# I need to ginish this fucking extended abstract about propaganda. Below is a transcript from a wo... ## Skills Used - calendar-add - harvest - open-note ## Retrieval Notes - Session id: `f56bfd61-be6e-430d-893e-0e0c16229051` - Last activity: `2026-01-24T14:04:19.239Z` - Files touched: `6` ## Artifacts **Created:** - `/Users/nickyoung/Downloads/mckenna-abstract.md` - [[Attachments/mckenna-abstract-temp]] - [[Attachments/mckenna-abstract-export]] - [[Attachments/charter-style.css]] **Modified:** - `/Users/nickyoung/.claude/WRITING-STANDARDS.md` - [[Notes/McKenna Propaganda Paper Plan]] ## My Notes <!-- Add your notes here. This section is preserved across syncs. --> ## Conversation ### User I need to ginish this fucking extended abstract about propaganda. Below is a transcript from a work in progress I did on these ideas from a a couple of days ago. As you can see, it didn't go well. But... I don't really give a fuck any more. I want to get this thing done, basically I want you to think of how to get this closer to the cev of itself, through easy to make changes that I can just drop in. I also want to give myself enough space to actually write a good paper over the next six months (which is when this paper is due) so I don't want to make promises I can't keep, or box myself in too much. you know what I mean. Please give me a long, deep, and detailed answer, with lots of options/ideas: as we are in the 'kicking ideas around' stage, I want as much feedback as you can give me so that i have a lot to consider. transcript: I can just introduce you. It's feeling very this is going to take 20 minutes. This is not gonna be a long time. Yeah, yeah, that's what I'm hoping. Uh, so welcome to the peer seminar and in an unusual, hybrid formaster. And our speaker is our colleague Will give a very long. Thank you for that introduction. Um, So yeah, this is a horrifyingly large. Number of people to be witnessing. This, this is my probably, my least prepared work in progress ever. Um, So what has been happened? Basically, the way this paper started was I was invited to a conference in Prague last year on propaganda and AI. I knew nothing about propaganda, but I thought maybe this will be interesting to learn. I got Prague for the conference and was violently ill and missed the entire conference and spent in the hotel. Um, since that time I've now Um, to revise. What I want to do a few times because I need to send an extended abstract for this edited volume. Based on the conference that I missed. And, I have discovered that I do not like the topic of propaganda at all and I found it very, very hard to find anything. Interesting. So this is the most interesting thing for me at the moment. I suspect especially seeing as there are now people in this audience who know a bit more about propaganda than I did. A few months ago. Um, this might seem a little bit thin, but hopefully you can help me improve it slightly. The first paper, I found that, I found interesting was by a guy called Robin McKenna, um, and he's written a little bit about propaganda. Uh, and he's written also about this paper, which is he talks about it not being propaganda. Um, well Alex, I think he's wrong and that's what I'm going to explain first. So what McKenna is interested in is a worry, we might have about what he calls persuasive, AI And McKenna thinks we should be worried about it. So, He is not thinking that AI llms are much better at persuading people. Of political ideas or anything like that, that ordinary human Persuaders, okay. There's some studies you can see and people are not very well fooled by this sort of stuff or well persuaded by this sort of stuff. Um, but what he does think, what he thinks is going to happen is Or suspect might happen. Is you have certain types of AI system? I'm going to talk about llms mainly. Which are extremely Advanced these days and are very advanced in terms of their reasoning capacities, okay? They're much better than they were a few years ago at working through problems logically. And he says, well look this what might happen here? Is that? Um, Even if Powerful political players. Use this technology. For good, in a way to give good correct. Reasonable arguments. He still thinks that this is going to be a Um, A problem for society because he's worried that the powerful peoples are the powerful governments? I don't know. I don't know some, maybe some Barons in Silicon Valley will just have all of this sort of CPU argumentative capacity. And this will be this potentially could lead to a politically problematic situation because they will have all of the argumentative power on their site. Okay, the idea here is that AI persuasive content. So I don't know, think of a silly thing. Abuse isn't isn't. Socialism greater isn't neocapitalism great or something like that and he's tries to persuade users. You have an LM saying no, no this makes sense. This is the correct way to go stuff like that and he thinks that if you have powerful people in Silicon Valley, they could just roll out all of this persuasive contents on mass. Whereas if you don't have all of this AI processing power to make reasonable true arguments, You're going to be at a massive disadvantage and he worries that this is Um, Sort of flood what he calls the marketplace of arguments. Which differentiates between the marketplace of ideas. So the marketplace of ideas is just I think that various political or sociological ideas you might have, but here he's suggesting, there's also a Marketplace of arguments where you can just crowd out, other people's arguments with your, you know, correct in a sense. Okay. So this is his worry. Okay. And the the phrase he uses he talks about, he talks about llms as sophist Street on steroids, okay? So, if you think about ancient sophists, They were arguers for hire apparently that you could employ them or get them to argue your case or some case you wanted argued. And he says llms are like this as well. They have no morals. They have no skin in the game. You can just make them produce persuasive content Advanced. Yeah. Vast amounts of persuasive content within moments. It would be the best value in the world, but it'll be made for by this massive increase in quantity as the marketplace of arguments. Is flooded. This is his he's he's taught this sort of bad situation. This situation, he thinks is not a good situation. He describes as a best case. Okay, so you know, the people in Silicon Valley or wherever are not Trying to do anything nefarious. They're not trying to lie. They're not trying to manipulate. Um, This is still the best case, which is still reason getting arguments aimed at rational persuasion but just deployed at scale by concentrated actors. He thinks that this is going to cause us. Yeah, he thinks this is not a good position despite it not being So, in seconds, Okay yeah he said so it's still rational persuasion. Okay in the best case, okay, I know if he uses Mitchell and Douglas's, definition of rational persuasion which is a referee persuades B. If a brings it about that big adopt attitude is that Phi, always forget my real numbers Alpha, excuse me, of course. Sorry a brings about the BL2's. Altitude Alpha a does o only by providing reasons for Alpha B dots this based on recognising, those reasons. Okay, so that's what rational persuasion is supposed to be. And also he is saying, well, again we're trying to bracket all of these nefarious things, okay? So it's just rational persuasion and so he thinks he's talking about something which is not Propaganda. Okay, he thinks his best case. Is still problematic, but not problematic because it is Propaganda. Okay. Um, So one question, my paper asks, not the question Um, is, can can this really be made sense of? Can we have this politically troubling Marketplace domination? And it's still not propaganda. So, sort of the second. Third of the paper. I'm trying to write here is going to be to show that it is propaganda. This is I am talking I'm going to take a definition of propaganda which is from the recent Bernard, fantasi and Marcus. Marquez paper where they sort of. Yeah, Revitalise. And re-examine Tuttle Ross's, epistemic deficit, defectiveness model of propaganda. Okay, so I'm not going to draw very much on all of their arguments here, but this is sort of the, the version I've been working off. Okay. So key musician act is propaganda. If and only if it is epistemically defective we'll get to that. There is some sort of political intent. So probably from institution or a political power of some description. It is targeting a socially significant group and the person Distributing it or saying it or producing it, um, is aware or should be aware of the defectiveness. Okay, so The most interesting one here for my purposes is what they mean, what is meant by Epistemic defectiveness. And there's a, there's a sort of the obvious one. If I sort of convince you of something. But it's epistemically defective, the most easiest way to do that is to to lie about something. Okay. So I can create this epistemically defective state in Enrico by lying to him, okay? And then he comes to believe something which is not true. Just by me saying something, which is not true. But there are, of course, other ways you can sort of get people to Produce beliefs, which might not be quite correct or might not be held for the right reasons. Um, so you can appeal to their emotions as well. Um, but what's more interesting from my purposes here is you can you can mislead people Um, by omission. Okay, so You can if you just cherry pick true things, this is still, I could imagine. Imagine this is a true sentence. I'm saying our army won a major battle. Okay, it's true. But um, If I don't tell you that the Army has lost every single other battle this year and the battle that they did when was Chinese, Oh really. Is it that other times? My doctor? But we can cheat. Oops. Because maybe it's a bit loose if you want. We can change to my. Yeah. Or maybe it's because it's like but why don't we try this? Yeah. Should be just the same. Where is it here? And allow. Do I need to make it big? Oh yeah. Yep. Okay good. So, yeah, you can, you can sort of induce beliefs in people in an systemically defective way through things like Omission. Okay, so I think that is that is clear as well. Now this is what I think McKenna's issue is here. Um, so if you think about his best case, And he was remember, he wants to say his best case is not a case of propaganda, but Troubling somehow. He's a state of the world. We don't want to to have happen, okay? But I don't think this works. Um, I mean, If you think about re so I mean it's maybe easier to see the second horn of this problem. First of all, Okay, which is It seems to me that if you say, you were political power and you flood the marketplace with, True. Correct. Arguments. Um, but you do so in such a way that leaves out or crowns out all of these other arguments, it seems to me. You're doing something very much the same as Propaganda by omission and I just don't see how he can