政治的暴力は決して許されない

Nor is the threat or implication of violence. Period. Ever. No exceptions.

It is completely unacceptable. I condemn it in the strongest possible terms.

It is immoral, and also it is ineffective. It would be immoral even if it were effective. Nothing hurts your cause more.

Do not do this, and do not tolerate anyone who does.

The reason I need to say this now is that there has been at least one attempt at violence, and potentially two in quick succession, against OpenAI CEO Sam Altman.

My sympathies go out to him and I hope he is doing as okay as one could hope for.

Awful Events Amid Scary Times

Max Zeff: NEW: A suspect was arrested on Friday morning for allegedly throwing a Molotov cocktail at OpenAI CEO Sam Altman’s home. A person matching the suspect’s description was later seen making threats outside of OpenAI’s corporate HQ.

Nathan Calvin: This is beyond disturbing and awful. Whatever disagreements you have with Sam or OpenAI, this cannot be normalized or justified in any way. Everyone deserves to be able to be safe with their families at home. I feel ill and hope beyond hope this does not become a pattern.

Sam Altman wrote up his experience of the first attack here.

After that, there was a second incident.

Jonah Owen Lamb: OpenAI CEO Sam Altman’s home appears to have been the target of a second attack Sunday morning, a mere two days after a 20-year-old man allegedly threw a Molotov cocktail at the property, The Standard has learned.

The San Francisco Police Department announced (opens in new tab) the arrest of two suspects, Amanda Tom, 25, and Muhamad Tarik Hussein, 23, who were booked for negligent discharge.

Stephen Sorace Fox News (Fox News): An OpenAI spokesperson told Fox News Digital Monday morning that the incident was unrelated and had no connection to Altman, adding that there was no indication that Altman's home was being targeted.

We have no idea what motivated the second incident, or even if it was targeted at Altman. I won’t comment further on the second incident until we know more.

Nor is this confined to those who are worried about AI, the flip side is alas there too:

Gary Marcus: One investor today called for violence against me. Another lied about me, in a pretty deep and fundamental way. They are feeling the heat.

It also is not confined to the AI issue at all.

As Santi Ruiz notes, there has been a large rise in the salience of potential political violence and violence against public figures in the past few years, across the board.

That holds true for violence and threats against both Republicans and Democrats.

This requires a non-AI explanation.

Things still mostly don’t spiral into violence, the vast vast majority of even deeply angry people don’t do violence, but the rare thing is now somewhat less rare. A few years ago I would have been able to say most people definitively oppose such violence, but polls indicate this is no longer true for large portions of the public. This is terrifying.

Indeed, the scariest reaction known so far has been a comments section on Instagram (click only if you must), a place as distinct from AI and AI safety spaces of all kinds as one can get. This is The Public, as in the general public, for reasons completely unrelated to any concerns about existential risk, basically cheering this on and encouraging what would become the second attack. It seems eerily similar to the reaction of many to the assassination of the CEO of United Healthcare.

The stakes of AI are existential. As in, it is likely that all humans will die. All value in the universe may be permanently lost. Others will be driven to desperation from loss of jobs or other concerns, both real and not. The situation is only going to get more tense, and keeping things peaceful is going to require more work over time. It will be increasingly difficult to both properly convey the situation and how dire it is, and avoid encouraging threats of violence, and even actual attempts at violence.

Then on the other side are those who see untold wonders within their grasp.

This goes hand in hand with what Altman calls the ‘Shakespearean drama’ going on inside OpenAI, and between the major labs.

Most Of Those Worried About AI Do As Well As One Can On This

The vast majority of major voices in Notkilleveryonism, those worried that we might all die from AI, have been and continue to be doing exactly the right thing here, and have over many years consistently warned against and condemned all violence other than that required by the state’s enforcement of the law.

Almost all of those who are worried about AI existential risk are very much passing this test, and making their positions against violence exceedingly clear, pushing back very hard against any and all extralegal violence and extralegal threats of violence.

