My guest, John Alioto, is an independent AI engineer with a computer science degree from UC Berkeley and 25 years building real-world AI systems at companies like Microsoft and Google.
In the wake of an attack on Sam Altman’s property, John Alioto came on the show to argue that Yudkowsky’s words are violent rhetoric that helped create this moment. Since I completely disagree with that characterization, we had a lot of fuel for a passionate debate.
For the record, here’s my position on why AI doomers are NOT “calling for violence”:
Are we acting like we actually think there’s an urgent extinction risk? Yes.
Are we calling for lawless violence? Absolutely not. At least not me, or the leaders of the movement, or anyone I’ve ever personally interacted with.
Are we calling for violence as a last resort if a government policy has been established and then egregiously violated? Yes… but that’s just standard for any governance proposal! A proposal for a strictly enforced treaty isn’t a call for violence — it’s a call for doing everything we can to make sure no one breaks the treaty, with zero violence, unless rogue actors decide to break the treaty and bring the consequences on themselves.
Thanks to John for having an extremely respectful and good-faith debate on this heated subject.
Links
John P. Alioto on X (Twitter) — https://x.com/jpalioto
Eliezer Yudkowsky, “Pausing AI Developments Isn’t Enough. We Need to Shut It All Down” (Time Magazine, March 2023) — https://time.com/6266923/ai-eliezer-yudkowsky-open-letter-not-enough/
Sarah Haider’s view on doomer beliefs (tweet, April 2026) —
Timestamps
00:00:00 — Cold Open
00:00:37 — Introducing John Alioto
00:03:02 — Setting the Stage: Recent Acts of Violence & AI Discourse
00:05:53 — Eliezer Yudkowsky’s 2023 TIME Article
00:11:16 — John’s Two-Part Argument
00:14:37 — Conditional on High P(Doom), Is Eliezer’s Policy Bad?
00:17:46 — Be Like Carl Sagan — Win in the Arena of Ideas
00:21:12 — No Carve-Outs for Non-Signatories
00:26:15 — Hypothetical: What If the UN Voted for a Treaty?
00:30:46 — What’s the Correct Interpretation of Eliezer’s TIME Article?
00:32:42 — Liron’s Interpretation: Same Structure as Any International Law Proposal
00:42:23 — What Should Eliezer Have Written? “Airstrikes” vs “Strong Deterrent”
00:49:57 — How John Would Rewrite the TIME Piece
00:50:54 — Carve-Outs: Allies, Civilians, Consequences
00:52:52 — Closing Arguments
00:56:27 — Last Q: Does High P(Doom) Imply Violence?
00:59:40 — Closing Thoughts
Transcript
Cold Open
Liron Shapira 00:00:00
This whole debate that I was hoping to have is just I claim that I’ve never had a call for violence. The policy proposal that I have, which does include a branch of the possibility tree where a violator of the international democratically voted on treaty—
John P. Alioto 00:00:13
Yes.
Liron 00:00:13
—a violator does get enforced, up to and including airstrikes. When I make that claim, I just wanted to come argue with you that I wasn’t calling for violence as a last resort.
John 00:00:22
You’re restating his position, sir. Sir, with all due respect, you’re restating his position incorrectly.
Introducing John Alioto
Liron 00:00:37
Welcome to Doom Debates. My guest today is John Alioto. He has a bachelor’s in computer science from UC Berkeley, same as myself. He is an independent AI engineer who’s been building real-world AI systems for twenty-five years, and he’s worked with major companies like Microsoft and Google.
Let me explain the context that led to me inviting John onto the show. As you’ve probably heard, a few days ago, there were some very tragic events at Sam Altman’s house. Lawless violence, including a Molotov cocktail thrown at his house in San Francisco when his family was inside, and then the next day, a shooting at or near his house. I don’t have the full details. I think they’re still being investigated.
And so right off the bat, a lot of us people who are concerned with AI existential risk, our first reaction was, “Oh no, this is horrible. This is not what we wanted. We didn’t want lawless violence. This is a tragic act. There’s no excuse for this act. There’s no silver lining to this act. It’s very bad.” That was my position.
And then, after communicating that and having a lot of soul searching — why did this happen? Is there anything that we did as people who raise awareness of existential risk that should place blame on us? What can we learn from this? After we were doing that kind of soul searching, we noticed that there was a lot of argument happening on social media where a lot of people were blaming, like somehow these events are actually connected.
And to be fair, there was some fuel added to the fire when it turned out that the Molotov cocktail thrower was apparently a member of Pause AI on Discord and joined briefly and left a couple dozen messages, and luckily didn’t get agreement there. But that was treated as a smoking gun where people as high up as Marc Andreessen were tweeting something like, “Aha, checks out.” Stuff like that.
And so it’s been a very intense week, very heated week for good reason. Very high stakes. A lot of heat being generated in this discussion. And I actually don’t count my guest John as part of the heat. But what happened was that I tweeted, “Hey, who wants to come and debate me and take the position that people like myself who are worried about AI extinction have in any way, shape, or form made a call for violence?” So that is the context for our discussion.
I wanna thank John for stepping into this fraught space and being willing to voice his opinion, even though there are just risks in even participating in the conversation. John Alioto, welcome to Doom Debates.
John 00:02:57
Sir, thank you very much for having me. I really appreciate it. I’m excited to meet you and get to know you.
Setting the Stage: Recent Acts of Violence & AI Discourse
Liron 00:03:02
Likewise, man. Okay, so I think maybe it’s good to just start with some points of consensus. I strongly believe that both me and you are people who are clearly in the mainstream of morality and would respect each other on that front. Do you get that sense?
John 00:03:18
Absolutely.
Liron 00:03:20
So even though we’re gonna have a somewhat intense disagreement on who said what and what this all means and who’s to blame, I think we have some significant disagreements there. I think a good starting point is, neither of us think lawless violence is good in any context, correct?
John 00:03:35
Absolutely agree, hundred percent. Having you culpable for the bad acts of others or the criminal acts of others makes no sense. You didn’t do it. You’re not involved in it. You disavow it. I do too. This is not about blame, at least from my perspective. This is about understanding — you understanding me and me understanding you. That’s why I wanted to come here, so I could better understand you.
Liron 00:03:57
Awesome. Yeah, likewise. So this debate is gonna be productive, high-quality debate where two people who are both rational, mutually respectable people are going to compare their analysis. Maybe one of us will swing the other’s position a little bit. Who knows?
Worst case, the viewers will at least see what these two people believe, so they better understand the position. Because I feel like a big problem in the discourse right now is a lot of people really don’t understand at least my position. Do you feel like people understand your position?
John 00:04:22
I agree with you a hundred percent. It’s why I’m excited to be here. I don’t get the opportunity that often to speak to someone as intelligent as you, who’s got these positions that I believe you hold very honestly and want to speak about them.
I’m interested in understanding that and understanding how you think about it because I just don’t get the opportunity to speak about that often. I do think if there’s a “my side” — I hate to think about it that way — but I do think a lot is misunderstood in AI in general in the community, and there’s various different reasons for that and causes for that. I do think there’s a lot of misunderstanding in general.
Liron 00:05:07
Yep. All right. That’s fair enough. That’s what we’re here to discuss.
Eliezer Yudkowsky’s 2023 TIME Article
Liron 00:05:10
Let’s rewind the clock back to 2023. On March 14th, 2023, OpenAI officially launched GPT-4 in the wake of GPT-3 coming only a few months before and GPT-3.5. So it was this rapid escalation that made a lot of us take notice — “Oh, holy crap. I think the singularity’s starting.” That was a very poignant moment, and I would argue today is still a poignant moment.
