This is a special post for quick takes by Holly Elmore ⏸️ 🔸. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
There's a serious courage problem in Effective Altruism. I am so profoundly disappointed in this community. It's not for having a different theory of change-- it's for the fear I see in people's eyes when considering going against AI companies or losing "legitimacy" by not being associated with them. The squeamishness I see when considering talking about AI danger in a way people can understand, and the fear of losing face within the inner circle. A lot of you value being part of a tech elite more than you do what happens to the world. Full stop. And it does bother me that you have this mutual pact to think of yourselves as good for your corrupt relationships with the industry that's most likely to get us all killed.
Big +1 to the gist of what you are saying here re: courage. It's something I've struggled with a lot myself, especially while I was at OpenAI. The companies have successfully made it cool to race forward and uncool to say e.g. that there's a good chance doing so will kill everyone. And peer pressure, financial incentives, status incentives, etc. really do have strong effects, including on people who think they are unaffected.
As an EA who is profoundly disappointed at the level of Bay Area EA collaboration with the labs, I must agree. As a technical researcher, I work with them when it improves safety, but I'm a big proponent of Dan Hendrycks when he says that regulation shouldn't be pleasing for companies since they are the antagonists to any regulation in this field.
And every time I meet policy professionals in the field, they are infinitely scared to ever say extinction risk, pause, or anything like this (happy to hear retorts to this from other insiders), which is incredibly critical. In that sense, it becomes a power-seeking move, but once you are in power, if you've never had an opinion on your way to the top, no one will listen to you.
Very relevant to mention that there are definitely some EAs that are honest and pragmatic, but it's too rare.
Looking forward to be proven wrong but this seems like a profoundly misguided strategy.
I agree with this problem, that there are many folk who lack courage, particularly within AI governance, but I don't really see these kinds of comments as being that likely to change things.
There are times when it can make sense to shout at people, but it normally works better when there's a clear and realistic path that's been sketched out[1]. And when this isn't the case, it tends to simply create strife for very little gain.
- ^
Someone really needs to write an article along the lines of "The case for being bold".
This topic is painful for me because I opened my heart to EA as fellow travelers who could share my moral burden. And I feel betrayed by my former community when they let themselves be charmed and intimidated by the AI industry. And then turn all their rhetorical tricks against me not to listen to me ("You're not being nice and/or smart enough, so it's actually you who has the problem.").
I realize now the rationalists were always cowards who wanted the Singularity and to imagine colonizing Galaxies more than to protect people. But I can't quite accept EA's failure here. I'm like a spurned lover who can't believe the love wasn't real. I thought, at its core, EA really got it about protecting beings.
Have you seen any polls on this? I would guess that the majority of EAs and rationalists would not support a pause (because they think it would increase overall risk, at least at this point, e.g. Carl Shulman), but they would also generally not be supportive of what the labs are doing (racing, opposing regulations, etc).
That’s not what Carl Shulman said, and the fact that people want to take it that way is telling. He messaged me recently to clarify that he meant unilateral pauses would be bad, something I still kind of disagree with but which isn’t something PauseAI advocates, and he said it way at the beginning of Pause talk. EAs just don’t want to arrest the tech momentum bc they see themselves as technocracy elite, not as humble grassroots organizers. They are disappointed at the chance we have to rally the public and want to find some way they don’t have to contribute to it.
People who identify as EAs in other countries might not be supportive of the AI companies, but they aren’t the ones on the ground in the Bay Area and DC that are letting me down so much. They aren’t the ones working for Anthropic or sitting on their fake board, listening to Dario’s claptrap that justifies what they want to do and believe anyway. They aren’t the ones denying my grant applications bc protests aren’t to their tastes. They aren’t the ones terrified of not having the vision to go for the Singularity, of being seen as “Luddites” for opposing a dangerous and recklessly pursued technology. Frankly they aren’t the influential ones.
Though Carl said that an unilateral pause would be riskier, I'm pretty sure he is not supporting a universal pause now. He said "To the extent you have a willingness to do a pause, it’s going to be much more impactful later on. And even worse, it’s possible that a pause, especially a voluntary pause, then is disproportionately giving up the opportunity to do pauses at that later stage when things are more important....Now, I might have a different view if we were talking about a binding international agreement that all the great powers were behind. That seems much more suitable. And I’m enthusiastic about measures like the recent US executive order, which requires reporting of information about the training of new powerful models to the government, and provides the opportunity to see what’s happening and then intervene with regulation as evidence of more imminent dangers appear. Those seem like things that are not giving up the pace of AI progress in a significant way, or compromising the ability to do things later, including a later pause...Why didn’t I sign the pause AI letter for a six-month pause around now?