Demands for impossible standards here are common, where someone who did not cause the problem is attacked for not condemning the thing sufficiently loudly, or in exactly the right away. This is a common political and especially culture war tactic.

Perhaps the worst argument of all is ‘you told people never to commit or threaten violence because it is ineffective, without explicitly also saying it was immoral, therefore you would totally do it if you thought it would work, you evil person.’

They will even say ‘oh you said it was immoral, and also you said it wouldn’t work, but you didn’t explicitly say you would still condemn it even if it would work, checkmate.’

The implicit standard here, that you must explicitly note that you would act a certain way purely for what someone thinks are the right reasons or else you are guilty of doing the thing, is completely crazy, as you can see in any other context. It is the AI version of saying ‘would you still love me if I was a worm?’ and getting mad that you had to ask the question to get reassurance, as opposed to being told unprompted.

The reason why people often focus on ‘it won’t work’ is because this is the non-obvious part of the equation. With notably rare exceptions, we all agree it is immoral.

Andy Masley offers thoughts, calling for caution when describing particular people. He draws a parallel to how people talk about abortion. Here is Nate Soares at length.

This is Eliezer Yudkowsky’s latest answer on violence in general, one of many over the years trying to make similar points.

Some Who Are Worried About AI Need To Address Their Rhetoric

Almost all and vast majority are different from all.

There are notably rare exceptions, where people are at least flirting with the line, and one of these has some association to this attempt at violence, and a link to another past incident of worry about potential violence. Luckily no one has been hurt.

Speaking the truth as you see it is not a full free pass on this, nor does condemning violence unless it is clear to all that you mean it. There are some characterizations and rhetorical choices that do not explicitly call for violence, but that bring far more heat than light, and carry far more risk than they bring in benefits.

Everyone involved needs to cut that right out.

In particular, I consider the following things that need to be cut right out, and I urge everyone to do so, even if you think that the statements involved are accurate:

Calling people ‘murderers’ or ‘evil.’

Especially calling them ‘mass murderer’ or ‘child murderer.’

Various forms of ‘what did you expect.’

Various forms of the labs ‘brought this on themselves.’

Saying such violence is the ‘inevitable result’ of the labs ‘not being stopped.’

You can and should get your point across without using such words.

Also, no matter what words you are using, continuously yelling venom at those you disagree with, or telling those people they must be acting in bad faith and to curry lab favor, especially those like Dean Ball and even myself, or anyone and everyone who associates with or praises any of the AI labs at all, does not convince those people, does not convince most observers and does not help your cause.

Note, of course, that mainstream politicians, including prominent members of both parties, very often violate the above five rules on a wide variety of topics that are mostly not about AI. They, also, need to cut that right out, with of course an exception for people who are (e.g.) literally murderers as a matter of law.

Also: There are not zero times and places to say that someone does not believe the things they are saying, including telling that person to their face or in their replies. I will do that sometimes. But the bar for evidence gathered before doing this needs to be very high.

Please, everyone, accept that:

Those who say they are worried that AI will kill everyone are, with no exceptions I know about, sincerely worried AI will kill everyone.

Even if you think their arguments and reasons are stupid or motivated.

Those who say they are not worried AI will kill everyone are, most of the time, not so worried that AI will kill everyone.

Even if you think their arguments and reasons are stupid or motivated.

A bunch of people have, in good faith, concerns and opinions you disagree with.

(Dean Ball there also notes the use of the term ‘traitor.’ That one is… complicated, but yes I have made a deliberate choice to avoid it and encourage others to also do so. It is also a good example of how so many in politics, on all sides, often use such rhetoric.)

My current understanding is the first suspect was a participant of the PauseAI (Global) discord server, posting 34 messages none of which were explicit calls to violence. He was not a formal part of the organization, and participated in no formal campaigns.

We do not know how much of this is the rhetoric being used by PauseAI or others reflecting on this person, versus how much is that this is him being drawn to the server.