And then a couple weeks after that date, on March 29th, 2023, a very well-known article came out in Time magazine, a contributed article by Eliezer Yudkowsky. And the title was “Pausing AI Developments Isn’t Enough: We Need to Shut It All Down,” basically making the public case for stopping frontier AI capabilities, which is related to the Pause AI movement. Although Eliezer Yudkowsky said he doesn’t believe in Pause AI, he’s not a member of the Pause AI movement. He just thinks stopping AI is good, so I wanted to draw that separation.
Okay, so this article came out in Time. And everything that people remember about that article, I think comes down to one quote basically — the airstrikes on data centers quote. So I should read it in Eliezer Yudkowsky’s exact words, so you guys have the firsthand source here, and if you want a link to the article, it’s gonna be here in the show notes of this episode.
So here’s the quote from Eliezer Yudkowsky: “Shut down all the large GPU clusters, the large computer farms where the most powerful AIs are refined. Shut down all the large training runs. Put a ceiling on how much computing power anyone is allowed to use in training an AI system, and move it downward over the coming years to compensate for more efficient training algorithms. No exceptions for governments and militaries. Make immediate multinational agreements to prevent the prohibited activities from moving elsewhere. Track all GPUs sold. If intelligence says a country outside the agreement is building a GPU cluster, be less scared of a shooting conflict between nations than of the moratorium being violated. Be willing to destroy a rogue data center by airstrike.”
Okay, so there’s the incendiary line at the end of that paragraph, and you can see he’s kind of escalating — worst case if this happens, here’s how you might respond. All the way up to “be willing to destroy a rogue data center by airstrike.”
And then immediately after this was published, you have people going all the way up to Marc Andreessen posting on Twitter even back in 2023, and of course again recently with this wave of violent attacks on Sam Altman’s house. But back in 2023, I distinctly remember Marc Andreessen and other intelligent people tweeting, “Okay, this is a call for violence. Eliezer Yudkowsky is calling for violence. People who believe in AI doom are calling for violence.”
And so that brings us to our debate today. I’m gonna let you give your own interpretation of why you agree with Marc Andreessen and the others. You agree that the paragraph I just read in the context of this article in Time Magazine was in fact a call for violence, correct?
John’s Two-Part Argument
John 00:07:50
Well, with your take, how would you define a call for violence? What does that mean?
Liron 00:07:54
Okay, so I think maybe one starting point is if an airstrike is in fact launched at a data center and blows it up, I agree that if you draw a circle around that event, that is indeed violence.
John 00:08:06
Okay.
Liron 00:08:06
So I guess where I would disagree is mentioning that there is an escalation path, an enforcement path that gets you to a violent end state. In the tree of possibilities of what you’re calling for, the fact that there’s one unfortunate branch of the tree where somebody’s violating your proposed treaty, and then you have to enforce the treaty, and then you have to use violence — zooming out and looking at the whole tree of if/then that you’re proposing, and then calling the tree as a whole, the policy as a whole, calling the policy a call for violence, I think is actively misleading.
John 00:08:38
Aren’t you a paperclip guy? Just explain to me. Aren’t you a paperclip guy?
Liron 00:08:41
So I am somebody that has a high P(Doom), and my main — I go around saying fifty percent—
John 00:08:47
What does that mean? What does that mean?
Liron 00:08:48
My probability that the world is going to end soon in the sense that we’re not going to leave descendants, it’s just gonna be overrun by AIs that have values that we would think are monstrous or just null — as bad as dust — or worst case even create a bunch of suffering consciousnesses. That’s a very real possibility.
John 00:09:05
And you’re modeling that as a probability?
Liron 00:09:06
Yeah. I think the probability that by 2050 we get on this irreversible path to a paperclipping or just a value destruction of the whole universe is in the ballpark of fifty percent. I often say ten to ninety percent because I’m not trying to put a precise probability on it. I just think it’s crazy to be really confident in—
John 00:09:20
But a very high probability.
Liron 00:09:23
Very high probability, yeah.
Conditional on High P(Doom), Is Eliezer’s Policy Bad?
John 00:09:24
You have derived some number, and it’s a very large number — a coin flip on the end of the world. Okay? I need to understand how you’ve come to that conclusion. So can you just describe without bringing in somebody? I don’t want to debate anyone else in absentia. They’re not here.
Liron 00:09:42
I want to focus our discussion on the call for violence part. Is it okay if I just restate my argument without invoking Bayesian epistemology?
John 00:09:48
Are you asking me if it’s okay for this discussion for you to assume the world will end with a fifty percent probability? No, it is not. We can get to the question of have we gone kinetic.
And the question I will ask you is, if you can get from where you are right now to the end of the world via AI, how can you not get from airstrikes against potential — there’s no carve-outs for allies. Data centers have people in them, working in them, so there’s no carve-out for allies. There’s no carve-out in his quote for other, for being at war or acts of war or military.
So if the data center underneath the Kremlin is doing too much GPU, we’re gonna airstrike it regardless of the consequences because extinction. Well, okay, because extinction. Now I ask you to justify “because extinction,” and you can’t.
So now you say, “Well, it’s kinetic. I should go do this with no carve-out for anyone or anything because extinction,” and you can’t justify “because extinction.” So really there’s no argument. It’s sort of unsound reasoning — from false implies anything.
Liron 00:11:13
Yeah. So let me just—
John 00:11:14
Your assumption is false.
Liron 00:11:15
Okay.
Be Like Carl Sagan — Win in the Arena of Ideas
Liron 00:11:16
In these debates, a really productive exercise I find is I’m gonna spend a lot of time just repeating your point back to you to make sure I’ve got it. I think that’s useful.
John 00:11:26
Yes, a hundred percent. Love it.
Liron 00:11:26
So here’s what I’m hearing. Two different arguments. You’ve clarified part one of your argument. Part one is: you high P(Doom) people, you’re making an epistemological mistake. You’re upgrading being concerned, which is legitimate, upgrading that to thinking you know way more than you can know and being way overconfident. That’s part one of your argument.
And then part two of your argument is: you’re proposing a policy which is inherently reckless and violent, even separate from your belief state. The nature of your policy — that you can have an international treaty enforced by airstrikes — there are circumstances like your example of, “Oh, really? You’re gonna do an airstrike under the Kremlin if Putin is working on that stuff even if Putin never agreed to the treaty?” So you’re saying the policy itself, regardless of part one of your argument about the belief state about AI doom, you’re saying the policy itself is inherently violent, correct? Those are two parts of your argument.
John 00:12:16
That is correct. And so you have this situation where you have a stated policy, Time magazine, as you said. And then you have books everywhere that said, “If anybody builds it, everybody dies,” or something like that. So everybody dies, death but dignity, airstrikes.
And then you struggle to find your way to how that sort of kinetic, violent rhetoric would affect other people, and yet you don’t struggle to find your way from where we are now in AI to the end of the world via paperclip maximization. So if you’re going to have it both ways, it’s very confusing.
If you’re going to follow point A to point B to point C in this sort of absurd — you’re gonna call my extrapolation absurd while saying your extrapolation is fine — I object, because you’re saying my extrapolation is absurd. I think it’s not.
I think the first-order consequences of airstrikes against an ally or someone building GPU data centers who are not part of a treaty is not an absurd, abstract extrapolation. I think it’s very clear that a first-order consequence of that would be sort of allowing or condoning violence.
It’s sort of like being confused about how bringing in violent rhetoric doesn’t do harm — that is confusing to me.
No Carve-Outs for Non-Signatories
Liron 00:13:56
I wanna ask for a clarification here, because when I restated your arguments, I was expressing my understanding that the two parts of your argument are separable. So we can have a whole argument—
John 00:14:05
Hundred percent.