But in terms of expending political capital or what asks would I have of policymakers, indeed, this is going to be quite far down the list, because its political costs and downsides are relatively large for the amount of benefit — or harm. At the object level, when I think it’s probably bad on the merits, it doesn’t arise. But if it were beneficial, I think that the benefit would be smaller than other moves that are possible — like intense work on alignment, like getting the ability of governments to supervise and at least limit disastrous corner-cutting in a race between private companies: that’s something that is much more clearly in the interest of governments that want to be able to steer where this thing is going. And yeah, the space of overlap of things that help to avoid risks of things like AI coups, AI misinformation, or use in bioterrorism, there are just any number of things that we are not currently doing that are helpful on multiple perspectives — and that are, I think, more helpful to pursue at the margin than an early pause."
So he says he might be supportive of a universal pause, but it sounds like he would rather have it later than now.
They aren’t the ones terrified of not having the vision to go for the Singularity, of being seen as “Luddites” for opposing a dangerous and recklessly pursued technology. Frankly they aren’t the influential ones.
I see where you are coming from, but it think it would be more accurate to say that you are disappointed in (or potentially even betrayed by[1]) the minority of EAs who are accelerationists, rather than characterizing it as being betrayed by the community as a whole (which is not accelerationist).
- ^
Though I think this is too harsh as early thinking in AI Safety included Bostrom's differential technological development and MIRI's seed (safe) AI, the former of which is similar to people trying to shape Anthropic's work and the latter of which could be characterized as accelerationist.
To get a pause at any time you have to start asking now. It’s totally academic to ask about when exactly to pause and it’s not robust to try to wait until the last possible minute. Anyone taking pause advocacy seriously realizes this pretty quickly.
But honestly all I hear are excuses. You wouldn’t want to help me if Carl said it was the right thing to do or you’d have already realized what I said yourself. You wouldn’t be waiting for Carl’s permission or anyone else’s. What you’re looking for is permission to stay on this corrupt be-the-problem strategy and it shows.
To get a pause at any time you have to start asking now. It’s totally academic to ask about when exactly to pause and it’s not robust to try to wait until the last possible minute. Anyone taking pause advocacy seriously realizes this pretty quickly.
As Carl says, society may only get one shot at a pause. So if we got it now, and not when we have a 10x speed up in AI development because of AI, I think that would be worse. It could certainly make sense now to build the field and to draft legislation. But it's also possible to advocate for pausing when some threshold or trigger is hit, and not now. It's also possible that advocating for an early pause burns bridges with people who might have supported a pause later.
What you’re looking for is permission to stay on this corrupt be-the-problem strategy and it shows.
I personally still have significant uncertainty as to the best course of action, and I understand you are under a lot of stress, but I don't think this characterization is an effective way of shifting people towards your point of view.
As Carl says, society may only get one shot at a pause. So if we got it now, and not when we have a 10x speed up in AI development because of AI, I think that would be worse. It could certainly make sense now to build the field and to draft legislation. But it's also possible to advocate for pausing when some threshold or trigger is hit, and not now. It's also possible that advocating for an early pause burns bridges with people who might have supported a pause later.
This is so out of touch with the realities of opinion change. It sounds smart and it lets EAs and rationalists keep doing what they’re doing, which is why people repeat it. This claim that we would only get one shot at a pause is asinine— pause would become more popular as an option the more people were familiar with it. It’s only the AI industry and EA that do not like the idea of pausing and pretend like they’re gonna withdraw support that we actually never had if we do something they don’t like.
The main thing we can do as a movement is gain popular support by talking about the message. There is no reliable way to “time” asks. None of that makes any sense. Honestly, most people who give this argument are industry apologists who just want you feel out of your league if you do anything against their interests. Hardware overhang was the same shit.
It should hurt to work with AI companies. You should be looking for excuses not to even if you think there’s an important reason to maintain the relationship. But instead it’s always the other way around because working with corrupt tech elites who “joke” about being evil is ego-syntonic to you. That’s an extremely serious problem.
I ran a successful protest at OpenAI yesterday. Before the night was over, Mikhail Samin, who attended the protest, sent me a document to review that accused me of what sounds like a bait and switch and deceptive practices because I made an error in my original press release (which got copied as a description on other materials) and apparently didn't address it to his satisfaction because I didn't change the theme of the event more radically or cancel it.