avoid saying that because if he wants to say his best case, if we want to say, well in the best case they're not crowding out any of these. These sorry. The the people in Silicon Valley are not crowding out. Any other rational views and they're not doing anything nefarious whatsoever. So that. Yes they are producing lots and lots and lots of logical. Correct. True arguments. Um, but they're doing so in such a way that doesn't crowd out all the other positions and all the other rational arguments. I don't understand how they're doing that without it being entirely untroubling for us if they are. So if they don't give it, if they are sort of throwing out lots of correct, true. Rational arguments without clouting out anybody else, it seems to me. They must just be doing a public service. They must just be telling us true, reasonable things and producing epistemically, non-defective states of belief in us. Understand how we can say that this is not propaganda because he is describing, okay? Because if it's if if you make it too idealised, then I don't see what the problem is. If you make it less idealised, it immediately turns into propaganda. This is the idea. Now, I want to skip that. So, just Okay, so just a few more ways that his sort of best cake. Well, a few more details about how his best case scenario. Just collapses into propaganda as far as I can see which is. Yeah. So I've done the crowding. Out idea if you just have so many rational correct arguments on your side and no one can see other peoples then that seems propagandistic. Quantity style misle leaving this. Um, I think I'll skip that one for now and distorted salience if you just see this. Arguments in favour one thing again, and again, and again. It kind of makes potentially could make people think that these arguments are the best ones, just because they're the ones, you always see. Okay, which is subtly different to one. Yeah, so I've already kind of said this, ideally remove the effectiveness but it's benign or then become propaganda. Okay, so that's the second third of my paper. And now all it already come into the last third and the last third is something I was hoping to get some feedback from you guys on. Okay, because Basically. I have an instinct. Mckenna's idea in general, is just something which doesn't Okay, so let me Try and say where we're at. So by the end of the second, third of the paper. I want to say, well actually McKenna's view is His best case is a case of propaganda, okay? In the in the final part I want to say that is this case even Sort of feasible, could this ever really happen? Okay, and I'm somewhat sceptical that you can really dominate the marketplace of arguments. It seems to me that we don't arguments are not the sort of thing that there are that many of that there are innumerable arguments to be made. I've done a little bit of research in particular. This paper sort of book by Walton Reed and McDaniel. Um, and they talk. I mean, these guys say they're about, I think they say 45 arguement structures and then they have like another 44, sub arguement structures. But then, so we've got around about 100 arguement structures in there at the very most and people tend to disagree with this. But it seems to me that what I need to be doing. In this third part is saying, well, if you think about what arguments are Just the idea that Silicon Valley could sort of hoard argumentative capacity, it just doesn't seem like something you could do. It seemed like there were not that many arguments for any position. You would want to make So I'm afraid and again so the people have attended online this is more or less. What I have here and this is the the third this third part is what I'd be interested in hearing what people think of But at the moment, I haven't got that much more to say about it. Um, Because each schema is a finite then. Yeah, we see the same patterns repeatedly. Um, And so we're not. Yeah so I can I just don't see how this is. How A Marketplace of arguments could be flooded. Okay. And of course, No. This is just a little footnote which is saying yeah, even if Uh, there are limits number of arguments schemas to use. This doesn't sort of rule out that non-reason-based propaganda such as rhetorical or emotional training could be used. And I'm afraid that is it because I knew it was going to be a short presentation. So, I gave you two other mini presentations at the end with the other papers I've been working on but if we want to talk about this for the time being, maybe this would be a good place to stop. And again I'm sorry it's a bit thin. I've really tried to find something interesting to say about propaganda. And this is, as far as I've got so far. That's it. Thanks. Maybe, can I ask if you can, maybe just somewhere because I'm not sure I've understood even if it's short. I, I'm not sure and probably others. So if you can just watch McKenna saying, what's your objection? Just a sort of clarification question. Okay, I, I, I that's true. I, I, so that's also a way of making it. Bad side of the clarification. Question is, what were you talking about? Mckenna. Okay, so we are living in an age of extremely powerful elements. These llms are now very, very good at logical reasoning. Okay? You can look at these benchmarks and you can see these they will solve, they'll always do better than human reasoners, the average human reasoner these days, okay? If you can make these machines, which are good at reasoning, these machines are good at making rational arguments. Okay, rationally, persuasive arguments Is. There might be a unbalance between the people who have access to this technology and the people who don't If the people, if this is correct, then. Mckenna worries that the powerful people even if they use their rational arguement making machines to create, Correct. Arguments. True arguments non-misleading arguments. He thinks this is still an outcome, we should want to try and avoid because he think he doesn't want the people with all of this processing power to have all of this power to make arguments. Okay, that is his concern. Once people the owners of yeah, the owners or political entities, such as, you know, China or America, or something like that, who could potentially flood the marketplace of arguments with rational correct persuasive arguments, and he thinks this is misleading. He thinks this is. Not a good thing to happen to the world. Okay. And what he's trying to do is he's trying to balance this thing. He's saying, look, they're not misleading by deploying all of these arguments, okay? Because these arguments are rationally, persuasive, they're correct. They are sound, they're based on true premises, okay? So then he says that they're still deploying these things, but they're not misleading anybody, okay? Or they're not deliberately misleading anybody but he's saying that's still not a situation, we should want to get. Okay, so my first problem with this You if you'd think of his best case with these people deploying all of these arguments, I think it's very hard to understand what he's trying to say without thinking of it, as, in fact being misleading or in fact being propaganda. Okay, and this is because I think you have two forms of the Dilemma here. If the the people with all the rational powers are behaving as benevolently as he seems to think they are okay, by just releasing rational true correct arguments into the motherplace of arguments then. I don't see what the problem is, okay? If they're doing everything above board and they're not leaving out the wrong arguments, okay? Then it seems to me you just have Silicon Valley or China. Distr Distributing. Correct. True. Rational arguments. Okay, in which case, I don't think it's problematic, okay? But then, as soon as you start sort of removing from this super idealised case, and you start saying, well, if they flood the arguement, if they flood the marketplace with just one type of arguement, it's going to crowd out other arguments, then we're already into misleading propagandistic epistemically defective territory. So that's that's my arguement number one against McKenna, okay. And then the last few slides which is again the thing I think is more interesting but the thing I haven't got interesting ideas about yet. Just incorrect to think. That such a situation could even arise, okay? However, benign or non-benign, and my intuition for that is that Argumentative power or arguments or persuasive power. Is not the sort of thing. You can hoard, you can, you can hoard. Cpu power and processing, power and llm power. But it seems to me Arguments themselves or argumentative capacity. This it's just not something where there's just more, it's not like it's not accountable noun in that, sorry, maybe it is a countable noun, but there's just not that many arguments you can make, which interestingly different to each other in favour of one point of view or the other And my preliminary work on this is looking into what philosophers have said about arguments structures and arguments schemas. Okay. And as far as I can tell, People do seem to think that there is a limited number of types of arguments you can make, okay, even if they disagree over how many arguement types there are. But if there is only a finite number of arguement types around there, then this adds to my suspicion that McKenna's wrong to think that argumentative power is the sort of thing that can be hoarded. Is just not that many arguments to make if you see what I mean. So, that's the entire presentation. It's clear, okay? And again I'm sorry it's a bit flat, I was far more interested in typography. Thank you, thank you for your talk and please don't apologise for it. So it, it strikes me, there's something very odd about talking about a Marketplace of arguments. Um, I think Marketplace of ideas is already a bit of a stretch but I can accept it but Marketplace of arguments because the arguments are secondary to the premises or to the ideas, right? So could you maybe talk more about that? What is the same word? I mean I do I do have that same worry. I don't think it's So, I mean the idea with the premise is Because I It's easy to think of. Arguments of any type. And the same arguement structure even the same arguement structure being applied to try and make the same points. Sure, a different point. And there are a million difference of variations on that because you can slightly change the premises or new premises are discovered. Would imagine, right? So, you can say, well, we now have new empirical data showing us that this political intervention works or does not. Okay. And so then you can say, okay, well now it's a slightly different arguement because we've changed we've added a little bit more evidence in there. Okay, but what I Is gonna work that well. I don't think that really saves the. I didn't think that makes the marketplace of idea just has more and more arguments. Sorry, I don't think the marketplace for arguments means, it's going to have more and more arguments. If you just change the