Liron 00:14:05
—about how confident do we deserve to be epistemologically about AI doom. And then part two of the argument is, if I understand correctly, it’s independent in the sense that part one, let’s say hypothetically I totally convince you, “Oh, wow, P(Doom) is really high.”
John 00:14:19
Yes.
Liron 00:14:19
But your part two, if I understand correctly, is you’re saying even if P(Doom) is super high, this policy is still a call for violence. And the reason I’m a little confused is because, if I remember correctly, you tweeted something or you agreed with something to the effect of, if somebody believes that AI doom is high, then this policy is rational.
So my question for you now to clarify is: the policy itself, is it rationally downstream of high P(Doom), or is it a separately bad part of what us doomers are proposing? Set on its own, the fact that we’re proposing this policy even with high P(Doom), is that bad?
John 00:14:57
Again, from false comes anything. So if you say, “I’m gonna make all these false assumptions and then draw a conclusion” — yes, if the sky is pink, you’re Bill Clinton. Okay. I grant you that is logical. It is not sound reasoning. So your logic is intact. Your reasoning is poor. You don’t have sound reasoning.
Liron 00:15:19
Oh, geez. Well, I don’t model this as a situation of anything follows from a false premise. This is more like—
John 00:15:24
Oh.
Liron 00:15:24
—which model of the world is accurate, and there can be logical implications that are downstream of a model of the world. So that’s my question to you—
John 00:15:33
Yes. Many logical implications that are unsound, and you’re choosing one that is unsound.
Liron 00:15:36
Okay, so that’s my question to you. Hypothetically, if you were to change your mind on part one and be like, “Oh, wow, Liron was right all along. There’s obvious reasons to think AI might kill everybody” — hypothetically, I know you won’t, okay? But if you did, would you then say, “Well, given the situation, given my new model of the world, I am part of the call to do this treaty that could be enforced by airstrikes in the worst case”? Would you then flip your opinion on the treaty, or do you think it’s overly violent?
Hypothetical: What If the UN Voted for a Treaty?
John 00:16:01
I’ll tell you why the answer is no. I’m not an expert in this matter. I’m an expert on AI, machine learning, mathematics around machine learning, how large language models actually work. I’m not an expert in politics, military interventions. Would I be okay with that policy? How many angels dance on the head of a pin? It’s an irrelevant question, because I’m not an expert in this matter.
However, I do see from historical precedent as well as what’s happening right now in the world that violent rhetoric is not valuable here. And when you start to use violent rhetoric instead of winning in the area of ideas — you can win in the area of ideas, sir. You are a brilliant man. You can convince me one hundred percent. I would be so happy if you falsified me because I would learn something that I don’t know right now from you, and that is the best possible outcome—
Liron 00:16:59
Okay.
John 00:16:59
—of today’s conversation, that I learned something I did not already understand from you. However, I will state uncategorically: violent rhetoric does not help.
Liron 00:17:11
Okay.
John 00:17:11
And I’ll tell you why. You silence your own voice. Now, this is very hard for me to explain without somebody gonna clip this like I’m a — okay. But I’ll just tell you.
I was at Cal, you were at Cal. I was at Cal at the time of the Unabomber. Do you know the Unabomber?
Liron 00:17:31
Yeah, yeah, yeah.
John 00:17:31
Ted Kaczynski? Sending packages. Whether or not anything about that guy — I’m not agreeing with him, I’m not disagreeing with him — my point is simply abstracted away from that. You can never, ever, ever say he had any good ideas whatsoever because he used violence. That invalidated entirely anything he possibly could have said on any subject matter of all time, forever, irrevocably.
Liron 00:18:04
What if he proved a mathematical theorem? I mean, I hate to play devil’s advocate here, but I don’t fully agree with that statement.
John 00:18:08
Well, you can agree with it or not, but mathematics itself is verifiable. I’m talking about opinion. So now you’re saying, if you agree with my opinion, be Carl Sagan. The guy is one of the most well-respected scientists of all time.
You’re getting nervous here, but let me help you understand. He was against nuclear war. He was a very outspoken proponent of nuclear disarmament and against nuclear war. And then there was the Weather Underground that sent bombs to the US Capitol in opposition to the same thing.
And in the mid-’80s, when there was — I’m pulling these numbers — but sixty, seventy, eighty thousand nuclear weapons in the world, you may have thought Carl lost. But then you look at it today, there’s what? Maybe ten, twelve thousand in the world.
So these things take time to progress. From the ‘70s until now, you might have thought, “Well, Carl lost, Oppenheimer lost. They should have done something. They should have upped their rhetoric instead of winning in the area of ideas.” But no, they’re actually winning, and they still stand a chance to win completely. The number has been coming down steadily.
Liron 00:19:18
Yeah.
John 00:19:18
My point is, don’t be the Weather Underground. Be Carl Sagan. It’s absolutely okay to say... Yeah, okay. You get my point.
What’s the Correct Interpretation of Eliezer’s TIME Article?
Liron 00:19:26
Yeah, and I — look, you were saying, just to summarize what I’m hearing now: don’t be the Weather Underground. Calls for violence are bad. Violent rhetoric is bad.
John 00:19:34
Yes.
Liron 00:19:34
Those are some major things you just said—
John 00:19:35
It’s unnecessary.
Liron 00:19:36
Yeah, and guess what? I agree. Three for three. I agree. But I’ve never had vi— that’s this whole crux of disagreement that we’re having now, is that I claim that Eliezer’s article in Time, and similar statements that I personally have ever made related to the subject of Pause AI — all of these people, I claim we’ve never called for lawless violence.
John 00:19:58
I agree.
Liron 00:19:58
We’ve never—
John 00:20:00
I agree. You don’t call for lawless violence and neither — but you do use vio— He, not you. And he’s not here. I don’t, I’m not going to debate him in his absentia. I’ve seen it. The title of his book, “Everybody Dies,” is in big letters.
Kinetic action against data centers is in Time Magazine. And then there’s a fifty-thousand-word post on X where I gotta go in and parse the footnotes to understand the subtlety. It’s like, guys, you can’t have it both ways. And I know that stuff like that sells books and it sells podcasts and it’s good for money. Again—
Liron 00:20:44
When you say parse the subtleties—
John 00:20:44
But that violent rhetoric—
Liron 00:20:45
What’s the subtlety?
John 00:20:47
“Everybody dies” is the least subtle possible thing you could ever say. The least subtle thing you could ever say.
Liron 00:20:52
Oh, sorry. I thought you meant that the subtlety was the subtlety of calling for airstrikes on data centers. You think the headline is airstrikes on data centers—
John 00:20:59
Also very unsubtle. It’s not subtle.
Liron 00:21:01
You think the headline is airstrikes on data centers, and the asterisk is “after there’s an international treaty and somebody breaks the treaty.” Whereas from my perspective, no — the asterisk is that you enforce the treaty with an airstrike. The headline is the treaty.
Liron’s Interpretation: Same Structure as Any International Law Proposal
John 00:21:12
You don’t get “breaks the treaty.” You don’t get “breaks the treaty” because there are non-signatories to the treaty—
Liron 00:21:16
Okay.
John 00:21:19
—including non-signatory allies. So if it was an ally, if it was in Paris, France, and France is not a signatory—
Liron 00:21:23
Yeah.
John 00:21:23
—to the treaty, we should just airstrike France because we don’t like their data centers. Regardless of what the data center’s actually doing — it could be running the modern version of Stadia, it could be streaming video games — but since there’s a lot of GPUs there doing whatever in the middle of France, we’re just gonna airstrike Paris for no reason.
And you can’t find your way to where this is not the right approach for responsible people in the public sphere. That position boggles my mind quite a bit.
Liron 00:21:53
What you just said now, I think there’s something that I partially agree with here. So here’s the scenario that I partially agree with. Imagine that only the US and China get together and have this treaty, and we don’t bring any other countries along. Or we try to bring Russia along, we try to bring France and the UK along, and they’re like, “No, we don’t like this treaty.”