My error: I made the stupidest kind of mistake when writing the press release weeks before the event. The event was planned as a generic OpenAI protest a ~month and half ahead of time. Then the story about the mysteriously revised usage policy and subsequent Pentagon contract arose and we decided to make rolling it back the "small ask" of this protest, which is usually a news peg and goes in media outreach materials like the press release. (The "big ask" is always "Pause AI" and that's all that most onlookers will ever know about the messaging.) I quoted the OpenAI charter early on when drafting it, and then, in a kind of word mistake that is unfortunately common for me, started using the word “charter” for both the actual charter document and the usage policy document. It was unfortunately a semantic mistake, so proofreaders didn’t catch it. I also subsequently did this verbally in several places. I even kind of convinced myself from hearing my own mistaken language that OpenAI had violated a much more serious boundary– their actual guiding document– than they had. Making this kind of mistake is unfortunately a characteristic sort of error for me and making it in this kind of situation is one of my worst fears.
How I handled it: I was horrified when I discovered the mistake because it conveyed a significantly different meaning than the true story, and, were it intentional, could have slandered OpenAI. I spent hours trying to track down every place I had said it and people who may have repeated it so it could be corrected. Where I corrected "charter" to "usage policy", I left an asterisk explaining that an earlier version of the document had said "charter" in error. I told the protesters in the protester group chat (some protesters just see flyers or public events, so I couldn't reach them all) right away about my mistake, which is when Mikhail heard about it, and explained changing the usage policy is a lot less bad than changing the charter, but the protest was still on, as it had been before the military story arose as the “small ask”. I have a lot of volunteer help, but I'm still a one woman show on stuff like media communications. I resolved to have my first employee go over all communications, not just volunteer proofreaders, so that someone super familiar with what we are doing can catch brainfart content errors that my brain was somehow blind to.
So, Mikhail seems to think I should have done more or not kept the no-military small ask. He's going to publish something that really hurt my feelings because it reads as an accusation of lying or manipulation and calls for EA community level "mechanisms" to make sure that "unilateral action" (i.e. protests where I had to correct the description) can't be taken because I am "a high status EA".
This is silly. I made a very unfortunate mistake that I feel terrible about and tried really hard to fix. That's all. To be clear, PauseAI US is not an EA org and it wouldn't submit to EA community mechanisms because that would not be appropriate. Our programming does not need EA approval and is not seeking it. I failed in my own personal standards of accuracy, the standards I will hold PauseAI US to, and I will not be offering any kind of EA guarantees on what PauseAI does. Just because I am an EA doesn't sign my organization up for EA norms or make me especially accountable to Mikhail. I'm particularly done with Mikhail, in fact, after spending hours assisting him with his projects and trying to show him a good time when he visits Berkeley and answering to his endless nitpicks on Messenger and getting what feels like zero charity in return. During what should have been a celebration for a successful event, I was sobbing in the bathroom at the accusation that I (at least somewhat deliberately-- not sure how exactly he would characterize his claim) misled the press and the EA community.
I made suggestions on his document and told him he can post it, so you may see it soon. I think it's inflammatory and it will create pointless drama and he should know better, but I also know he would make a big deal and say I was suppressing the truth if I told him not to post it. I think coming at me with accusations of bad faith and loaded words like "deceptive" and "misleading" is shitty and I really do not want to be part of my own special struggle session on this forum. It wounds me because I feel the weight of what I'm trying to do, and my mistake scared me that I could stumble and cause harm. It's a lot of stress to carry this mission forward against intense disagreement, and carrying it forward in the face of personal failure is even tougher for me. But I also know I handled my error honorably and did my best.
See my post here.
because I made an error in my original press release
This was an honest mistake that you corrected. It is not at all what I'm talking about in the post. I want the community to pay attention to the final messaging being deceptive about the nature of OpenAI's relationship with the Pentagon. The TL;DR of my post:
Good-hearted EAs lack the mechanisms to not output information that can mislead people. They organised a protest, with the messaging centred on OpenAI changing their documents to "work with the Pentagon", while OpenAI only collaborates with DARPA on open-source cybersecurity tools and is in talks with the Pentagon about veteran suicide prevention. Many participants of the protest weren’t aware of this; the protest announcement and the press release did not mention this. People were misled into thinking OpenAI is working on military applications of AI. OpenAI still prohibits the use of their services to "harm people, develop weapons, for communications surveillance, or to injure others or destroy property". If OpenAI wanted to have a contract with the Pentagon to work on something bad, they wouldn't have needed to change the usage policies of their publicly available services and could've simply provided any services through separate agreements. The community should notice a failure mode and implement something that would prevent unilateral decisions with bad consequences or noticeable violations of deontology.