premises slightly, it's still going to be the same arguments again and again, okay just sort of tweaked or with different things brought in. Exactly. So I mean, maybe I need to think harder about the premises, maybe that is actually a problem for me. But I'm more. Yeah I was more thinking about what we if you think about what politics should say on the TV now and the sorts of arguments they make Okay? When often, when you hear them making it, you've heard something similar before. Okay, they may have slotted in slightly different premises or slightly different justification but they're kind of making the same case that they're always making for it. And so that that's Instinct is what makes me think. The McKenna and the marketplace of arguments is just kind of A non-Starter from the yeah. Non-starter Oh thanks. So I also find it quite odd to talk about Marketplace in this case, but And I guess that I, I made the other questions. So, if can you unshare the, the presentation in this way. Maybe we can see. Yeah, the speaker better. Yeah. Great. Cool. Um, so yeah, I guess that like, mine is a very naive and qualification question basically. So, but, uh, can you can you give me the four conditions for, um, First time into account. It's propaganda, please. It's fun, it had to do with intention, right? Um let me go back to I can see my screen with me for you cards. Oh um Okay, so you have Atomic defectiveness. Political intent, social targeting and awareness. And let me just put those in that chat as well. So, just so other people can. Yeah, so so I guess that you can imagine several scenarios in which is this condition, two and four. Are not satisfied by the sort of Marketplace domination case. Um, And I was also thinking that it doesn't have to be necessarily the case by presenting an arguement in favour of something. You are Automatically, you're meeting something. So well, but let's assume that, um, So, let's take the the case that McKenna gives Where actually there's nobody intention one, just wants to compete the truth and whatever. So, you were saying, well, in that case, I don't see the problem. Um, I think I see your reasoning but I was wondering whether It could say some, I mean, if if the analogy, if I if I have to take Seriously the analogy with the marketplace. Couldn't say something like, well, you know. Something, which is the counterpart of Monopoly happens here. And so concentration of some kind of power in the ends of the handful like very restricted Elite or maybe just like one institutional organisation and that's bad. That's bad per se because it's not too realistic. I don't know, something like that. So you would have a case. You don't have propaganda but it's still something trouble. That's what he wants to say. I think yes. And I I don't see why that's not epistemic Decepticus, still Because you're crowding out other ideas as it were and if you're not fine, Why why does it have to be epistemic Effectiveness, as defined by these four conditions? Uh, I don't see why. Why, why is it epistemically defective on the conditions? I've just given you. Yeah. So why does it have to be necessarily? Epistemically defective as get away that principles the notion of epistemist Effectiveness that is defined by this. Okay, so let's this is, this is good. By the way, it's good for me to think about these different ways. So imagine in the future. The political theorists have discovered without a shadow of a doubt. That socialism is correct. Okay. It is. Socialism is correct. Um, okay. And it's, it's just absolutely been proven to be correct, okay, now they then start Distributing. Very very, very many. Arguments. Uh, let's say on social media that have been made by Opus x point x or whatever it is Claude 10 or whatever. Okay. And all of the arguments are true, okay, all of them are sound correct, arguments based on true differences. So and they're just they're just informing the world that socialism is true and correct. And without a shadowing doubt, I don't see the problem here. Do you think that? Do you still feel like the, the in that sort of best best case scenario? There is still sort of something politically unpleasant about that? It's not like this. It's not like they're preventing other people saying anything deliberately I mean, I might be giving a very naive reply here, but So, so my instinct in this case would be to say that. There has to be some sort of decision about. What's the best? So what's the best way of structuring a society, maybe having like an open Society that is open to just Different voices somehow is better than just School being the truth? I don't know. I don't know. I don't know if this is correct, but But I guess, There has to be something of work along those lines there, right? So, so the conviction that, um, arguments are for persuasion but But no good arguement per se. As uh, is necessary undefeatable. Uh, and so you have to leave space for ground to replies or something like that. I I don't know. Now I'm just guessing but but I mean I see your point but um, I don't know how far it is analogy with Marketplace goals. Um, But in that case, I suppose that if that's an allergy, could be taken seriously. As I was saying, then one might want to avoid monopoly or something like that. And the reason why I might want to avoid, it might be grounded in the discussion that we've just had, okay? Uh, so, whether some kind of Monopoly is admissible or not, depending on the consequences, maybe monopoly approaches. Well, I don't know. But uh, Truth here is talking about augmentative monofies which is something that I will keep separated. Anyway, he's an interesting Comparison case. Okay, so AI is helping a great deal in The Sciences at the moment, sort of discovering new types of protein and stuff like that. Okay, so he's a, he's a sort of imaginary scenario where Ella basically llms are used to deploy Correct their rational true arguments about science and health. Worldwide. Okay. Would would we be feeling this worry about? Non-Pluralism here. If it's not, this wouldn't be propaganda anyway because it's not political. But would you be worried? Then in that case, that there's not pluralistic What about the horoscope people? It's difficult. It's it's but I might, I might Um, Oh yeah, so I I guess I see the point. Okay, but that's an interesting thing. Yeah, thanks. Yeah, thank you, Jaime. Great to see you. Nice to see you too. You're looking good. Um, So okay, so I mean a couple of uh caveats first. I was a little bit, uh, Hertz initially when you said I can't find anything interesting. Uh uh published in on propaganda as you can say. Yeah, there is one reason trusting paper. And so I really I guess it's really insults to you. I think it's funny. And but then you really collected corrected yourself by by choosing the right? The right model propaganda. And yes uh obviously I mean I have changed my mind slightly about Mean very, very timely about one condition. But it's it's not it's not it's not important to talk about it and and also I haven't I don't think I've managed to persuade uh constantine Theresa. Okay about that. But anyways, but so let me try. So I have one sort of Devil's. Advocate question, I guess I'm most mostly persuaded by what you're saying. One, two or three very rambling things. So hopefully you allow me to To save these things. They're not super tidy. The one is I mean one is obviously McKenna. The way I understand it, it's not really understanding. The real dangers of AI in propaganda. I mean, it seems to me like the main spies. I understand seems to me. The problem is not the number. I mean, I don't think I don't think AI is especially good. I mean, this has got to do with with my current project. On on the Aesthetics of knowledge I don't think AI is especially good at rush at sort of giving good arguments. I think AI is especially good at the moment at presenting these arguments in a very persuasive way right and but also I mean you should you should look up I guess There is a recent substack by Tom, Stafford on this stuff from the persuasiveness of AI that that you that you that. I don't know if you've seen it, but I think that is interesting because people that dance and testing and and that could be interesting. So anyway, so I don't think that the main danger of AI is could I could I just hear such of arguments onboarding. Could I just respond for a second or that, um, I completely disagree with you about the the rational arguement stuff, okay? Uh and I say this as, as everyone most people in this room, know I spend, I work entirely within AI. Now when I do philosophy and all my philosophy has done, and all my rational, arguments are developed through AI now. But I'm Yeah, I can really smash through philosophical arguments these days in my opinion. But anyway, just to note, Yeah, I mean that that is an empirical thing so and so it might you might be right and my my own use of AI. Is mostly in polishing. The descent this, you know, the way which I present stocks. So it's kind of aesthetic, aesthetic work. Um, not really sort of, um, Uh yeah, our arguement work anyway. I'm just about to start a sub stack of using energy points. Sorry. All right, you should subscribe to the sub stock. I'm about to launch about using it for philosophy, but that's by the bike. Carry on. I will I will. Yeah it's time to me, send me an alert when you do Um, Yeah, the other two. Tiny things are so one. Just one is again translator. It's just to understand that I've understood correctly. Um In which case, I think I agree with with your arguments sir. At least, when your first arguement the second arguement the the finite number of arguments. I think I have a devil's Advocate. Um, So you so McKenna says that they're going to the AIS might uh proud that might well might sort of have uh horde. These arguments I have so many of these arguments out there, push them out. To that alternative arguments are not going to be. As well represented is that is that the idea? He's because he's very much, he also said, but they don't cry out the other ones and I don't see how they can't try out the arguments. If it's this is this is what I think his problem is. Okay, this is, I think he, he's not, he's somewhat ambiguous as The marketplace with a rational, true correct arguments would be and I think, yeah, I think it sounds like maybe you agree with me here, which is It's hard to make sense of that situation without it becoming Of flooding pushing other things out. Yeah. Right. Yeah, yeah. So I so I agree on that. On that thing. On the second thing about the number of arguments So it seems to me like yeah, maybe the structures, I don't even. I mean, I I'll go and look up at this, um, this book that you they they, they quoted from Interesting, there's only 45 or something. So actually, I'm not sure exactly what structure of arguement that means their means, but I found three books. I'll send you the most, it seems to me like Sorry sorry, thank you. So it says to me. No, no, so it seems to me. Um, So, every time Every