if the task seems overwhelming, I can help you pick up the skills and decompose tasks.

36

11 comments, sorted by Click to highlight new comments since: Today at 1:20 AM
New Comment

I wrote about this on EA Forum a few days ago. I'm glad others are starting to think about this. I do think archiving all existing alignment work is very important and perhaps equally important as efforts to keep alive people who represent existing experts & talent in the field. It would be much better for them to be able to continue their work than for new people to attempt to pick off where they left off, especially since many things like intuitions honed over time etc. may not be readily learnable.

I'm increasingly inclined to think that a massive "shock" in the near future (like a nuclear war or a severe pandemic) which effectively halts economic progress, perhaps for a few decades or more, then restarts it at a lower baseline, may be one of the few remaining scenarios we can reasonably expect to survive AGI, taking into account the grim present strategic situation as Eliezer outlined in the recent sequence. Such a world might especially favour alignment since AI work (prosaic AI especially) seems to be much more capital intensive than alignment work, so in a post-shock world with less capital available it would be disadvantaged or impossible to continue carrying out at all. There are a few other reasons such a catastrophic shock may actually increase our collective odds of success re: AI risk, such as a greatly reduced population implying fewer AGI projects & race pressures, etc., morbid as it is.

Given this, the OP's project is doubly important.

Assuming your beliefs as stated above are truly held, why shouldn’t I be worried that you’ll try to deliberately induce such a “shock,” and thereby undertake action to kill a significant percentage of the (currently living) population?

(Apologies for being horribly blunt, not sure how else to word this)

[This comment is no longer endorsed by its author]Reply

Interesting! One potential downside my mind immediately goes to is public perception, in the (hopefully probable) case that such a contingency plan isn’t needed. In popular culture, the idea of a privileged (usually very wealthy) class of people escaping to an “ark” as the world ends for everyone else is generally considered a classic evil villain trope. For instance, in Don’t Look Up (a recent Hollywood blockbuster involving a GCR), the good guy scientists are offered a refuge in the evil president’s secret escape spaceship, but refuse. This is presented as the heroic and correct thing to do, even though refusing was an effective act of suicide (within the context of the movie). Not that your idea is actually in any way a bad one, but I would wager that the similarities between your proposition and what evil Hollywood villains stereotypically do is likely to increase the public perception of EA folks being cult-like (if your plan captures any press attention), which could potentially drive talent away, and discourage outsiders from cooperating with the community. All that being said, this is ultimately a rather minor concern compared to, say, the possibility of human extinction, so take the above with a grain of salt. If you do plan on going ahead with this on a large scale, I would definitely talk to some people outside the community with PR experience, so as to minimize any possible negative social effects. Good luck!!!

I'm hopeful that most people would see the difference between "rich people trying to save their own skin" and "allowing researchers who are trying to make sure humanity has a long-term future at all to continue their work", but I would be very happy to have leads on who to talk to about presenting this well.

EA is already committing to caring about AI alignment, and spending large sums of money in what would be otherwise perceived as unusual things. EA will also inevitably need to compete against others for finite sources of power* - including other charities and non-EA causes - if it must fulfill its aims better. I doubt its possible to hide this forever, even if it were desirable to do so (which idk).

*civ ark isn't that finite tbh, you can carry all sorts of research and researchers not just AI alignment work. But other sources of power are even more finite, such as being able to influence govt policy.

(x-post from EA forum)

Nice post, but if I may add: It's not just alignment research that needs to be preserved. For instance here's Linch Zhang's comment on civilisational restart manuals. Would be cool to have one coordinated megaproject on all aspects of this.

Various forms of embossing/etc on metal sheeting can also be decent, although beware the tradeoff of 'cheap metals corrode; expensive metals have a tendency to get melted down because they are expensive'.

Stainless steel is not that expensive, and pretty corrosion resistant. Although laser etched glass may be a better option. 

Stainless steel is an option. It does still corrode long-term[1]. It works fine over decade-to-century timescales for structural applications[2]; I don't know if we can trust it to retain fine details[3] over long timescales[4].

Laser etched glass is interesting, though brittle. 

  1. ^

    Said corrosion is slow, especially in proper conditions (in still dry air, no other metals around for galvanic corrosion, etc); it is not, and cannot be, non-existent.

  2. ^

    ...most of the time. Salt water destroys everything.

  3. ^

    1mm of corrosion in a 2cm-deep structural member is far less of a problem than 1mm of corrosion on 0.1mm-deep lettering.

  4. ^

    The longest study I found on atmospheric exposure of stainless steel was 10 years. Somewhat surprising considering that stainless steel has been around for ~180y at this point (1840s or so).

Footnote #5 seems as if it cuts off too soon.

Fixed, thanks.