PauseAI has indeed unequivocally condemned this attack, which is good, and I believe those involved sincerely oppose violence and find it unacceptable, which is also good.

I think they still need to take this issue and the potential consequences of its choices on rhetoric more seriously than they have so far. Its statement here includes saying that PauseAI ‘is that peaceful path’ and avoiding extreme situations like this is exactly why we need a thriving pause movement. This is an example of the style of talk that risks inflaming the situation further without much to gain.

There is one thing that they are clearly correct about: You are not responsible for the actions of everyone who has posted on your public discord server.

I would add: This also applies to anyone who has repeated your slogans or shares your policy preferences, and it does not even mean you casually contributed at all to this person’s actions. We don’t know.

For the second attack, for now, we know actual nothing about the motivation.

But yes, if you find your rhetoric getting echoed by those who choose violence, that is a wake up call to take a hard look at your messaging strategy and whether you are doing enough to prevent such incidents, and avoid contributing to them.

Similarly, I think this statement from StopAI’s Guido Reichstadter was quite bad.

Speak The Truth Even If Your Voice Trembles

If one warns that some things are over the line or unwise to say, as I did above, one should also note what things you think are importantly not over that line.

Some rhetoric that I think is entirely acceptable and appropriate to use, if and only if you believe the statements you are making, include, as examples:

‘Gambling with humanity’s future.’

‘If [X] then [Y]’ if your conditional probability is very high (e.g. >90%), or of stating your probability estimate of [Y] given [X], including in the form of a p(doom).

Calling Mythos or something else a ‘warning shot.’

Calling Mythos or other similarly advanced AIs a ‘weapon of mass destruction.’

Most of all: To call all the act of creating minds more powerful than humans an existential threat to humanity. It obviously is one.

If you believe that If Anyone Builds It, Everyone Dies, then you should say that if anyone builds it, then everyone dies. Not moral blame. Cause and effect. Note that this is importantly different from ‘anyone who is trying to build it is a mass murderer.’

I could be convinced that I am wrong about one or more of these particular phrases. I am open to argument. But these seem very clear to me, to the point where someone challenging them should be presumed to either be in bad faith or be de facto acting from the assumption that the entire idea that creating new more powerful minds is risky is sufficiently Obvious Nonsense that the arguments are invalid.

Here is a document about how Pause AI views the situation surrounding Mythos. It lays out what they think are the key points and the important big picture narrative. It is a useful document. Do I agree with every interpretation and argument here? I very much do not. Indeed, I could use this document as a jumping off point to explain some key perspective and world model differences I have with Pause AI.

I consider the above an excellent portrayal of their good faith position on these questions, and on first reading I had no objection to any of the rhetoric.

False Accusations And False Attacks Are Also Unacceptable

There has been quite a lot of quite awful rhetoric in the other direction, both in general and in response to this situation. We should also call this out for what it is.

There are those who would use such incidents as opportunities to impose censorship, and tell people that they cannot speak the truth. They equate straightforward descriptions of the situation with dangerous calls for violence, or even attack any and all critics of AI as dangerous.

At least one person called for an end to ‘non-expert activism’ citing potential violence.

We have seen threats, taunting, deliberate misinterpretation, outright invention of statements and other bad faith towards some worried about AI, often including Eliezer Yudkowsky, including accusing people of threatening violence on the theory that of course if you believed we were all going to die you would threaten or use violence, despite the repeated clear statements to the contrary, and the obvious fact that such violence would both be immoral and ineffective.

This happened quite a bit around Eliezer’s op-ed in Time in particular, usually in highly bad faith, and this continues even now, equating calls for government to enforce rules to threats of violence, and there are a number of other past cases with similar sets of facts.

At other times, those in favor of AI accelerationism have engaged in threats of and calls for violence against those who oppose AI, on the theory that AI can cure disease, thus anyone who does anything to delay it is a murderer. The rhetoric is the same all around.