time someone talks about it in, you know, so this public person raised this issue. Uh, and so You wanna in a particular way. And in order to fight that particular, arguement, you're going to have a slightly different version of an arguement that, you know, we all kind of know. But that's to me that might sound that might count as a slightly different. Um, Arguement or, you know, there's a robbing good whatever, what was the person of the? The person that was shocked by eyes, uh, caring, good or something like that. In there, she got shot in this event. People are going to build out arguments on this event pro and con ice. Let's say and so every time I'm going to fight one of these arguments at the pro work on, I'm going to have to make a reference to what the particular thing that happened in in the good in the good incidents and again that might count as another arguement so it seems to me actually arguments might be. I'm maybe I'm not explaining this correctly, it seems to me like arguments are Migrain level. Um, are actually kind of a law, right? I mean, maybe potentially I don't know if infinite but because the because these arguments are going to have to touch on particular, cases and incidents. And this person said that and this other person said yeah. Actually be more arguments than than you seem to to be thinking. That, that might be the the devil's Advocate. Um, question do the second part. Yeah. Good. Thanks. No useful. And yeah, thank you because this is what I want to this. Third part is what I find. Interesting. So this is I would yeah I appreciate the the pushback here. And by the way, I should say your paper is very interesting when I thought I found it hard to basically. I find it very hard myself to say something interesting about this Um, So the arguement stuff. So by the way, I'm not sh, do you mind if we change the example, because that example is, The cuz the the evidence in that case is just so obvious that I find it kind of hard to to get very right, because you can see it in the video, right? So think about another sort of thing you might politically argue about. Okay, here's a really low stakes one. Smoking in bars. Okay. And so you can imagine somebody before it was banned, you can sort of say, well, here's a rational, arguement it's is unhealthy. It is anti-social, it makes everyone smell. And on the other side, you've got well, you know, libertarianism and, you know, it adds to the atmosphere or something like that. How could that? Political, disagreements. Get that many more arguments which are not just variations on the ones. I just gave you. Okay, so you could say okay, the health one You might change it that arguement might not work anymore. If we discover that secondhand smoke is not as bad for people as we thought. Okay, but it's still, you know, it's the same arguement but it's just got new premises or new additions to it. Okay, well and you can do the same with the other, the arguments on the other side and maybe this is just a lack of imagination on my part, but it just seems to me with lots of Political questions, either low or high politics. There's not that many different rational arguments you can make in favour. I don't know. Have I convinced you at least a little bit? Not really. No. I mean so the the Sorry, it's just I'm not sure their name so I mean people are not. So people are what's really important here is what people are persuaded by, right? So no, not the logical. Uh, you know, Ivory Tower ideal rationality type of uh, arguments, right? So, people are persuaded. In lots of different. I mean I'm not you know, I'm gonna say something that I that I that I feel is right but you know, Maybe, maybe I'm not not a particular super expert. On this but it seems to me like people are frustrated by case cases emotions particular. Uh, situations and what you know, especially with ethics. I mean I've always had a very And some sort of the leaning towards particularism in ethics seems to be like, Even slides reference, even references to particular cases in these arguments, which to you count as slightly different versions of the same. Arguement might actually be, uh, very persuading, uh, Action people even though maybe from a, you know, a kind of Uh, yeah, from a philosopher's ideal rationality point of view. Uh, they might count as well. Okay, you just said the same thing. Well, actually, not really because it's a person. Might be persuaded by this like different version. Making reference to this particular case. Yeah, so say, you know, this person entered into the bar. Had a transformative experience from Um, I don't know feeling the, you know, smelling the the smoke Uh, in the bar and that made. Uh, I don't know. Discover new a new chemical component in which you really want. Uh, lose out in all of these cases, in which people discover, you can components by Having these transformative experiences. I mean, I'm making this way more outrageous than it needs to be, but you can find things that are more sensible. Anyway, I'll shut up. Yeah, thank you for that though. And, um, your new project, sounds cool. So, uh, yeah, I'll have to catch up with you about that. Okay, Miss Michelle. Um, thanks thanks. Nick, can you hear me? Uh, yes I can. I'm just seeing if I can click on the screen. Oh yeah, there you are. I can see you now. Yeah. Hey, hey, hi everyone. I guess I I by now I I'm gonna probably repeat uh, the first two questions you got in a slightly different Framing from someone in the room and from David. But let me give it a try and then at least you, you can help me understand a bit better, what's going on? So the the first thing is, About the weather, arguments can really be something you can have propaganda about and And the, the thought I had was that one of the main reasons, why you? I mean, what you, you can draw a distinction between, what what's um Propaganda. And what's educating people by? Uh, by saying something alone, I mean, this is pretty rough, but the idea would be, I mean, if if you aim at under, I mean, if you use arguments to, sorry if you aim at an education, what you want to do is promote someone's understanding of something, right? And and if you aim a propaganda, if you want to make what you want, is making sure that people form the belief that you want or whatever. Yeah, but it's really hard to think that providing arguments. Easy way of. Obscuring the education name. So I mean, it seems like it arguments almost by their nature if they are good arguments. And I mean broader understanding, so I guess this is along the lines of whether arguments can really be uh the subject of Of propagandistic games. Sort of anything like that. I hope I managed to make myself fairly fairly clear. Yeah, you have but I think we agree. Um, yeah. Because Um, So first of all, I think, I mean, maybe Philippo can correct me here, I think. Perfectly. I think arguments can be utilised in propaganda as in rational if they stand out. Bye bye. Okay, but yeah, and again maybe Philippine great. They would be propagandistic if I deployed them. In such a way, in a sort of, A less than honest way. I, you know, I didn't tell you about Alternatives. I didn't tell you all the information. I didn't tell you all the facts, something like that, okay? And that I think would be propagandistic because it's epistemically defective. Okay, so I think we agree there, then in that case, maybe I have missed. Okay. No, no, then there is no disagreement. Yeah, yeah, no, disagreeing whatsoever. So yeah. Then I, I, I just probably misunderstood what something of what was going on before. Yeah, certainly my other question, we had Yeah, the other question I had was about Um, the dialectic of the And probably it will turn out to be again quite difficult question but the dialectic of your response to Robin and McKenna and So the the, the quick way of of putting, the question is I I didn't get whether Robin signed up signs up for, um, Filippo and others definition of propaganda. And if not I wonder whether he can simply drop the epistemic effectiveness condition and and get out of the Burst horn, the benign horn of your dilemma. So, he could just say, So, yeah, I think I understand. Could you could you give me a bit more detail? So, the idea would be, I mean, if, if McKenna doesn't sign up for an account of propaganda that involves epistemic defectiveness as a condition, he could go. And here, I'm getting back on David's line, I guess. Um, I mean messages information that is not defective at all and is still propaganda. So, what I'm wondering is, is it you that you're signing up for English definition of Philippine colleagues definition of propaganda or even Robin accept that definition? Because if he doesn't, I suppose there might be some other account of propaganda that doesn't involve epistemic defectiveness as a condition. This is where we're getting to the limits of my knowledge about the propaganda literature. Um, so Robin McKenna has written about propaganda specifically in the paper that I'm working with. However, He doesn't say completely explicitly, but he very much seems to imply that I had some quotes on him on the slides. He was seemed to be thinking that this is there's nothing dishonest or untoward in his best case scenario. Which she still thinks is problematic. Okay, so he seems to be thinking that his problematic case is not a case of propaganda still. Filipe has a finger and he might be able to help me out. Do you mind if I ask him to what he thinks right now? Yeah, yeah, that's true. Philippo, can you help me? Um, Filippo. Can't hear you. I can't hear you. Anyway, okay, let me get this. I think pushing technical progress is problem of ours that his music still. I think his connection. Yeah, I think he's feeling Michelle. Can you just speak? Say something just to see whether the problem is our connection or Filipus You can hear me, right? Yeah, yeah we can. You can, we can hear Michelle perfectly but we cannot, we can't hear Filippo, but I don't even know. Felipe was trying to speak. He's still muted. He said, finger it appeared in the chest. Philippo from five minutes ago at this point. I don't know what to do. Um, Maybe in the while working Philippines. I could see as you look right. Not yet. No, he doesn't know. Uh, would you mind if we come back? Mikhail, would you mind if we come back to you once? Uh, Philly pose. Whatever it is, he's going through. So, um, I'm I'm agree with you. When you say that. A limit case of arguement that you can do. Is one kind of topic and any day, a little case of type of arguement, I think that and type of album that are relevant. When you say rational hardness because Russia Island became live hiding in a type of hybrid. But I think that maybe the point. I don't know. I don't know because I don't read the paper or kind of love. Maybe what is? That is not a problem of kind of arguement but tokens arguement so propaganda. Maybe it's So when I make an arguement in a specific place in a specific time, so maybe people change our mind and take some choice because