But the US and China shake hands and agree. And then, like you said, France goes off and does something, Russia goes off and does something, and the US and China go airstrike their data centers. We airstrike the Kremlin. I do actually agree. I’m sympathetic to the idea that in that scenario, that’s not really much of an international treaty. That’s a US and China treaty, and it’s dragging along these sovereign nations, and that’s not the right way to do it. And you could even argue that’s an egregiously, excessively violent way to do it. I somewhat agree.
John 00:22:36
It’s violent. There’s no question it’s violent. People work in data centers. It’s violent.
Liron 00:22:41
I somewhat agree with that position because you can argue, yes, in the context of international order, that’s an internationally lawless thing to do when the UN — all these countries in the UN have veto power but you’re not letting them exercise it, and you’re just having two countries — so you’re not giving the international order the monopoly on violence. So I’m actually sympathetic to that argument.
If your whole argument is, “Hey, Eliezer should have made sure to say that so many different countries need to sign the treaty, and you don’t want too few countries signing the treaty,” that’s totally valid.
What Should Eliezer Have Written? “Airstrikes” vs “Strong Deterrent”
John 00:23:11
My argument is you can see your way to the end of the world from where we are right now via AI, but you can’t see from violent rhetoric to what’s going on now. This is an absurd extrapolation.
I know, sir, that you have not called for violence and that you would absolutely and roundly abhor lawless violence, and so probably does the vast, vast, vast majority of people in any movement or any group abhor this and don’t want it.
I would state to you that it is a likely consequence of violent rhetoric where that may be misunderstood. Now, that’s not on you, and that’s not on anyone who says the violent rhetoric. That’s on the person who breaks the law. That’s on the person who does the bad acts.
However, saying you can’t see your way to that eventuality when you can see your way to the end of the earth by something that yet doesn’t exist is a strange way you’ve frankly resolved the cognitive dissonance of your position.
Liron 00:24:25
That seems like a stretch. You’re saying, “Hey, you have different epistemology about P(Doom), and also you’re struggling to understand exactly what my position is on this call for violence,” and you’re drawing a connection between those. I encourage you to try to compartmentalize and keep being charitable about my understanding.
John 00:24:40
No, I’m saying the first-order consequences of a stated policy are not absurd extrapolations. Your point is the first-order consequences of a stated policy are absurd extrapolations, and my point is they’re perfectly foreseeable.
How John Would Rewrite the TIME Piece
Liron 00:25:01
I’m still in the mode where I’m trying to load your entire position into my head. We’ve made some progress. In order to help—
John 00:25:05
Well, just state your position. Don’t worry about repeating my position back to me.
Liron 00:25:09
No, no, no. I could state my position all day, but if I want to debate you—
John 00:25:14
Please.
Liron 00:25:15
—I need to know what position I’m debating, and I’m seventy percent done understanding what your—
John 00:25:19
All right. Keep going. Let’s get to a hundred.
Liron 00:25:22
So a position has branching structure. I do need to probe your position with a couple hypotheticals to be like, “Okay, got it. This is now sufficient.”
John 00:25:32
Let’s do it. Let’s do it.
Liron 00:25:32
Okay. So I think an important hypothetical to probe your position is: imagine a lot of people, a lot of world leaders get convinced, wow, there’s a very uncomfortable chance that AI is going to end the world soon. A lot of people hypothetically get convinced of that, even if you disagree with them.
And then there is a UN meeting, or a the most unanimous imaginable kind of vote on the international stage where all of these countries are like, “Yes, yes, yes, yes.”
John 00:25:55
Yes.
Liron 00:25:55
I know this is a hypothetical. This isn’t realistic, but—
John 00:25:58
No, no, I’m with you. Keep going.
Liron 00:25:59
Okay. So in the scenario, just to review: the Time Magazine article comes out—
John 00:26:03
Right.
Liron 00:26:03
—a bunch of world leaders talk to Eliezer, and they unanimously agree, “Wow, Eliezer makes sense, and the people that we represent, they support us.” Sixty percent of every country supports us, and they unanimously vote to have this treaty. And then the implication of the treaty is that if rogue actors violate the treaty and they refuse to stop, then they get worst case an airstrike on their—
John 00:26:23
Yes. Yes. Yes. I agree. Your point is correct.
Liron 00:26:25
Look at that scenario. When you look at that scenario and your analysis is this scenario includes violent rhetoric because the part in 2023 when Eliezer wrote his Time Magazine article, that was violent rhetoric.
John 00:26:36
Correct, because in your scenario that you just handed me, 2023 predates the decision by all these world leaders to have a treaty. So yes, the violent rhetoric predates the treaty. The violent rhetoric is occurring now, and there is no treaty.
So not only is it not a hypothetical, it’s actually existing, and to your point at the beginning, people are operationalizing violence. Now, that’s no one’s fault but the person who operationalized that violence. However—
Liron 00:27:04
Yeah, but we’re talking about the scenario. I’m still talking about the scenario. I’m trying—
John 00:27:06
Yes, I know you want to have a hypothetical where—
Liron 00:27:08
So if I talk about the scenario for a moment—
John 00:27:10
I agree. I do.
Liron 00:27:10
Okay. In this scenario, is the problem that Eliezer called for something that was more extreme than what ended up happening? Are you analyzing it like, “Yep, Eliezer called for violence, but then the countries did better than Eliezer. They ended up on something that wasn’t violent.” Is that your analysis of the situation in this scenario?
John 00:27:27
No. My analysis of the situation is prior to the agreement, prior to the treaty, prior to anything else, there is no call for violent rhetoric because there are no exigent circumstances, and violent rhetoric is counterproductive to your position.
My position is very simple, sir. It’s not caught up with all these complicated scenarios and situations. It’s very simple. You go in Time Magazine, and you use kinetic language to describe a situation. You do that intentionally. You write a book, and you put “Everybody dies” on the cover. You do that intentionally.
Said actions predate not only AGI but any sort of international coalition against GPUs. Now, if there’s an international coalition against GPUs, and this international coalition, fantasy as it may be, decides to do something, am I okay with that? Well, maybe I’m okay with it, maybe I’m not. I’m not gonna sign on to that sort of hypothetical in the future because I don’t have all the details.
But what I will tell you is prior to that event, violent rhetoric in Time Magazine is not called for. Go to the President of the United States, go to the president of all our allies, and use all the violent rhetoric in private you want. There’s no reason to put it in Time Magazine. That doesn’t help. It hurts. It hurts your position.
Carve-Outs: Allies, Civilians, Consequences
Liron 00:28:48
You remember when I factored your argument into part one and part two, where part one is it’s inherently violent to assert a high P(Doom)? That was part one of your central claim. And part two of your central claim is, conditioned on having this belief that P(Doom) is high, it’s inherently violent to propose a treaty that’s enforceable by airstrikes. I tried to factor your argument into two major claims.
John 00:29:11
You’re not restating it quite correctly.
Liron 00:29:15
Okay. Yeah, correct me.
John 00:29:15
The first part of my argument is you have arrived at a number, and I asked you to derive that number, and there was no derivation. So the first part of my argument is I don’t understand your derivation.
The second part of my argument is simply this: violent rhetoric hurts your position. It has no value. It has zero value.
Liron 00:29:39
No, but by the time you’re already saying — when your claim is “violent rhetoric hurts your position”—
John 00:29:42
Yes.
Liron 00:29:42
—that’s outside the scope of what I wanna argue about. I was trying to argue whether—
John 00:29:45
If that’s perfectly within scope, in my opinion. You asked me, is this a call for violence, and should we do it? The answer to you is no, we should not do it.
