Equilibria against outright violence seems like one of the great accomplishments of civil society. We don't let might make right, and there's an argument that things could devolve pretty quickly once violence is on the table.
I suspect that when people imagine violence on the margin, they're just assuming they do some violence and nobody else responds in kind. More realistically, violence gets violent retaliation, people stop talking or arguing (which I think there is hope for), and stuff gets a lot worse.
Asymmetric weapons and "argument gets argument, not bullet" are relevant here.
You might claim that building unsafe AGI is itself violence, and I see the case for that, but that's on a non-universally accepted set of beliefs (contrast sticking bullets in people), and one could also claim that for every day AGI is delayed, millions more die, and therefore anyone contributing to those delays is committing violence that justifies violence against them.
I'd rather stay out of worlds where things go that way. The strong deontological taboos are there for good reasons. Humans run on corrupted hardware and our civilization largely seems sane enough to say "no murder, no exceptions". Well, for private individuals. Some people do need to be stopped and we have institutions for that (police, governments, etc) that do accomplish a lot (compare places with functional law enforcement and not). And within that approach, getting everyone to agree that if you take action X that is agreed up on bad, the state violence monopoly will stop you, is a good, kinda asymmetric outcome. Hence, AI policy and government intervention, which is not a bad idea if done right.
To get a little more philosophical, I'm staunchly of the "day-to-day actions get driven by deontology and virtue ethics by and large, but the deontology and virtue ethics are justified by consequentialist reasons. And in this case, I think there's solid consequentialism backing up the deontology and taboo here, and only myopia makes it seem otherwise.
The same reason there are not constant assassination attempts in general: it's outside the Overton Window of acceptable actions. One of the many benefits of civilization is that we've all agreed not to kill each other, even when it seems strategically beneficial in the short-term, because that's what's necessary to create the world we'd like to live in long-term. Defection against this norm is harshly punished to maintain the norm.
There are probably a lot of reasons, but a big one is that it probably wouldn't help much to slow down progress toward AGI (no one individual is that important), and it might hurt efforts for safety/alignment quite a lot. The PR repercussions would be enormous; it would paint the whole safety movement as dangerous zealots. It would also make others in the AI industry tend to hate and fear the safety movement, and some of that intense emotion would spread to their attitude toward any individuals with those opinions. Those creating AI would thus become less likely to listen to safety arguments. That would hurt our odds of alignment and survival quite a lot.
The PR repercussions would be enormous
To add to this: Just Stop Oil can afford to make people angry (to an extent) and to cause gridlock on the streets of London because the debate about climate change has been going on long enough that most people who might have an influence on fossil-fuel policy have already formed a solid opinion -- and even Just Stop Oil probably cannot afford the reputational consequences of killing, e.g., Exxon executives.
As long as most of the possible decision makers have yet to form a solid opinion about whether AI research needs to be banned, we cannot afford the reputational effects of violent actions or even criminal actions. Note that the typical decision maker in, e.g., Washington, D.C., is significantly more disapproving of criminal behavior (particularly, violent criminal behavior) than most of the people you know.
Our best strategy is to do the hard work that enables us to start using state power to neutralize the AI accelerationists. We do not want to do anything hasty that would cause the power of the state (specifically the criminal justice system) to be used to neutralize us.
Making it less cool “to be AI” is an effective intervention, but crime is not a good way to effect that.
I got the distinct impression (from reading comments written by programmers) that Microsoft had a hard time hiring programmers starting in the late 1990s and persisting for a couple of decades because they were perceived by most young people as a destructive force in society. Of course, Microsoft remains a very successful enterprise, but there are a couple of factors that might make the constricting effect of making it uncool to work for an AI lab stronger than the constricting effect of making it uncool to work for Microsoft: first, it take a lot more work (i.e., acquiring skills of knowledge) to start to be able to make a technical contribution to the effort to advance AI than it takes to able to contribute to Microsoft. (Programming was easy to learn for many of the people who turned out to be good hires at Microsoft.) Second, what work is required to get good enough at programming to start to be able to contribute at Microsoft is readily transferable to other jobs that were not (and are not) considered destructive to society. In contrast, if you've spent the last 5 to 7 years in a full-time effort to become able to contribute to the AI acceleration effort, there's really no where else you can use those skills: you have to start over career-wise basically. The hope is that if enough people notice what I just said before putting in that 5 to 7 years of effort, a significant fraction of the most talented ones decide to do something else (because they don't want to invest that effort, then not be able to reap the career rewards because of their own moral qualms or because AI progress has been banned by the governments).