the experience of that particular arguement in that Particular space. Make them change their mind because the product propaganda and the topic of programming is extremely perfect because The people use programmer, in many different ways. Usually I I prefer the Bernard ways of described propaganda, so it's not Only political arguement, but it's more. Okay. So I I think that maybe It's interesting to to make this kind of distinguishable difference from types. Are women token, arguement and maybe could you explain why? That's could you explain? I understand? Why would we want to be making a distinction? Because if Paul can argue like are relevant, maybe EI can have a market Dominion. Used too many too many token arguments soon. Maybe I need some specific arguement but not type armour. Not talking about so it's like personalisation, right? Yeah. Okay. Now we see what you mean. Yeah. Yeah. Okay. Um, Yeah, I don't think I disagree with that but then Did I? I don't know. I start to lose my grip on whether that's So I I wonder whether that's what McKenna's trying to get at, when he's talking about hoarding this sort of argumentative power, Okay, because Rational persuasion. Okay, so maybe that doesn't work then maybe but maybe you do have to add the personification thing. Personalisation thing in there as well. Okay, you exclude my arguement but then the trouble with this, sorry, it's a good point. The problem with this, my paper is it makes my paper far less interesting because people have already written about the personalisation of propaganda via AI, so good point, but thanks for nothing of that. Something I understand anything, what's going on? First of all, what's the role of AR is it supposed to be that the leaders of the world, use AI to get a good arguement? And then with all their money and power, they use that arguement to direct public policy or is it that as Leaguers they get to fiddle with the AI. So that when average people use it, the AI gives them, a particular kind of arguement also. No. That they use their AI. To generate a very large number of persuasive arguments. And what can they do with them when they get they distribute them over the internet or something like that? Okay, they just would be the mode of plot against each other. Yeah but I mean AI isn't going to change what we have now. I mean the idea would be the AI is allowing them to make really good arguments. I mean the extra The arguement is going to be lost. This is the paper, this is the paper. This is what I'm saying. He could already employ philosophers to give Philip. So this figure from these arguments and they'll have as much or as little persuasive power as what? They get from AI. Which I get with the worries and maybe I'm making this too easy for myself, okay? Because, but it is a paper by McKenna but um, I don't know, but now we agree but that's that's the idea that they're just sort of put yeah, pushing out correct rational arguments across the world and they're making these arguments via AI. And so, just try and think about Sort of Devil's out over McKenna. Llms. Are good at deploying arguments at at scale and can come up with quite complicated arguments, very, very quickly. Okay, so there is there is something to it there. I don't know. But I think we agree. Got your. Thank you. I was wondering, do you have any specific reason why you focused on arguments rather than images or pictures? Because major May can produce very versatile photos that can be used in 25 discourses. Uh, I agree. Yeah, yeah. So I was just like I just couldn't think of any interesting things you just said the idea, you know and then Be interesting to say about images and propaganda and AI because yeah, they can make loads and loads of images. Yeah, the problem is not necessary related to arguments can be created. Probably speaking everything that said hey I can produce and use the in the propaganda context so I guess I guess you can do. There's a couple so yeah, you can use AI to produce vast amounts of propaganda. Okay, at scale. And as the vertical saying personalised, okay? And it can be images that can be text and stuff like that. But then, what I was trying to get at now, by the way Filippo, I've just seen you, we'll come back to you when Rico is ready. Um, What McKenna is saying is he's saying, no, no, this is other thing to do with, which is not propaganda, but is still problematic. And what I try to do in this talk. Um, is to say, first of all, it is still propaganda. But second of all, Maybe the worry, he's trying to describe is misplace in the first place. The stuff about using images to make propaganda. I agree. And you can also use llms to make text propaganda. But I don't know what else to say about that philosophically other than Yes, you can use them. Sorry. No. It's okay. Before going back to Philippe was I have a follow-up on someone point, and I think it made you were considering contrast cases, most historical contrast cases for instance, a friendly second in Germany was sort of enlighted, the Monarch, so he was relying on Enlightenment philosophers. The best Minds of his age working for you. I'm simply fine. And then yes also control on the Press and the schools, the universities. And so since or I don't know other cases. Marcusa Reyes and was a philosopher. He had very good arguments, the story called the story schools working with him. And again, as a as an emperor, he also had control on the communication things. Or even the Italian televisional television before the the race of Beros con, he saw in the 50s and in the 60s, it was a very good television with a strong control, a strong educational. Power was also somehow Being and bringing garments and somehow controling the arguments. And there was again, people are converted to Echo chaliva. Altimo very good. Minds working in this television. So this is all so it seems just a matter of the of degree and again I find it hard to see what what sees the difference is to be that AI is just covering the whole uh, spectrum of of good arguments. Whereas can't or Roberto Eco. There were just or Marco Reyes were just having a limited region of the space of arguement is just that right at the difference. Uh, but I know since that, again, find tabs to see what what's really new in that, it seems that what the Cure propaganda is rather the control of the, of the media, to spread the arguments, rather than the control of of the the instrument producer commands. That's the idea that there are these two two layers in in, propaganda producing, good arguments. And spreading having power to spread good arguments and social AIS, is more about the first thing, But the games I really kind of what is specific of AI that in principle one may have a, a lot of University professors and scientists working for them and seems to produce a quite similar effect in terms of logical law in terms of rational arguement generation. Yeah. Scale and the number. And the variations you can do So I mean now it is quite easy I think for students to to cheat on their, their reading philosophy, right? Because you can copy the paper in, and you can say tomorrow in the, in the group, with Enrico, I'm going to have sort of five arguments against this case, give me five arguments generally, I'll give you some pretty good arguments now, right? And so, I've decided what I think is correct. I've just you, I've gone to an AI and said, give me some good compelling arguments that I can use to push my case. And I think what McKenna is getting at is That's perhaps vast vastly large Escape. Okay. So like I said I was going to say Trump has a political ideology but there's no ideology. Think about George W, bush and neo-conservativism or something like that? You could think of someone like him saying. Well, I'm an idiot. So I can't argue the case myself but I'm going to go to a good AI, hey, AI, give me Rational persuasive. Correct arguments. To argue in favour of neoconservatism. And McKenna would say that that would not be a good state of affairs because George W. Bush has all the power to make his case and there might be equally good arguments against neoconservatism. But because the people who believe them don't have access to the new GPT then they don't do this. But again I don't see it. Seems that the the big power is more in the control of social media televisions newspapers, because once the the Also, because it's unless, so hypothesis is that Trump may have a better version of chargpt than the Dynamics is that also, the point that's completely. That's what you think they use more powerful. Potentials. You're holding the argumentative power. Yeah. Okay. And you don't you just think that doesn't work as well? No, no, I guess I can seem better now. It seems feel that this is a sort of neck but Bottleneck. Because it's even if you have all these arguments, the social media, or the media, the time of the people not allow them to, to So you always need to just to to spread some arguments that these totality seems even beyond the Simpson to me that the big thing is to to have control on the on the media rather than on the generation of arguments. But Maybe this is not incompatible with this point, it just yeah, it's not denying that that the media also are important. They think that this is a further Dimension, this possibility of control on the whole space of arguement can can give further true. But again he's so Um, vague about exactly what his case is going to be. Okay, because as soon as you say, A company has control of the arguments and and suppresses other arguments then moving to propaganda territory already. Yeah, just this is more to conclude and Victoria point I think examples may help a lot in the paper. Maybe you have even in this discussion in in the presentation I should have given more experiments examples. Otherwise it's very hard to understand what's going on. What this is the smoking example or other other all contrast cases, I I think that that can can help to have a. I don't know if in the McKenna paper, there are examples, so not really. Have to be finished until December. I'm I'm a week later. My 800 word extended abstract so, this is very early stage. Um, because I've only wrote like six papers on propaganda. Yeah. So And we can go back to Philippe pass. Yeah, Felipe. Still there. Yeah, can you hear me now? I can let me see. Okay, yeah, sorry about that. The, for some reason, maybe, maybe the the powers that be the llms controlling everything, uh, stopped my computer. And so that it froze and I couldn't speak. Um, I guess I wanted to say the thing about maybe in case itself, so Robin I got in touch with Robin McKenna. Well, well somehow, I think, maybe a year ago on Twitter, we we spoke uh, He was reading our paper. But didn't seem and he wanted to write propaganda and I I haven't caught up on whether it's published anything but it didn't seem particularly convinced by by our people. So it may well be. Uh, as Michelle was saying that um, he doesn't subscribe to our Theory. He's, he's written a subject about I