Liron 00:29:55
Okay, but I actually did not ask should we do it. This whole debate that I was hoping to have is just I claim that I have never had a call for violence. The policy proposal that I have, which does include a branch of the possibility tree where a violator of the international democratically voted on treaty—
John 00:30:11
Yes.
Liron 00:30:11
—a violator does get enforced, up to and including airstrikes. When I make that claim, I just wanted to come argue with you—
John 00:30:17
You’re restating his position, sir. Sir, with all due respect, you’re restating his position incorrectly. So his position has no carve-out for signatories. It has no carve-out for civilian targets. It has no carve-out for any nation.
His position is simply: if they’re violating a treaty that I have, I will use kinetic enforcement of a treaty that they may not even be signatories to. So you’re saying, “Oh, a member of the coalition violates the treaty.” This is not Eliezer’s argument, sir.
So let me give you a more proper reading of his scenario. We come to a broad coalition, and there are many signatories. There is a non-signatory ally who does not come to the same decision. The non-signatory ally builds a data center beneath their capital in a civilian population that is doing AI.
Intelligence, according to the Time magazine quote, decides unilaterally — there’s no carve-out for a vote. His quote doesn’t say the coalition decides. He says unilaterally, the United States or other country that is a signatory — so let’s abstract away the country, it could be any signatory — decides to do an airstrike against a non-signatory violator on the basis of intelligence.
So now that is a correct interpretation of the quote. Your interpretation of the quote is simply incorrect.
Closing Arguments
Liron 00:31:59
My interpretation of the article is no, it’s not violent rhetoric because the whole methodology of the article was the same methodology that one uses to make the argument — the same structure of an argument that one uses for any type of international law proposal. Any type of serious international law proposal that comes with enforcement has a violence case. Normally, the violent case is the asterisk. It’s the details.
John 00:32:24
This is what’s known as a straw man argument, sir.
Liron 00:32:28
Okay.
John 00:32:28
So you’re presenting now a straw man argument, which I’m not going to go beat up on the straw man. I’m not an expert in international law. I’m asking you a very simple question, and that simple question is: the text you read, regardless of the intent of the author—
Liron 00:32:42
Yes.
John 00:32:42
—in your interpretation of these words, an intelligent man, very highly intelligent man, by the way—
Liron 00:32:50
Yes.
John 00:32:50
—your interpretation of these words, regardless of author’s intent or who said them, is that violent rhetoric? Is that what we would categorize as violent rhetoric?
Liron 00:33:01
Yeah. That’s what I wanna argue about. I claim that a proposal—
John 00:33:05
But you won’t answer.
Liron 00:33:06
—for domestic or international law, even though governments have a monopoly on violence — but in general, the general category of proposals of laws, “this should be a law, an enforced law” — when somebody makes a proposal in that category—
John 00:33:19
Yes.
Liron 00:33:19
—that person is not, quote-unquote, calling for violence. That would be a—
John 00:33:24
I didn’t ask that. Again, you’re answering a question I didn’t ask.
Liron 00:33:28
Okay.
John 00:33:28
The question’s simple. Is it violent rhetoric? I’m not saying it’s a call for violence. I’m asking you this question.
Liron 00:33:34
Okay. Yeah, and I—
John 00:33:34
Is it violent rhetoric?
Liron 00:33:34
No, I don’t think it’s violent rhetoric. I don’t even think it’s close to being violent rhetoric.
John 00:33:39
Well, okay. Why not? Now we’re getting somewhere. Explain to me why airstrikes is not violent.
Last Q: Does High P(Doom) Imply Violence?
Liron 00:33:42
Other examples of international law, like what if all the countries—
John 00:33:44
No, no, no. No examples.
Liron 00:33:45
—came together. If a country is—
John 00:33:46
Why isn’t that violent rhetoric? Give me a rational reason why that is not violent rhetoric.
Liron 00:33:51
Okay, so maybe this will make it obvious if I just use an analogy. If all the countries came together and they said, remember CFCs? I think we did a good job reducing CFCs and preserving the ozone layer. But imagine some countries are like, “No, I have to keep emitting CFCs.” So sufficiently many countries, let’s say hypothetically every country except one, came together and said, “Hey, we do have a no CFC policy. We as the UN have voted for that, and at some point, if you’re still emitting this much after a year, we start doing sanctions if you have some crazy CFCs you emit.”
John 00:34:19
This is what’s known as a straw man argument. This is what’s known as a straw man argument. You continue to make them. I don’t know anything about carbon emissions. I’m not here to talk about carbon emissions.
I’m asking you a very simple question, and I know why you won’t answer it, because if you answer it, the entire linchpin of your argument collapses. So if you admit — and it’s very, very clear, prima facie, that this is violent rhetoric. Anybody who reads that who’s independent, any reasonable interpretation of the quote you read is violent rhetoric.
Now, you agreed violent rhetoric is not valuable in a discussion, and if you agree that the quote in Time magazine is violent rhetoric, then you agree that it’s not valuable, and you won’t do that.
So I’m not calling it a call for violence. You jump to “call for violence.” You present many straw man arguments because you don’t wanna answer the simple question: is this violent rhetoric?
Closing Thoughts
Liron 00:35:20
Well, at the end of the day, I always actually hate when the argument devolves into semantics, and if the whole crux of our disagreement is we say, “Okay, it’s not a call for violence, but is it violent rhetoric?” Even though I definitely think the semantics are no, it’s not violent rhetoric — if you think the semantics are yes—
John 00:35:35
Which I do.
Liron 00:35:35
—then we can always just find a compromise where we just redraw definitions of terms and just call it a day. But I just think you’d be—
John 00:35:39
I just wanna understand your mind. I want to understand how that is not violent rhetoric. Just give me one, without analogy—
Liron 00:35:47
Okay.
John 00:35:47
—without bringing up chlorofluorocarbons, without bringing up international treaties that don’t exist, without bringing up AGI, which has no definition, or ASI that has no definition. Without any sort of additional framing whatsoever, explain to me how those words are not violent rhetoric. Very simple question.
Liron 00:36:08
Sure. I mean, imagine that somebody is gonna go plant a bomb in the Hoover Dam—
John 00:36:15
This is an analogy.
Liron 00:36:15
—and a guard shoots him. And I support a policy of shoot on sight for somebody who’s trespassing in our country.
John 00:36:24
This is an analogy. Violent rhetoric. Violence in words. Is this violence in words?
Liron 00:36:31
Is this violence in words? I say no, because when violence is in the context of centralized government policy of law — legal violence is not... Calling for legal violence in the context of there’s a law, and that’s why it’s enforced with violence—
John 00:36:49
Agreed, agreed. I didn’t say it was illegal. I didn’t say it was unlawful. I didn’t say it was immoral. I asked if it was violence—
Liron 00:36:55
Okay.
John 00:36:55
—and your response was, “Yes, but it’s legal violence.”
Liron 00:37:00
If you’re willing to use your terminology and say every call for an enforced law or international policy of any kind is violence, then I will happily agree that because this is that, then this would be violence. I just think—
John 00:37:11
Thank you.
Liron 00:37:11
—that’s a pathological way to define—
John 00:37:12
It’s not pathological at all.
Liron 00:37:14
—call for—
John 00:37:14
I object to that characterization. I think it’s very simple.
Liron 00:37:19
Okay.
John 00:37:19
We now agree this is violent rhetoric. Now, I don’t find violent rhetoric to be valuable in this discussion. I don’t think it should be used because it can be misinterpreted by individuals, and that has happened. That has actually occurred.
Testing the Definition
Liron 00:37:37
So just to be clear, if somebody says, “Hey, if a trespasser makes it past a certain point that’s very dangerous, toward blowing up infrastructure, toward being within striking distance of infrastructure, the law says that there’s enforcement with shooting or whatever.”
