Theory of Change / Macrostrategy

I was thinking of creating and discussing a theory of change/ victory to pause AI. I think it would help to learn how we think are the most effective ways to achieve it so we can focus our limited resources in the best ways and realize (if we haven't already) about all the actions that have the highest chance of saving the world. I don't know much about ToC but it seems good to also being in the same page and specially redirect better money and people that don't have much preferences about what to do. Are there cons to it? To make it public?

I was thinking on basic stuff like pausing AI -> politicians agreeing to it and enforcing it -> (being informed about the risks -> lobbying) & (people being in favor of it -> outreach)...

And I was thinking for example that it would be better to outreach rather than lobbying right now given that we are far away from the summits so the next questions would be how to outreach better and how to grow the community for that.

I also just don't see politicians agreeing to a fricking pause on AI without a lot more of negative public sentiment about it and I don't know how much the polls actually represent that?? But maybe I'm wrong in either of the two.

hurt-tomato•12/22/23, 11:44 AM

I've been thinking of this, too. It's probably a good idea to make our strategy highly public.

I think my strategy is very simple:

Be honest about what the risks are and what needs to happen (this is a surprisingly rare view in the field - most often there is a belief of needing to play 4d chess, not saying what needs to happen but packaging it some some specific manipulative way)
Enable volunteers to easily join and contribute in a productive way.
Stimulate a positive, action-oriented and welcoming culture where people feel happy to be part of it.
Enable peaceful protesting, prevent violent forms of protest by channelling the energy to other things
Adapt to what is needed. The field moves quickly, so respond to changes.

The basic gist of the theory of change is also quite simple:

Get more people to understand the risks and how to mitigate them
Both bottom up support (influence politicians indirectly) and top down support (influence politicians directly) are valuable.
Protests often work as an 'in' with politicians - they get attention, which could lead to politicians reading the full proposal, learning more about the risks, or talking about what needs to happen in a one-on-one.

There are also a couple of things I'm not quite sure about:

Radical flank. I've heard people from AI governance orgs say that we are a useful radical flank. We push the overton windown by being extreme, and that makes otherwise extreme solutions appear more moderate. I don't like this frame for a couple of reasons. 1) I don't think what we're asking for is extreme. Most people support a pause, and it is just what is needed. 2) If we would accept this position / strategy, that means we should push for things that are more extreme than we really want as outcomes. That means lying. I don't like that one bit.

colossal-harlequin•12/22/23, 7:50 PM

Of course those governance folk should speak for themselves... but it seems the most extremity they might infer is that we're saying "first pause all frontier model training, then determine its safety requirements" while others prefer "first determine the safety requirements for training frontier models, then pause unsafe training".

Not much difference in assessment of timelines and risk can be sufficient to prefer one order to the other. I agree this does not seem extremely extreme.

sacred-roseOP•12/23/23, 8:35 AM

I think I don't disagree with your points and given that you didn't really comment on my points I think you don't disagree with them. I will repeat the most crucial one just in case: I think we should focus on growth that btw includes general soft support of a pause AND more strong support from people that would be willing to protest and lobby.

Which connects with the "extreme" notion that you were talking about. Because I don't really know how to interpret the polls. Do people really support a pause? Do they even fear X-risks? Then why I believe nobody talks about it? Of course I don't live at the important countries to observe that are the US and UK, but influencers aren't talking about it and that is where I get an idea of those places.

So what do the polls actually represent? What is your model about what do the people think about Joep?

sacred-roseOP•12/23/23, 8:40 AM

Maybe people are well informed about AIXR but for some reason they don't talk about it.

full-green•12/23/23, 6:45 PM

My theory about the polls is that the vast majority of the public:

Are not well-informed about AIXR
Don't think about human extinction
Are viscerally freaked out by AI a little bit
Are worried about deepfakes and fake news
Have a vague sense that powerful AI could be dangerous in the wrong hands (or unto itself)
Don't want to be replaced by machines
Don't want to lose their sense of worth / meaning in their lives
Don't want to lose their income

sacred-roseOP•12/23/23, 9:17 PM

Isn't viscerally the opposite of little bit?

sacred-roseOP•12/23/23, 9:24 PM

so they're not well informed but they fear it anyways based on the 4th and 6th point here https://pauseai.info/polls-and-surveys ? Do they vocalize it though?

Because let's say the public like the idea of a pause. Do we need more than that for the governments to implement it, right? That they support it more strongly?

sacred-roseOP•12/23/23, 9:26 PM

I mean there's a difference between someone mentioning something and people responding "yeah, that sounds fine" and going out and expressing it themselves

sacred-roseOP•12/23/23, 9:30 PM

For what I understand at least you don't need them to feel strongly to the point of determining what people vote.

Ssacred-rose Isn't viscerally the opposite of little bit?

full-green•12/24/23, 1:03 AM

I should have said "instinctive".