believe he's written a sub stack on propaganda recently, which maybe mentions your paper, I can't remember. No, he doesn't. I I just looked probably, doesn't the, the native anglophones tend not to site us. So the, the problem is that, um, The problem is that even if You know, unless he's explicitly connected to a theory of propaganda, Uh, you can always say, look, this is the, this is the correct Theory and therefore um therefore you're wrong. And then but if and if he does submit sorry, if he does commit to a specific uh Theory program that I guess he has to show you that um, his his views are compatible. With these few and not with your sorry, they're acceptable. According sorry. They, they don't make it count as propaganda, according to his viewer propaganda, not yours, right? So, all of this work if it hasn't been done, from from the purposes of your paper, I think, um, Is this work hasn't been done but I I can't tell you for absolutely sure. Then then your paper still is still golden. I think. Okay. So Thanks. I'm the first thing to say is So, I'm when I'm When I'm criticising McKenna, I'm not trying to say, aha, you're hoist by your own batard because this contradicts your your view on Paganda what I am criticising him for though, is him thinking that what he's describing as the best case in this paper, I guess a even conceivable. And be conceivable in such a way that it avoids. Epistemic Effectiveness. Okay? Because the way he writes in the paper, he seems something he can do this. And that's the first point. Okay, I'm trying to make that it falls into it and then if you want to choose your, the sort of the the view of propaganda that you endorse, then it comes out as propaganda. Okay? That would be the step one. Yeah, what I'm more interested in is using that as a springboard into the more interesting. Second part of the paper about arguments counting and whether it really is something to think about. But you know, I need to read a bit more McKenna and I definitely need to read more about propaganda. It's okay. Other questions. Thank you. That's great. Thank you. Thank you. Cheers. I still have something related to what I said before is the analogy with lawyers can add. For instance, lawyers often provide good arguments in favour of their clients and and usually powerful people can entire bad. Best better lawyers. And can, this can create a form of Justice. Yeah. Because they, they are more more chances to, to, to, to, to win a trial just because they are the best lawyer. And so this, the the mechanical points seems to be something like GPT is the best lawyer possible? Because we produce the best arguments ever for whatever thesis you want to defend. Then the diseases can be epistemically good or not but that, that seems Dr. One of the no idea one, but in any case you, you are regardless of, whether you are innocent or not, you you are the best lawyer, they create the best arguments for your. And and so that's that's the point seems to be that that um Just not lawyer at the political lab so you can have The best people who who give Darkness for your your clients, you always, The the trial. Yeah, he he talks about Because of the name of the paper is sophistry on steroids. Yeah, exactly. And the idea is that they were a lawyer. Yeah, exactly. The idea is, you could hire a lot of sophists and he's saying that Yeah, but that would be the, the sort of The very low level way of understanding his example, right? Imagine if Donald Trump, just employed all the best soffits in the world, would that give him, but then promise to use them to make true arguments correct? Yes. Would that be politically? Yeah, undesirable would be the question. And so you can just I mean, you can disagree there that it's I mean, does it change at all, if you swap out llms for soft armies of sophists, Not really, I think it still comes out as propaganda. It still comes out. Hard to understand that you can have that many arguments towards the same conclusion. Kind of just follow up on that. Okay. Nobody said anything about this but I wonder if if at the back of people's minds for not wanting to pursue the analogy that Enrico and I was saying myself. Why isn't that just what we've got? Is that I mean, despite A fact of people's corruptness and weakness and so on, if you have an army of lawyers or philosophers helping you, they still are people. And we think they've got a conscience and a moral set. And now, LM doesn't have any of that. So, in theory, very long limits to what it'll argue for, oh, completely, he mentions this in the paper as well. He says, the llms don't have any skin in the game. Yeah, although that again, that's maybe actually a point to argue against I'm not sure. It's hard to I don't know. But the difference is also that if even if you are the best philosophers, there is still some chance for others in the debate to, to counter arguement against that. Whereas, with, with this superpower for La Land, the idea is that they just, it's a bit like playing chess. They always win. They are so so bad. And so in this case, there is the control is full. Whereas with philosophy, even if you are the best lawyer, the best pillars, if it may be that like in Hollywood movies, the the, the poor lawyer and Steve fans some some way of fighting against. Yeah. Yeah, some chicken. Yeah, whereas in that case, is the McKenna scenario is that they just add the certainty of winning in the game of arguments and so that, that's great decade. So that's a nice big analogy by the way. Difficult in understanding your presentation of mechanics is you have the two cases in, which the the arguments are are sound, the premises are true and let's say also the aims are progressive or liberal are flourishing of humanities promoting Justice. And yeah. And that that seems the interesting case, right? Because the other case it seems really There is no problem. If the Trump case is, in the case of propaganda, which is strengthened by the use of any land. So, in that case, I, I don't see why why one should disagree. So the the critical case the the controversial case is the case in which are using this way to to saturate and dominate the space of arguement, but Epistemically morally. Everything is where the the intent are noble. The the Yeah, the the fact that the epistemology is correct, so that, that's, that's the problematic case, that's the case in, which there are where you are a mechanics disagree, right? He thinks that that that's problematic case. And you think it's not I've lost trackings out in which case we're referring to specific case because I really don't see where the problem is in the fact that just instead of just using television and social networks to to do propaganda in all the, the negative senses that filling post paper, are I liked, they also use a AI to to have arguments in favour of A racist claim. Or get politically aggressive. So in that case, Is just propagandas traditional propaganda strengthened by by AI, which offer more and more arguments in favour of of false. Morally, bad claims. So they're not me not in. Okay. Yeah. Okay. Yeah. The the case that seems create tension is the case in which uh, the premises are true, and the aims are politically neutral or even uh positive and morally valuable. And that's, that's the case, uh, that, that seems to to, to create different. So one may say, oh, this is still propaganda because you are, you are all to go dominating the space of arguement. And that's a kind of propaganda or something. Anyway, even we don't want to call it propaganda, but it's something bad. Yeah. Whereas your point is that not that in that case, Not in particularly bades, I think is happening. Yeah, I think if you idealize too much, everything's great, sounds good. I mean, if you move even slightly away from the idealised case, it turns into Epistemic Effectiveness propaganda. Yeah, okay. And then the second arguement about whether this is even a considerable situation. So, but in the, in the game of development, I think there is one more problem because Okay, yes. So we have a political arguement, the AI produced a good arguement. So premise, premise, a true conclusion, true screen. So I can, what can you can we see space is up to you to like that Angelo but we're better speaking, what I can I can see the the icon. No what what Man in the room here, Roberto speaking. And I was thinking about York. I was completely organised. I've been to lock effect. I understand, you know, and so we are a natural political arguement and so There could be a real game arguement in this kind of examples or else can be better. I try to explain it better. So You cannot you have a destinic problem about good afternoon because you can in some sense cannot produce. Are too many good arguments because some are maybe are good. But people cannot really Uh, understand the arguement because maybe they are too difficult. They are the structure of deductionary to Probably maybe there are perfect, arguement maybe better arguement, but In the practise of the aesthetic practise, they are not with armed because people don't understand it. I think the main cases in the Mathematical issues. So maybe a really good arguement but not every people can understand it. So only a few people maybe we, I produce good arguments. And that's actually that's an interesting thing to think about, actually, because You know what I mean? By the Monty Hall problem. No, sorry. How can I explain the modules? It's really hard. It's like a logic puzzle, like almost like a riddle, I'll show it to you afterwards or something like that, and it's one of those things where it seems so the wrong answer is so incredibly intuitive. It's extremely hard, but I don't know whenever I understand why why that answer is wrong and the right answer is right. I feel like I can only keep the explanation in my head for a set amount of time before the other one. So maybe there's something similar like that. Are fighting thoughts. And again, hopefully taking me into something more interesting. Is, there's also, Is it move 37 in gold? Do you know what I mean by move? 37? In go, the game, go and move, 37 was the when it beat the world champion and he did this, extremely eccentric move, which nobody understood. And everyone thought it made a mistake and afterwards. They realised. It was this. Yeah. Mind-blowing move. So maybe yeah. Maybe that would also be an issue about having argumentative power. Arguments, which make no sense to the average person. So no that is that's good thought, thank you. That's And Angela was an excellent. Yes. Um, yeah. Yeah, thanks. Uh, thanks Nick for for for this uh presentation. Uh maybe I'm I will just I will repeat the questions that I have already been are asked about some The the analogy with this office, maybe again like made me again now under think about how like the problematic one problematic aspect of. Um, Uh, floating with arguments or this kind of stuff is like. And maybe Michelle mentioned, something like that. But is it isn't it? The case that I mean, what