A security policy, okay? If I propose, “Yeah, we should secure our infrastructure,” would you say that is violent rhetoric?
John 00:38:00
Yes.
Liron 00:38:01
Okay. All right.
John 00:38:01
No, it’s justified.
Liron 00:38:03
So you’re extending the definition quite a lot. Okay.
John 00:38:05
No. Sir, violence is violence. You’re confusing justification, lawfulness, the reasoning, the mens rea of the individual. Violence is violence.
Liron 00:38:16
All right.
John 00:38:17
And we seem to be unable to agree on what is violence. Okay, now that we agree on what is violence—
Liron 00:38:24
Okay.
John 00:38:25
—and we agree that violent rhetoric is not valuable, I think what you want me to say is the gentleman who spoke those words in Time Magazine is somehow culpable for the unlawful actions of individuals.
Liron 00:38:36
Okay, can I—
John 00:38:38
I will say absolutely not. He is not responsible for those acts.
Liron 00:38:39
Do I have this right? You’re saying talking about enforced laws is violent rhetoric.
John 00:38:44
No.
Liron 00:38:45
Therefore, talking about any kind of laws is bad?
John 00:38:48
No. Enforced laws is write me a parking ticket, so don’t restate my argument.
Liron 00:38:55
Okay, talking about any kind of law—
John 00:38:56
I said violent rhetoric is violent rhetoric.
Liron 00:38:57
Yeah. Okay.
John 00:38:59
There’s no need to redefine it. It is what it is. It’s normal English.
Liron 00:39:03
Right, so it’s talking about any law that carries the death penalty.
John 00:39:05
This is subject, object, verb English. Yes. Why do you continue to wanna redefine words that simply don’t need to be redefined in this discussion? It makes no sense.
Clarifying the Terminology
Liron 00:39:12
So just for the record, okay? Just so viewers understand.
John 00:39:15
Yes, yeah. For the record.
Liron 00:39:15
You’re saying that any kind of discussion of any law that is enforced up to and including the death penalty or serious enforcement like that—
John 00:39:24
Enforced with violence is violent rhetoric.
Liron 00:39:25
Yeah. Okay. Got it.
John 00:39:26
No. It includes violence. It includes violence. Okay? I’ll agree with that.
Violent rhetoric is the call for violence against allies, non-signatories, data centers. That is a call for violence. Now, is that a call for unlawful violence in the community? Absolutely not. He never got up and said, “You should go commit crimes based on this issue.”
He said what he said. I’m not gonna restate or recharacterize it. That is a call for violence in my definition. Yes, it may be completely lawful, completely justified morally. It may be within the bounds of a hypothetical treaty or laws. It is a call for violence. It is not a call for vigilante justice.
Liron 00:40:23
Okay.
John 00:40:23
It is not a call for lawless violence. It is not a call to go assault someone’s home or firebomb. It is not a call for any of that, nor does that individual bear any culpability whatsoever for the violent and unlawful acts of other people.
I am merely stating that is a call. In other words, we should use violence. In absence of the definition of “call for violence,” that is what I would place there. An appeal — that’s a call — for violent actions. That’s an airstrike. So this is an appeal for violent actions.
Liron 00:41:01
Okay.
John 00:41:01
Yes?
Liron 00:41:01
I think I actually understand your position.
John 00:41:03
Do we agree?
Liron 00:41:04
Okay. So that’s good. It’s good that we’ve clarified the semantics. We’ve clarified how we draw boundaries around things, and I—
John 00:41:10
Do we agree?
Liron 00:41:11
I agree that you can operate semantics—
John 00:41:13
Excellent.
Liron 00:41:13
—the way you’ve been operating them. I mean, that’s the most I can say besides—
John 00:41:15
What does that mean, “operate semantics”? I’m speaking English.
Liron 00:41:18
But you can draw—
John 00:41:19
Subject, object, verb English.
Liron 00:41:19
I agree that there is a logically coherent way that you can say, “Hey, violent rhetoric includes somebody proposing—” Remember, we said, “Anytime there’s a law whose enforcement goes up to and including—
John 00:41:29
Good.
Liron 00:41:29
—the death penalty or an airstrike, anytime that law is on the table, then that counts as violent rhetoric.” If you wanna draw that semantic distinction, you can. I disagree with it, but I don’t think there’s more to say besides which—
Why Airstrikes Specifically?
John 00:41:40
Well, now explain — there is more to it. There is more because there’s an entire other side of why calling for airstrikes is not a call for violence. So now please explain to me—
Liron 00:41:50
Yeah.
John 00:41:50
—why it is not a call for violence.
Liron 00:41:53
When Eliezer wrote his post, the scenario my mind went to as the mainline scenario — not even a hypothetical scenario, but a realistic scenario — is that a bunch of countries sign the treaty, some critical mass, all the major AI powers sign the treaty, and then nobody dares build these malicious data centers, and that’s it, and everything’s peaceful.
John 00:42:12
Well, he should’ve just said that without—
Liron 00:42:14
No, he did.
John 00:42:15
—but that would be a great point. No, he—
Liron 00:42:17
That is what he said.
John 00:42:18
It isn’t, sir. He said... If he had said, “We can put together a broad coalition of governments and governance bodies,” that would be a strong deterrent for violation of the treaty. Strong deterrent is not violence. Strong deterrent is a strong deterrent.
Now, there are many ways to deter. We do it all the time — sanctions, political pressure, all sorts. I’m not an expert in this area. Again, I don’t like to go outside of my area of expertise.
However, saying “a strong deterrent” is not violent rhetoric because it contains no violence. It is simply a deterrent. So I agree with you 100%. He should have said... I’m not gonna characterize that. I’m not gonna debate him in his absentia. I would have liked to see some more thoughtful rhetoric than airstrikes. I know he wanted to—
Liron 00:43:21
Yeah.
John 00:43:21
—make a point, and he was using appeal to emotion to do that. I get it.
Eliezer’s Strategic Reasoning
Liron 00:43:27
I don’t think he’s appealing to emotion. He’s addressed this. So imagine what would’ve happened if Eliezer didn’t say airstrikes and he was just like, “Yep, we need international coordination on this front.”
John 00:43:35
A strong deterrent.
Liron 00:43:36
Yeah, or even said deterrent. The scenario he was worried about is: imagine you got a bunch of lawmakers being like, “Wow, what a reasonable claim. I’m going to support this.” And the UN votes, and they’re like, “Okay, any country that is training frontier AI models in violation of the central body that we make, any country doing that, they’re gonna get slight sanctions that are going to cost them a billion dollars a year off their economy,” which is small potatoes.
And everybody shakes hands like, “Yes, we did it, guys.” And of course, meanwhile, some people are like, “Hey, I’m gonna get rich. I’m gonna train a super intelligent AI, and I’m happy to have my country pay the billion dollar fine.” And then we all die. That was the scenario he was trying to guard against.
John 00:44:14
And this position’s absolutely reasonable. The position you just took — that it would be a slap on the wrist, and therefore everyone would do it — absolutely fine.
I would probably agree. However, you don’t need to escalate to specific calls for violence to make your point. You can simply say, “Sufficient deterrent.”
Liron 00:44:38
Gotcha.
John 00:44:39
We’re intelligent people. He has the opportunity. I don’t know that he has the opportunity. So when you have something published in Time Magazine, I think it’s fair to say you have a very broad reach. You have an opportunity to have a very broad reach.
And when you have the opportunity to have a very broad reach, you should probably choose your words quite carefully, and I’m sure he did. ## The Language Choice
John 00:45:02
For the purposes that he wanted to achieve. And to your point, you feel, and I think you’re right, that there are some elements of government agreement which would appear to be a solution, which would appear to be a deterrent, but would not be, because the punishments and the enforcement is not sufficient compared to the value of building these things.