Some Examples Of Attempts To Create Broad Censorship

This is from someone at the White House, trying to equate talking about logical consequences with incitement to violence. This is a call to simply not discuss the fact that if anyone builds superintelligence, we believe that it is probable that everyone will die.

I that kind of attack completely unacceptable even from the public, and especially so from a senior official.

One asks what would happen if we applied even a far more generous version of this standard to many prominent people, including for example Elon Musk, or other people I will decline to name because I don’t need to.

Here is the Platonic form:

Shoshana Weissmann, Sloth Committee Chair: This is insane behavior. And those promoting the idea of AI ending humanity are contributing to this. It has to stop.

As in, you need to stop promoting the idea of AI ending humanity. Never mind how you present it, or whether or not your statement is true. No argument is offered on whether it is true.

This is the generalization of the position:

florence: It would appear that, according to many, one of the following are true:

1. It is a priori impossible for a new technology to be an existential threat.
2. If a new technology is an existential threat, you’re not allowed to say that.

Indeed, one of the arguments people often literally use is, and this is not a strawman:

You straightforwardly say sufficiently advanced AI might kill everyone.

But if someone did believe that, they might support using violence.

Therefore you can’t say that, or we should be able to use violence against you.

While I don’t generally try to equate different actions, I will absolutely equate implicit calls for violence in one direction to other implicit calls for violence or throwing your political enemies in jail for crimes they obviously are not responsible for, indeed for the use of free speech, in the other direction, such as this by spor or Marc Andreessen.

Nate Soares (MIRI): “even talking about the extinction-level threat is incitement towards violence”

No. High stakes don’t transform bad strategies into good ones. Let’s all counter that misapprehension wherever we find it.

michael vassar: This is probably my number one complaint about the current culture. The false dichotomy between ‘not a big deal, ignore’ and ‘crisis, panic, centralize power and remove accountability’.

That’s the same thing or worse, especially in this particular case, where the accusation is essentially ‘you want government to pass and enforce a law, we don’t like that, therefore we want the government to arrest you.’

There is also the version, which I would not equate the same way, where #3 is merely something like ‘so therefore you have a moral responsibility to not say this so plainly.’ For sufficiently mid versions, as I discuss above, one can talk price.

A variation is when someone, often an accelerationist, will say:

These people claim to be worried about AI killing everyone.

But you keep condemning violence.

Therefore, you must not care about these supposed beliefs.

Or here’s the way some of them worded it:

bone: Nice to see all the LessWrong people fold completely on their philosophy. Very good for humanity. They have no beliefs worth dying or killing for. It's nonsense from a guy who never had the balls to stand up for his words once push came to shove.

Yudkowsky stands for nothing.

bone: remember: if they actually believe all this stuff and they are unwilling to be violent, it means they are cowards, that they refuse to measure up to their own words, that they will not do what they believe needs to be done to save mankind.

they are weak, they believe in nothing

Zy: AI doomers are like “attacking key researchers in the AI race is an ineffective strategy to prevent AI doom which pales in comparison to my strategy of paying them $200 a month to fund capabilities research”

Lewis: Rare Teno L. If you actually think Sam Altman is going to genocide children then it makes sense to try to hurt him. So you need to pick one. It’s either completely insane or it’s totally sensible. Which one is it?

L3 Tweet Engineer (replying to Holly Elmore): If you’re such a good person, and stopping AI is so important, why don’t you go bomb a data center? Why waste your breath tweeting about this stuff and writing grand narratives, go make it happen.

phrygian: You've already talked about how it would be moral to nuke other countries to stop asi. The only logical reasons you have for not engaging in smaller forms of violence to stop ASI is that they aren't as effective. On a fundamental level, your views justify violence of any kind.

Ra: maybe this is just me and explains some things about me, but *personally* i would much rather be seen as a potentially dangerous radical than as a feckless and insincere grifter, especially if i believed the world was ending soon and was personally responsible for stopping that.

Trey Goff: Look do you people not realize how silly you look

“AI is going