one of the problems of propaganda is not just Quality of beliefs, the propaganda offers. But also the thing that maybe I don't know. Propaganda seems to try to tamper with. How people are reasoning rights? These are just uh kind of you know, limiting the way one could reason about an issue or, you know, limiting the possibilities of how the issue could be framed or argued for stuff like that. So, Uh, I can see why one would would want to say, I mean, what's the problem when this is the best case scenario when everything is correct and and stuff like that, but uh, I was wondering like if you just give people and let me try to rephrase that you were, you were kind of, um, Is where somebody goes to charge PT asking for. Uh, reasons or arguments for whatever certain conservative political view or whatever it can be socialism, whatever, right? I'm not taking sentences on the, on the, on the, you know, left. I'm just, I'm just trying to think. It seems and I'm not sure this is in in McKinney's paper. It's just what I'm trying to reason about is like uh unless you are a philosopher expert or whatever, like lay person might use GPT. To, you know, try to Have, uh, you know, I try to get ideas, you know, like uh, it's some sort of saying, hey, give me an idea on this, give me an idea on that. So if we only have this, you know, the similar kind of arguments This might might be, you know, problematic in the sense. We just give, you know, ready-made idea to, to people. And this doesn't really help with, uh, between the sort of time, it's like, Mike Thumper the the way people reason about stuff uh, Say it's maybe the variety doesn't even really change how people interact with llms and agents and stuff like that, you know. But so maybe having Uh, you know, this flood of same arguments, might be pejorative in one sense, right? So Do you see like the problem if propaganda could be sort of framed, like that? Another question, what could I could I ask you to go a couple of steps back? I'm not quite sure I follow. Yeah, so I was, I was wondering again, like is one aspect. It's like one problematic aspect of the way we can. Um, see propaganda is not just about the, you know, the truth kind of falsity of beliefs, but the fact that they kind of inhibit or tamper with the way people reason about stuff. So it's not just about uh, Paper talks about rationally, persuasive arguments, but it seems to me that the way Every every like common common people, like if you're not a philosopher, you're using charging PT, it's not to be rational rationally persuaded but it's more likely to have a ready-made answer of something, right? And so if you just get a tons of the same, Made answers maybe like this doesn't help you because it doesn't help people. This may be, you know, Might might make, You know, might make worse the the, you know, or Might not help people to kind of reason about the fields they are trying to look or search for So, so is it is your your worry, sounds more about llms than about propaganda, and it sounds like you're concerned that people will be. Oh, sorry, Philippe was going. So let's wait, goodbyes. Um, thank you for the comments, by the way, again. And um, your paper was great, by the way. Sorry. Um, thank you. Okay, and your project sounds cool. So hopefully we'll speak soon. Whenever you want by. Um sorry Angelo um so it sounded to me like you were saying, it seems to me that you're more worried about people. Not thinking for themselves because they're using llms. Was that part of the worry? Otherwise, I'm sorry, I'm still not quite following this. I can't hear you now. Really. Oh it's it's the battery is things are still on. Wait a second. Sorry. Any better. Should be muted because that might should be picking me up instead. Yeah. We lost the camera in the room as well. It says, yes. Because that's not a promising. Can you hear me now? Right, just two seconds. Still good night. Okay. Anchor work. Start. Any better. They just said, yes. Can you hear us? Yeah, you can hear me? Yeah. Okay, sorry about that. Okay, so And so that's what I understood. You sounded like you would worry people. Thinking reasoning themselves and using an allo. Instead, I was sorry not to follow. Yeah. Not not just in a general format. Yeah, I was Oh you gone quiet. We can't hear you. I'm doing this. One of the issues we we might think is Troublesome is also for that reason, you know. So you cut out, it's hard to hear. You hang on. So ah yeah sorry can you hear me now? Now? We can. Yeah. Yeah, my question was more. If we can understand one of the problematic aspects of propaganda being also, like aiming at not making people, you know, reason too much for themselves. You know, so tampering disability, you know? So, Giving them false beliefs, or whatever, you know, charge believes about whatever, but also kind of sucks maybe in a subsequent manner, or whatever. So, but I'm not, I'm not kind of arguing for that. I'm wondering if that could be one of the problematic aspect and maybe there's also happens in. I mean, I'm not sure in the best very best idea is you know, case scenario. But maybe We've lost the sound again. There you go. Normally what's happening in this globe? It's connected towards around. The internet connection here is bad. Probably Yeah, before he worked just fine. So maybe it needs a little bit of time to adjust, sorry, but I don't know, so it's still not working. Can you hear us? Angela? Can you hear me? Sorry. Actually now we can hear you as well. Do you want to say, try for, like the tenth time to say that was again, I was saying like, I'm not really arguing for that. I was just putting whatever, like maybe a suggesting to say, maybe one of the problematic aspect might be framed in this way. So uh, yeah, yeah. Okay, so all I was gonna say for that is I suspect that what you're describing, what it sort of some sort of political actor, sort of um, People's capacity to reason. I suspect. At least Felipe would say that that doesn't count as propaganda. Propaganda itself has to be yeah, sort of forming beliefs to an epistemically defective route. So I don't know if there's another name for that of sort of. Yeah. Trying to sort of yeah. Subdue people. Capacities. But I suspect it wouldn't count as propaganda. Yeah, sure. I, I need to, I think I've read something about this, so it's not my idea, you know? So if I find the paper, or if I recall, where please, I found it, maybe I can send it to you, please. Maybe it's not, you know, it's not convincing or persuasive, but just like, oh no. No, I need as much of my defensive. Yeah, I please send me as much material as you can. Yeah. Thanks for the reply. Uh I ever follow up with this, okay? It's the fact that what Angelo just said me writes, the the issue, whether it's maybe just a a case of a more General phenomenon which is the tendency of AI just to to replace certain human activities. Uh, so for instance, there seems to be a similar problem with philosophy. Is become so good in doing philosophy, just as it's so good. So better than us in playing chess or go what we should do. We can still play chess of God, but then just a minor thing. The big issue will be addressing. And so this seems just another instance of the same thing. If AI becomes so good in doing politics and taking decision and making arguments in the public sphere. Well, that that just with, it's another area of human activity in which AI will replace humans. So the problem doesn't seem to be a problem of propaganda, seems more a problem of replacement of humans in activities that were considered specifically human. Both rather than instead of Would you say so it's not only have we replaced human reasons, you will say that if this happens in philosophy or art or music or chess, or God, that just replacement, yeah, it happens in uh, then that's so super popular. Yeah, as well. It's a new front desk or something like that. Then, that seems to, that seems to be a mechanis. Like, what is your point? Seems to be more, not in that case, at least in the, in the positive. If, if the ideas are bad, then sure, if that, if his AI is used just to promote racist politics, that's propaganda. But if the point is just using AI to have a good arguments in the public sphere, the best arguments available in the public sphere. This seems just like using AI for doing science, you know, so if you're playing God, yeah, thank you, man. Okay, if there are no other thanks, everyone that was really not great. I the next work in progress, I can do will be far better. I've done so much more fun stuff over the holidays. Okay. So thanks Angelo. Yeah, thank you for the questions both of. You see you soon? Hopefully take care. Hopefully good job. Bye bye. Thanks. And now we have yeah, special events. But I I forgot them that a bottle and I forgot it in my, in my Pizza. So I I know about a lot. Yeah. What happened? Actually.com. No. Yeah, let's estás. They will be close to this side of the company. Take a change and get practise and yeah. ### Assistant I need to read the current note first to understand what you've already written, then I can give you detailed feedback on how to develop it toward its CEV. ### Assistant This is a rich situation. Let me give you a thorough analysis with multiple paths forward. --- ## What's Good Already The draft abstract is genuinely solid structurally. You have: 1. A clear target (McKenna's "best case") 2. A crisp dilemma (idealize fully → benign, less idealized → propaganda) 3. A second, potentially more interesting move (can the marketplace actually be dominated?) 