We will have broad agreement on that, sir. I think, again, I don’t say you shouldn’t be concerned, and I don’t say you shouldn’t have the positions of sufficient deterrence that you think are important because of your belief system, and I respect that.
My point is simply this can be made clear to intelligent people without violent rhetoric or without calls for violence, and I believe calling for airstrikes is a call for violence. It’s not a call for lawless violence. It’s not a call for vigilante violence. So please don’t restate my idea. It is simply, you asked me to come on and say why I thought it was a call for violence, and that’s why.
Liron 00:49:09
So if Eliezer had taken out the phrase where he says “up to and including airstrikes on data centers”—
John 00:49:15
Yes.
Liron 00:49:16
We can even go back to his—
John 00:49:16
Yes.
Liron 00:49:18
—exact phrase. Give me a second.
John 00:49:18
And he could have said military intervention.
Liron 00:49:20
So the quote says, “Be less scared of a shooting conflict between nations than of the moratorium being violated. Be willing to destroy a rogue data center by airstrike.” So what you were saying before is you were saying, “Man, if only Eliezer had just stricken out ‘destroy a rogue data center by airstrike’ and said something like, ‘Be willing to exercise sufficient enforcement’—”
John 00:49:38
Yes.
Liron 00:49:38
—then it would no longer be a call for violence, even if it would have—
John 00:49:41
100%.
Liron 00:49:41
—the same consequences.
John 00:49:42
It wouldn’t. It wouldn’t, because it gives us... Number one, it’s not violent rhetoric, and it’s not subject to misinterpretation. Now, that’s not his problem, and that’s not his issue.
But in my opinion, it would have been better because it adds reasonable consequences to an important issue without going over the top, without bringing violence into it. It doesn’t create alarm. And what I think, since you are at a very high number that the world is gonna end, you feel—and I don’t necessarily think from your perspective this is wrong—that alarm is warranted. I don’t think alarm is warranted here. I think concern is warranted.
I think as a community, we all need to be working to make sure AI brings the most possible benefit to the most possible people as quickly as possible. So on these things, we would broadly agree. However, I think bringing violent rhetoric into it brings in alarm. It brings in unnecessary interpretations. It brings in a lot of things that don’t need to be there, because you and I could talk about these things intelligently without that sort of rhetoric.
The Hypothetical Rewrite
Liron 00:49:57
Okay. So this is actually very—I didn’t expect the conversation to go this way. So if Eliezer had written that, instead of saying, “Be willing to destroy a rogue data center by airstrike,” if he had written, “Be willing to use sufficient enforcement against a rogue—”
John 00:50:12
I’d agree with him.
Liron 00:50:13
Okay. And you’d agree with him, and in your mind you’d be picturing, “Hey, worst case scenario, they go super rogue, and they’re gonna need an airstrike.” You’d be picturing that?
John 00:50:19
I would be picturing a broad coalition of international countries deciding at the time what the appropriate intervention ought to be. Could be as much as, “You know what? We’re gonna call this dude and ask him to stop.” It could be sanctions. It could be UN acts.
See, my sort of language and the more responsible sort of language leaves many, many, many options on the table, including military intervention. I never said there should never be military intervention. I said don’t use violent rhetoric.
The Carve-Outs and Nuance
John 00:50:54
Okay. Now, that position I would’ve agreed with. If there’s a coalition and a broad coalition and an international treaty—I’m not an expert in this matter. The governments enforce it. This happens all the time in history, many priors to this event.
Liron 00:51:08
Okay, so literally an airstrike is a hypothetical that you could picture if Eliezer talked about sufficient enforcement.
John 00:51:14
You said come on my show and tell me why it’s a call for violence. Yes, violent rhetoric is a call for violence. Non-violent rhetoric is not a call for violence. Very simple.
Liron 00:51:23
Okay. I think the only property that makes it a call for violence is to make—
John 00:51:27
Is it contains violence.
Liron 00:51:28
—the mechanism of sufficient enforcement more explicit, right?
John 00:51:31
Yes.
Liron 00:51:31
So the word “sufficient enforcement” could mean an airstrike, but he didn’t write “sufficient enforcement.” He wrote “airstrike.”
John 00:51:35
Yes.
Liron 00:51:35
And the fact that he wrote “airstrike”—
John 00:51:36
What makes it violence—what makes it violent language is it contains violence. It’s very simple.
Liron 00:51:42
Because I guess the word “airstrike” is a direct reference to violence, and the word “sufficient enforcement”—but “sufficient enforcement” is a category identifier, right? And we know that airstrikes are in that category, but it’s still not violent to you.
John 00:51:54
Even if he had said “up to and including military intervention.” Sure.
Liron 00:52:00
Okay, but the fact that he’s saying “airstrike” instead of “military intervention” is where you draw the line?
John 00:52:04
No. I draw the line at unilateral. He doesn’t do any carve-out for agreement among the allies. He doesn’t give any carve-out for civilian targets versus military targets. He doesn’t give any carve-out for a non-signatory ally. He doesn’t give any carve-out for the consequences of his actions of stated policy.
Okay? You seem to want to give him all these carve-outs.
Liron 00:52:31
I see. So when you want him to say “sufficient enforcement,” you see—
John 00:52:34
Yes.
Liron 00:52:34
—that term as not just being a category of types of enforcement, but also adding in restrictions like, you know, don’t attack it if it’s a civilian building, that kind of thing.
John 00:52:42
I don’t know. Get somebody on here who does international law for a living. I don’t do that. I do AI. I know what violence is and what it’s not.
Liron 00:52:48
Okay, I think this has been pretty illuminating. I certainly learned more about your position, and it’s more clear to me than it was at the beginning. At the very least, we’ve done that.
John 00:52:55
Thank you.
Testing the Final Distinction
Liron 00:52:55
I’ll just ask you one final clarification because I feel like this was an important moment in the conversation. So if Eliezer Yudkowsky had just replaced the phrase “Be willing to destroy a rogue data center by airstrike” with the phrase “Be willing to exercise sufficient enforcement,” and even if that phrase in some cases implied airstrikes—if he had done that, if the article had come out like that, you’d be like, “Yep, no calls for violence here. All good.”
John 00:53:18
Yes, I would have thought that was a much more thoughtful thing to do. I would believe it did include potential military intervention. I think it would be a reasonable position that he could win in the area of ideas.
And I think it is more responsible and less likely—although I’m not holding him to account for what people do or others do. That is not his responsibility, nor is it yours, nor mine. The responsibility falls squarely on the person who has violated the law.
I would think that is a more responsible use of language, and I would not call that a call for violence. I call the other a call for violence simply for one reason: it contains violence.
Closing Statements
Liron 00:54:09
Got it. All right. Let’s move into our closing statements. So I’ll kind of recap things from my perspective, and then I’ll give you the last word to recap. Sound good?
John 00:54:16
Absolutely.
Liron 00:54:19
Great. Okay, so the initial claim that I wanted us to debate, that you were very game to do, is this idea of was Eliezer Yudkowsky’s article and proposal in Time Magazine in 2023 a call for violence? And of course, that’s kind of a semantic distinction. We can just redefine terms, or people can have different interpretations of terms.
To me, I think it’s clear enough to tell when somebody is posting a call for violence versus when somebody is posting a proposed international policy. And then as a footnote—practically a footnote, technically it was in the article—but practically a footnote to point out that international policy is “we’re not messing around here.” This isn’t a slap on the wrist policy. This is a policy that, in the unfortunate situation where somebody insists on ignoring the warning shots and continuing to train, train, train, and they will not listen to reason—in that particular scenario, then you gotta bust out the enforcement, or yes, airstrikes as a type of enforcement.