Ssacred-rose so they're not well informed but they fear it anyways based on the 4th and 6th p...

hurt-tomato•12/24/23, 11:41 AM

I think @Nathan Metzger is right in that people are viscerally / instinctively freaked out a bit, but haven't really developed a serious worry about AI takeover / xrisk. For me personally, it took years to internalise xrisk. When I did, I could no longer sit still.

I think people don't necessarily have very coherent beliefs about all of this, I think most go on with their day after hearing about scientists warning about the end of the world. Perhaps the feeling of 'not my problem' is dominant. What can I do? I'm not an AI scientists. I'm not a politician.

Some people have formed strong opinions / worries about specific risks. The job risk is pretty common, as is the bio risk. People are also worried about privacy - which I'm not quite sure what that means for most people. I think it's our job to make sure every worry is represented at PauseAI. We should highlight the breadth of risks from AI, including misuse risks.

My intuition is that once you're internalising one risk, you become more open to considering all others.

sacred-roseOP•12/25/23, 12:13 AM

In that case PauseAI should be the answer to a lot of people fears and if we have limited time or space in some places to warn about risks vs say that a pause is possible and the most robust solution but needs their help we should focus on the latter

sacred-roseOP•12/25/23, 12:26 AM

So, following my logic that I think the most important things to do right now is growing. Outreach of PAI + making the people who support a pause in theory to get involved and take action.

sacred-roseOP•12/25/23, 12:38 AM

For the first thing I think the most cost effective thing is #Influencer outreach (e-mails) if its possible.

For the second thing, I'm not sure, I think luckily it's a reinforcing lack of hope loop that if PAI starts to grow we could have the opposite? Something close to a network effect. I have some ideas to better the Discord still and maybe we should do some surveys to know why people join and what are their obstacles if they want to help.

Anyways what would be good metrics to measure the movement? People in protests?

sacred-roseOP•12/25/23, 12:40 AM

And e-mails and comments on #projects-

sacred-roseOP•12/29/23, 8:35 PM

actually I've been thinking that recruiting (sending DMs) should be really useful to know what do people want/ need to volunteer more and I will start doing it next week

Ssacred-rose actually I've been thinking that recruiting (sending DMs) should be really usefu...

hurt-tomato•12/30/23, 1:14 AM

yay that's great!

sacred-roseOP•1/12/24, 3:50 AM

sacred-roseOP•1/12/24, 3:55 AM

an early draft of what it can someday be a Theory of Victory we can share for funding, to the general public to make people understand better what we are about and a possible plan that they can take part on, and in the discord as a reminder on what's the stuff different people can work on and what's their contribution to saving of the world :)

sacred-roseOP•1/12/24, 3:56 AM

I think we could technically add other ways in which advocating for a pause without getting it could be useful

sacred-roseOP•1/12/24, 3:56 AM

Please critique! and tell me about stuff I could've missed or forgot.

sacred-roseOP•1/12/24, 3:58 AM

I mean, every part could be simplified or expand it on but I don't know

sacred-roseOP•1/12/24, 4:04 AM

Ideally, we should end up having objectives, subobjectives, target audiences with their respective messaging strategies, way to measure progress, and maybe prioritized actions

sacred-roseOP•1/12/24, 4:09 AM

E.g. in the near future, reaching out to people unemployed because of AI could be pretty productive

sacred-roseOP•1/12/24, 4:12 AM

if you wanna try doing changes

Theory_of_Victory.excalidraw209.66KB

sacred-roseOP•1/12/24, 4:30 AM

Although to be honest, the cooperation with some AI Ethics concerns are complicated when it comes to say until when we want the pause, if we really want to ally with them we maybe need to start adding other requirements to unpause like "putting windfall clauses for the affected by AI replacements"

sacred-roseOP•1/12/24, 5:35 PM

That sounds like a bad idea now that I think about it. Too complicated and kinda dishonest. More considering that a pause can already do stuff for unemployment: slowing it down and allowing a smoother transition

full-green•1/12/24, 6:36 PM

Hmm, yeah, I suppose there is a difference between most people who support a pause also supporting popular AI ethics proposals, and Pause AI explicitly including those proposals in their asks. There's always room to pivot if we actually get our primary ask.

wet-aqua•1/23/24, 8:21 PM

@Pato Vercesi I agree that it's key to have a good model of the AI situation (timelines, labs, governments, PAI, e/accers) in order to make good decisions about how to intervene toward particular outcomes.

moderate-tomato•1/24/24, 11:38 PM

h/t to Greg Colbourn

Jaan Tallinn’s priorities for 2024:
https://jaan.online/philanthropy/

moderate-tomato•1/24/24, 11:39 PM

Includes Global Off-Switches

sacred-roseOP•3/1/24, 11:44 PM

I better my draft months ago but it seems we are not looking at that level of details at least for now.

sacred-roseOP•3/1/24, 11:44 PM

https://github.com/joepio/pauseai/blob/main/src/posts/theory-of-change.md?plain=1

GitHub

pauseai/src/posts/theory-of-change.md at main · joepio/pauseai

Website for PauseAI.info. Contribute to joepio/pauseai development by creating an account on GitHub.

sacred-roseOP•3/1/24, 11:47 PM

First of all. What is the idea? are we just gonna have a couple of sections about the "last steps"? should we develop it more but just have that in the page?

sacred-roseOP•3/1/24, 11:51 PM

I don't know much about Theory of Change but seems something incredibly cool. There seems to be lots of model on how to do it, is there a problem if someone without "expertise" does it?
Maybe there's other frameworks which are better to monitor progress and correct mistakes to plans?