Some people will disagree with my assertion that people who invested a lot of time getting the technical knowledge needed to contribute to the AI acceleration effort will not be able to use that knowledge anywhere else: they think that they can switch from being capability researchers to being alignment researchers. Or they think they can switch from employment at OpenAI to employment at some other, "good" lab. I think that is probably an illusion. I think about 98% of the people calling themselves alignment researchers are (contrary to their hopes and beliefs) actually contributing to AI acceleration. And I think that all the AI labs are harmful -- at least all of them that are spending $billions on training models. I feel bad saying so in what is essentially a footnote to a comment on some other subject, but I'm too tired today to do any better.
But my point is that an effective intervention is to explain that (namely, the destructiveness of AI acceleration and the non-transferability of the technical skills used in AI acceleration) to young people before they put years of work into preparing themselves to do AI engineering or research.
Disclaimer: This is *not* a call to action in any form, I neither endorse any form of violence nor the hold the radical anti-AI views that might potentially prescribe assasinations. The question was the result of a philosophical debate I had with my roommate yesterday and we did not come to a conclusion, leading to this post here.
Tldr: There's a lot of people in the EA/RAT space alone, who strongly believe in short AI timelines, have a high p(doom) and devote their entire career to slowing down AI. Many of those people have access to money and live in California. Why has nobody even tried to kill Sam Altman or other high impact individuals who heavily contribute to speed up AI?
Underlying Beliefs:
Within EA/Rationality community there's a lot of people with relatively short AI timelines and a high p(doom). Exaggeratedly speaking they might think "we're all gonna die in 5 years". A lot of those people are highly educated in AI, think they know a lot the market hasn't priced in and yet they feel relatively powerless in impacting the AI trajectory. Many of those people strive for having a high impact, devote a significant part of their lifes to an AI safety fight, many yudkowsky-like people think is hopelessly lost. Overall the course of how and how fast AI develops seems to be strongly determined by a very small set of people - even many highly educated AI researchers in SF have negligible impact, while a few high profile people like Sam Altman concentrate a lot of public attention, decision making power and attract capital. If one was to succeed in assasinating him that would probably significantly slow down AI directly for obvious reasons. But addionally even if it failed it would set a precedent making it much less attractive to be a shining AI accelerationist. Potentially this would push some of the decision making behind closed doors, but much of the acceleration happening only works because he's such a shiny individual that attracts investors, users, companies etc. So also indirectly there could be a significant slow-down from making it less cool "to be AI". This makes me think that many of those "high impact saving the world from AI" people, who feel paralyzed because of their complete absence of impact on short AI timelines and their belief that AI kills them soon anyways would spot an opportunity to actually have radical impact and survive, if they were to think that an assasination would be realistic to achieve.
Now many of the more radical rat/EA people live in california, where Altman lives too. Many of those people have access to some funds. Weapons are relatively easily accessible in the US and even organised crime is big enough in the US that professional assassines/hitmen are available for something in the ballpark of ~50k. Those AI leaders live under a much less strict security regime than partisan top politicians like donald trump. I would find it surprising that if a few highly educaeted rationally minded people with some funds who were to organise themselves well, would not stand a chance in succeeding in assasinating Altman and a couple of other leaders in a series of strikes on one day, radically changing the way AI is being accellerated at the moment.
For most people with normal ethical beliefs like myself that would obviously an extremely evil thing to do. However I reckon that not everybody shares my beliefs and in those communites there are many strict utilitarians who say "if I could have killed Putin 3 years ago, I would have done it" - and given their beliefs about AI Altman might be in an analogous position. However I haven't seen a single article or news headline about even an attempt to do so. Why is that not happening? Are there obvious errors in my reasoning I don't see?
I want to emphasise again that I do not endorse this at all and strictly oppose any violence. This reasoning was the product of a philosophical debate I had with my roommate yesterday and we did not come to a conclusion, leading to this post here.