So when he included that, in my mind I was like, “Yep, uh-huh. That’s what it looks like to enforce a policy. No problem here.” But then people jumped down our throat, and they said, “Oh my God, violence!” And I was like, “Wait, no, no, no.” The people making that claim, you are deliberately putting the cart before the horse and saying a policy that has an unfortunate violent component if necessary, if invoked, if people violate it—such a policy, calling for such a policy, you take the violence part and you promote it to be the headline of that policy. To me, that seemed like a malicious move, a disingenuous move.
And then I later changed my mind, and I’m like, “You know what? The people who did that, I don’t think they’re being disingenuous.” I actually think that in their mind, they just see it and they’re like, “Violence, violence, violence.” They see the difference between saying the word “airstrike” and saying the word “sufficient enforcement.” They, as you’re saying, see that as a huge, huge difference, and they think that gives them the legitimacy to say, “Oh my God, a call for violence. AI doomers are being violent.” Whereas to me, I’m like, “Wait, what? That’s not how you analyze it.” So that’s recapping the discussion from my perspective.
There was a part at the beginning where you mentioned that being high P(Doom) itself—there’s violent rhetoric even in the way that we talk about AI going to end the world. Fine. I think we kind of factored that out of the crux of disagreement in the scope of this particular conversation.
And lastly, I just wanna say, hey, you came on, you were super game. You’re a great participant in the discourse. I wish more people—I wish Marc Andreessen, who I see as one of the most prominent people taking a similar position to you—I wish he would come on and defend his claim. I think it helps the discourse when people try to understand each other, and I really appreciate you for that. All right. Giving you the last word.
John 00:54:48
Thank you, sir. I really, really have enjoyed our conversation. I really appreciate you having me on.
I agree with everything you just said. I don’t think you should be blamed for the actions of others. I don’t think anyone should be blamed for the actions of others. The actions of others are their responsibility. You are not calling for violence. You don’t want violence. You don’t want violence against individuals. You don’t want it to come to that.
I think the jump from violent rhetoric being used to “it is your fault” is wrong. So I absolutely agree with you. I think I’ve defended the position of why perhaps violent rhetoric may not be the correct choice here, due to—I think both of us can agree that there’s broad misunderstanding about AI, and we have an opportunity as practitioners to resolve that misunderstanding.
The Responsibility Question
Liron 00:56:27
Thank you, sir. Really appreciate it. One quick thing. This is actually outside the scope of our debate, but I forgot to ask your take on this. There’s been a lot of people on social media who—
John 00:56:36
Yeah.
Liron 00:56:36
Sarah Haider is one, who I think she’s coming at this in really good faith, and she’s one of the best representatives of this position where even though I disagree with her, I think the way she’s going about it is totally authentic and productive. Her argument, and a bunch of other people’s argument, is like, “Hey, if I was an AI doomer, if I had a high P(Doom), then I would be calling for violence. That’s the natural thing to do, or at least I would be condoning violence.” Like, you guys—
John 00:56:59
Of course.
Liron 00:57:00
—gotta own it. That it’s implied by high P(Doom). Do you think that that is a valid argument?
John 00:57:04
No, that’s an absolutely valid argument. Look, I’m not here to disabuse you of your beliefs. My question on your beliefs is not that they’re wrong.
Liron 00:57:11
All right.
John 00:57:12
My question is please help me understand—
Liron 00:57:14
But I’m saying the implication isn’t right. You say you agree with the implication that if somebody’s—
John 00:57:17
Of course.
Liron 00:57:17
—a doomer, then they... Okay.
John 00:57:19
Look, look, look. Let me—let’s make this clear. This is a very simple argument that you and I agree on. My question is how do you get to fifty percent, not what that implies.
Liron 00:57:31
You and I should actually make a clarification because we spent the whole conversation talking about how military enforcement of a certain kind—getting a little premature with airstrikes—that’s violence. I think I need to actually clarify. In the case of the people that I’m responding to right now, I actually think those people are claiming that being a doomer should actually make you condone things like the Sam Altman’s house Molotov cocktail attack—
John 00:57:52
Absolutely not.
Liron 00:57:52
—the Wallace—
John 00:57:53
Absolutely not.
Liron 00:57:54
Okay. So you don’t—
John 00:57:55
That is absolutely wrong.
Liron 00:57:56
So you actually don’t agree. Okay. You don’t agree with the implication in the—
John 00:57:59
I categorically reject it.
Liron 00:57:59
Okay.
John 00:58:00
You’re not responsible for the actions of people who hold your same beliefs.
Liron 00:58:09
No, but it’s not—I’m not saying I’m not responsible. I think we agree I’m... Well, it’s a stronger claim. Okay. Let’s say I’m not responsible, but given that I’m not responsible, isn’t it rational for me to condone it? I don’t think it is, but do you think that I should think that I should condone it?
John 00:58:20
Absolutely no. Look, you were very clear with me. I think you’ve been very clear in public from what I’ve seen, that you do not condone these actions. And it is not a strike against your beliefs that others do violent and criminal acts.
I don’t know what the gentleman who allegedly did this—what his belief system is. I don’t know if it aligns with yours. I don’t know anything about it. You are here as an intelligent man having intelligent debates. That is the right thing to do.
You’re like my hero Carl Sagan. You’re like Carl Sagan.
Liron 00:59:26
[laughs]
John 00:59:27
You’re against this thing. Yes. You’re against this thing, and you’ve got your channel, and you’re having intellectual debates where you win in the area of ideas. I praise you. I back you. That’s why I’m here.
Final Words
John 00:59:40
You’re doing the right thing. To say everyone broadly with this belief system is responsible for the actions of a single individual who may or may not have this... It appears based on the evidence that there are some similarities, but he certainly doesn’t have your morality or your intelligence or your propensity to win in the area of ideas.
If you can convince me that you’re right, you have taught me something I don’t now know, and my mind is always open to that. I invite falsification of my ideas. I would love for you to falsify my ideas. You are doing this exactly correctly. There’s nothing you’re doing that’s wrong. You have no culpability. I support you one hundred percent.
Liron 01:00:27
All right. Thanks so much. It means a lot that you’re granting me the good faith. You’re being charitable toward my motives and my way of going about it. I’m happy to extend the same to you and—
John 01:00:37
Thank you.
Liron 01:00:38
—many, but not all, of the people I see on social media who disagree with me. I’m happy to also extend charity that everybody’s doing their best. Thank you for helping take the temperature down, because I can say for many of us, it’s been a stressful week, very high stakes here—
John 01:00:49
Yes, I know.
Liron 01:00:51
—both in terms of let’s keep a civilized—
John 01:00:51
For all of us.
Liron 01:00:53
—law society, and also let’s not die from AI. So those are two things we have to both keep in mind.
John 01:00:58
We need to move things into the area of ideas. I would love to come back and talk to you about my philosophy about AI and how to resolve these issues and these types of misunderstandings. I think this was a spirited discussion. I appreciate it.
And I think in many ways you would find that we are allies and not opponents. So again, I appreciate the opportunity to come here and have my voice heard. Thank you very much. It’s really great to meet you. I find you to be an extraordinarily intelligent person, and it’s been my pleasure to meet you and have this discussion.
Liron 01:01:34
Thank you, sir. Really appreciate it. John Alioto, thank you very much for being on Doom Debates.
John 01:01:38
Thank you, brother. It’s great meeting you.
Doom Debates’ Mission is to raise mainstream awareness of imminent extinction from AGI and build the social infrastructure for high-quality debate.
Support the mission by subscribing to my Substack at DoomDebates.com and to youtube.com/@DoomDebates, or to really take things to the next level: Donate 🙏