Ssacred-rose https://github.com/joepio/pauseai/blob/main/src/posts/theory-of-change.md?plain=...

sacred-roseOP•3/1/24, 11:54 PM

With respect to what Joep wrote, I think growing the movement is actually one step back from Lobbying and Protesting. Alongside Getting funding.

sacred-roseOP•3/1/24, 11:57 PM

And Informing the public would be like a step before Growing the movement but I would put it inside of it

sacred-roseOP•3/6/24, 7:38 AM

https://www.socialchangelab.org/_files/ugd/503ba4_f2b72008b8fd47d087b019a620533236.pdf this and other publications from this EA (adj?) org could be quite useful to read @Joep Meindertsma

Ssacred-rose https://www.socialchangelab.org/_files/ugd/503ba4_f2b72008b8fd47d087b019a6205332...

hurt-tomato•3/6/24, 8:02 AM

I read that one (think I link to it from some places of the PauseAI site) but do you have other related publications?

sacred-roseOP•3/6/24, 8:18 AM

from https://www.socialchangelab.org/research the other one which seems useful is https://www.socialchangelab.org/_files/ugd/503ba4_6b9e237eafca404c9db67b5a41d08e8b.pdf

sacred-roseOP•3/6/24, 8:18 AM

but was published before, so maybe the findings were included in that one

sacred-roseOP•3/6/24, 8:22 AM

and this one https://www.apollosurveys.org/social-change-and-protests/

Ssacred-rose and this one https://www.apollosurveys.org/social-change-and-protests/

hurt-tomato•3/6/24, 8:35 AM

Interesting graphic from that study about effects of disruptive tactics.

Anti-AI stuff is probably low-awareness, bug high-support

Almost 60% experts believe disruptive tactics are effective, 30 % thinks they are ineffective.

X-risk risks are probably lower in support, which is more 50/50.

awake-maroon•3/6/24, 10:25 AM

Do you think that other risks / "anti-ai stuff" can work to get people to later consider x-risk? Because ultimately the other risks probably aren’t going to cut it if we want more people to talk to politicians about a pause

Aawake-maroon Do you think that other risks / "anti-ai stuff" can work to get people to later ...

hurt-tomato•3/6/24, 10:29 AM

I think pausing would also make sense even if there was no x-risk. Bioweapon and cyberweapon creation are both not x-risks, but definitely scary and catastrophic. But even without catastrophic risks, we'll have a huge set of issues to deal with in making our democracy function with superhuman AI models in them. How will we deal with inequality when most jobs are automated away? How do voters inform themselves when the web is becoming bloated with AI generated stuff designed to manipulate them? We've got a shit ton of issues to deal with - and pausing is the right call even if if we had only one of these issues.

awake-maroon•3/6/24, 10:35 AM

Hm, the problem is that in my mind the global moratorium we are calling for is very much a drastic measure against a drastic problem and politicians are forced to always pick the mildest solution to a problem. I can see bioweapons and cyberweapons being on the correct scale but I wouldn’t be surprised if someone finds a milder solution there ("only let the trusted good guys build it and don’t deploy it publicly" or something like that)

awake-maroon•3/6/24, 10:37 AM

The interview with Sarah in one of the Dutch articles about pauseai gave me some hope that worrying about other risks and then coming across xrisk can work even if it’s just anecdotal

technical-ivory•3/6/24, 10:38 AM

So what do you think we should do to convince more people? Talk about other risks? I really want to be useful but I don't really know how

Aawake-maroon Hm, the problem is that in my mind the global moratorium we are calling for is v...

awake-maroon•3/6/24, 1:03 PM

I would really like to hear other opinions on this

Aawake-maroon I would really like to hear other opinions on this

full-green•3/6/24, 7:15 PM

I think the ideal would be to partner with other orgs that are focused on other risks or that have a strategy that focuses on other risks (e.g. Control AI).

The hard part would be the convincing other orgs that partnering with PauseAI makes sense for them even if they aren't sure about x-risk. That might be feasible, though, if we focus on the policy ask and supply arguments from popular support in the various surveys we're aware of.

Theory of Change / Macrostrategy

Similar Threads

Similar Threads

Similar Threads