I am extremely saddened by the emotional impact this has on you. I did not wish that to happen and was surprised and confused by your reaction. Unfortunately, it seems that you still don't understand the issue I'm pointing at; it is not the "charter" being in the original announcement, it's the final wording misleading people.
The draft you sent me opened with how people were misled about the "charter" and alleged that I didn't change the protest enough after fixing that mistake. I think you're just very unclear with your criticism (and what I understand I simply disagree with, as I did when we spoke about this before the protest) while throwing around loaded terms like "deceptive", "misleading", and "deontologically bad" that will give a very untrue impression of me.
Hey Holly, I hope you're doing ok. I think the Bay Area atmosphere might be particular unhealthy and tough around this issue atm, and I'm sorry for that. For what it's worth, you've always seemed like someone who has integrity to me.[1]
Maybe it's because I'm not in the thick of Bay Culture or super focused on AI x-risk, but I don't quite see why Mikhail reacted so strongly (especially the language around deception? Or the suggestions to have the EA Community police Pause AI??) to this mistake. I also know you're incredibly committed to Pause AI, so I hope you don't think what I'm going to say is insensitive, but I even some of your own language here is a bit storm-in-a-teacup?
The mix-up itself was a mistake sure, but every mistake isn't a failure. You clearly went out of your way to make sure that initial incorrect impression was corrected. I don't really see how that could meet a legal slander bar, and I think many people will find OpenAI reneging on a policy to work with the Pentagon highly concerning whether or not it's the charter.
I don’t really want to have a discussed about California defamation law. Mainly, I just wanted to reach out and offer some support, and say that from my perspective, it doesn't look as bad as it might feel to you right now.
Thanks. I don't think I feel too bad about the mistake or myself . I know I didn't do it on purpose and wasn't negligent (I had informed proofreaders and commenters, none of whom caught it, either), and I know I sincerely tried everything to correct it. But it was really scary.
Mikhail has this abstruse criticism he is now insisting I don't truly understand, and I'm pretty sure people reading his post will not understand it, either, instead taking away the message that I ~lied or otherwise made a "deontological" violation, as I did when I read it.
It would be really great if EAs didn't take out their dismay over SBF's fraud on each other and didn't try to tear down EA as an institution. I am wrecked over what happened with FTX, too, and of course it was major violation of community trust that we all have to process. But you're not going to purify yourself by tearing down everyone else's efforts or get out ahead of the next scandal by making other EA orgs sound shady. EA will survive this whether you calm down or not but there's no reason to create strife and division over Sam et al.'s crimes when we could be coming together, healing, and growing stronger.
We all know EAs and rationalists are anxious about getting involved in politics because of the motivated reasoning and soldier mindset that it takes to succeed there (https://www.lesswrong.com/posts/9weLK2AJ9JEt2Tt8f/politics-is-the-mind-killer).
Would it work to have a stronger distinction in our minds between discourse, which should stay pure from politics, and interventions, which can include, e.g. seeking a political office or advocating for a ballot measure?
Since EA political candidacies are happening whether we all agree or not, maybe we should take measures to insulate the two. I like the "discourse v intervention" frame as a tool for doing that, either as a conversational signpost or possibly to silo conversations entirely. Maybe people involved in political campaigns should have to recuse themselves from meta discourse?
Relatedly, I'm a bit worried that EA involvement in politics may lead to an increased tendency for reputational concerns to swamp object-level arguments in many EA discussions; and for an increasing number of claims and arguments to become taboo. I think there's already such a tendency, and involvement in politics could make it worse.
What's so weird to me about this is that EA has the clout it does today because of these frank discussions. Why shouldn't we keep doing that?
I'm in favor of not sharing infohazards but that's about the extent of reputation management I endorse-- and I think that leads to a good reputation for EA as honest!
What's so weird to me about this is that EA has the clout it does today because of these frank discussions. Why shouldn't we keep doing that?
I think the standard thing for many orgs and cultures to start off open and transparent and move towards closedness and insularity. There are good object-level reasons for the former, and good object-level reasons for the latter, but taken as a whole, it might just better be viewed as a lifecycle thing rather than one of principled arguments.
Open Phil is an unusually transparent and well-documented example in my mind (though perhaps this is changing again in 2022)
I can see good reasons for individual orgs to do that, but way fewer for EA writ large to do this. I'm with Rob Bensinger on this.
Have you seen any polls on this? I would guess that the majority of EAs and rationalists would not support a pause (because they think it would increase overall risk, at least at this point, e.g. Carl Shulman), but they would also generally not be supportive of what the labs are doing (racing, opposing regulations, etc).