4. A tentative, non-overcommitting conclusion The seminar feedback reveals that the dilemma works as a dialectical move—people pushed back on various grounds but didn't refute it. --- ## The CEV of This Paper The best version of this paper does three things: **1. It makes the dilemma argument *structurally unavoidable*, not just rhetorically compelling** Right now the dilemma says: "If you idealize fully, benign. If you don't, propaganda." But McKenna could respond: "I idealize *content* (true, sound arguments) but not *distribution* (concentrated actors flooding)." The CEV sharpens this: *any* feature that makes flooding troubling (crowding out, distorting salience, licensing false inferences) is also a feature that constitutes epistemic defectiveness. It's not that McKenna's scenario *becomes* propaganda when you de-idealize; it's that *the troublingness just IS the defectiveness*. They're the same thing under different descriptions. **2. It develops the finite-schemas point into something genuinely novel** This is where your instinct is right—this is the more interesting part. The current draft gestures at it but doesn't develop it. The CEV would: - Take seriously the Walton/Reed/Macagno taxonomy - Distinguish argument *types* from argument *tokens* - Ask what "flooding" means if schemas are finite - Consider diminishing returns on repeated exposure **3. It points toward where the real concerns are** If McKenna's specific scenario is questionable, what *should* we worry about? The CEV doesn't just critique—it redirects. Options: - Personalization (token-level tailoring even with finite types) - Rhetorical/emotional packaging (finite arguments, infinite presentations) - Attention capture rather than argument dominance - Speed/scale of response to breaking events --- ## Concrete Options for Improvement ### Option A: Sharpen the Dilemma (Low Effort, High Return) Add one paragraph that makes the dilemma more explicit and harder to escape. Something like: > McKenna might respond that the relevant idealization concerns *content* (the arguments are sound) but not *distribution* (they are deployed asymmetrically). But this distinction does not save the best case from propaganda. The features that make asymmetric deployment troubling—crowding out alternatives, distorting what appears salient, licensing inferences like "these are the main considerations"—are precisely the features that constitute epistemic defectiveness on the Bonard, Contesi, and Marques account. Flooding is not a separable feature that makes sound arguments worrisome; flooding *is* the defectiveness. The campaign-level communicative act misleads through omission and salience manipulation even if the atomic arguments are locally sound. This makes the point that McKenna can't have *both* a troubling scenario *and* non-propaganda. The troublingness and the propaganda-ness are constitutively linked, not accidentally co-occurring. ### Option B: Engage with the Type/Token Distinction (Medium Effort) The seminar raised this (the personalization point). You could add: > One might object that even if argument *types* are finite, argument *tokens* are not. AI enables not just volume but personalization: the same schema can be instantiated in ways tailored to individual recipients, their particular concerns, their recent search history, their emotional vulnerabilities. Perhaps the marketplace can be dominated at the token level even if the type level is constrained. > This is a genuine concern, but it shifts the locus of the worry. If the problem is personalization, then the relevant danger is not marketplace saturation but *microtargeting*—delivering different argumentative tokens to different audiences in ways that exploit individual epistemic situations. This is a known concern about AI-enabled political communication, but it is different from McKenna's marketplace-flooding scenario. The shift suggests that McKenna's specific picture may not be the right place to locate the worry, even if nearby worries are genuine. This acknowledges the objection, absorbs it, and uses it to reinforce your tentative conclusion. ### Option C: Develop the Chess/Go Analogy (Low Effort, Potentially Interesting) You mentioned move 37 in Go in the seminar. There's something here about arguments that are *correct* but *opaque*: > A further complication: AI might generate arguments that are sound but whose soundness is difficult for human audiences to verify. If an AI produces an argument whose validity is transparent, audiences can assess it independently. But if the argument relies on chains of reasoning that exceed ordinary human verification, audiences face a choice between accepting it on authority or ignoring it. This is a different problem from marketplace saturation—call it *argumentative opacity*. It raises questions about epistemic trust rather than about flooding per se. This opens a door you might want to walk through in the full paper, without committing you to a specific position. ### Option D: Engage with the Historical Analogies (Medium Effort) The seminar mentioned Frederick the Great, Marcus Aurelius, Italian state television. This is a real challenge: if powerful actors have always been able to hire the best arguers, what's new? > McKenna's scenario invites comparison with historical cases of concentrated argumentative capacity. Frederick the Great employed Enlightenment philosophers; Marcus Aurelius had Stoic advisors; state broadcasters have featured public intellectuals. In these cases, actors with resources controlled disproportionate access to sophisticated argument production. What, if anything, does AI change? > The answer cannot simply be *scale*—printing, radio, and television all enabled scale. More plausibly, AI changes the *marginal cost* of argument production. Hiring philosophers is expensive and slow; generating AI content is cheap and fast. But this returns us to the question of diminishing returns. If political debates have finite argument structures, cheap production of argument tokens may not translate into proportional influence. The newspaper editor who decides to print the same editorial fifty times does not become fifty times more persuasive. This situates your argument historically and makes the finite-schemas point do more work. ### Option E: Flag the Empirical Open Questions Explicitly (Low Effort, Strategic) This is about not boxing yourself in: > Several empirical questions remain open: whether repeated exposure to the same argument schema produces cumulative belief change or fatigue; whether rhetorical and emotional framing can dominate even when argumentative content is stable; whether audience recognition of flooding triggers discounting. The present argument is conditional: *if* marketplace saturation is possible in the way McKenna envisions, *then* the troubling version counts as propaganda. The antecedent deserves scrutiny. This lets you write the full paper once the empirical literature clarifies, rather than committing to claims you can't defend. ### Option F: Point Toward the Real Concerns (Medium Effort, Good Framing) End with a gesture toward where the genuine worries might be: > If the argument here is correct, then McKenna's specific scenario—marketplace domination through sheer volume of sound arguments—may be harder to achieve than it appears. This does not mean AI-enabled political communication is unproblematic. But the real dangers may lie elsewhere: in *speed* (the ability to respond to events faster than human deliberation can keep up); in *personalization* (the tailoring of content to individual epistemic vulnerabilities); in *opacity* (the generation of arguments whose soundness cannot be independently verified); or in *non-argumentative persuasion* (rhetorical and emotional manipulation that bypasses argument altogether). Identifying which of these deserves priority is a task for future work. This turns a potentially negative paper into a constructive one and gives you a roadmap for the full version. --- ## What the Abstract Should Do vs. What the Paper Should Do For the *extended abstract* (which is what you need now), I'd suggest: 1. **Keep the structure you have**—it's good 2. **Sharpen the dilemma (Option A)**—one paragraph 3. **Briefly acknowledge the type/token objection (Option B)**—one sentence or short paragraph 4. **Flag empirical open questions (Option E)**—this gives you room to maneuver 5. **Gesture toward real concerns (Option F)**—good ending For the *full paper* (6 months from now), you can: - Develop the finite-schemas point with actual engagement with Walton et al. - Engage with the empirical literature on persuasion and AI (Tom Stafford's work that was mentioned) - Work out the type/token distinction properly - Consider the historical analogies in depth - Explore the "where are the real concerns" question --- ## Specific Edits for the Current Draft **Para 3 (McKenna's claim about scale):** Fine as is. Maybe tighten slightly. **Para 4 (best case):** This is good but could emphasize more that McKenna *wants* to isolate something troubling-but-not-propaganda, and that this is the dialectical target. **Para 5 (Bonard et al.):** Fine, does the job. **Para 6 (dilemma):** This is the core. Consider adding one sentence at the end: "The troubling features and the epistemically defective features are not accidentally co-present; they are constitutively linked." **Para 7 (finite schemas):** This is promising but underdeveloped. For the abstract, that's fine—you're flagging a direction. For the paper, this needs work. **Para 8 (hedge):** Good—acknowledge rhetoric/emotion, don't overclaim. --- ## Things to Definitely NOT Do 1. **Don't commit to a specific theory of how persuasion works empirically**—you don't need to and the literature is unsettled. 2. **Don't get into definitional battles about propaganda beyond what's necessary**—use the Bonard et al. account as a tool, not as a thesis to defend. 3. **Don't claim that AI-enabled political communication is fine**—you're not saying that, and readers will misread you if you're not careful. 4. **Don't try to solve everything in the abstract**—flag directions, don't close them off. --- ## The Underlying Issue Reading the seminar transcript, I notice that you were somewhat apologetic ("this is my least prepared work in progress ever," "I found it very hard to find anything interesting"). But there *is* something interesting here—the finite-schemas point is genuinely novel and worth developing. The dilemma is a real philosophical contribution. The CEV of this paper is one where you stop treating it as "I have to say something about propaganda for this conference I missed" and start treating it as "I noticed something about the structure of argumentation that people working on AI and propaganda haven't considered." That reframing might help. --- ## TL;DR: What to Do Right Now 1. Add one paragraph sharpening the dilemma (Option A) 2. Add one sentence acknowledging the type/token point (Option B, brief version) 3. Tweak the ending to gesture at real concerns (Option F) 4. Submit the abstract 5. Spend the next 6 months reading Walton et al. and thinking about whether argument schemas really are finite in the relevant sense Vuoi che ti prepari una versione rivista dell'abstract con queste modifiche già inserite? ### User my worry with option F is that I believe these types of suggestions have already been made. Option C is interesting can you fine tune your advice based on this? ### Assistant Good instinct on both counts. Let me recalibrate. --- ## On Option F You're right to be cautious. Speed, personalization, microtargeting, emotional manipulation—these are the obvious moves, and the AI ethics / political communication literature has been discussing them for years. Ending with "the real concerns might be X, Y, Z" where X, Y, Z are already well-trodden makes your paper look like a late arrival rather than a contribution. **Two ways to handle this:** 1. **Acknowledge rather than propose:** Instead of "future work should investigate these," say something like "These concerns are well-documented elsewhere (cite). The contribution here is different: to ask whether McKenna's *specific* scenario—marketplace domination through argumentative volume—is even coherent." 2. **Skip it entirely:** End on the tentative note you already have, which is honest and doesn't overclaim. --- ## Developing Option C: Argumentative Opacity This is where I think you might have something genuinely underexplored. Let me work through why. ### The Core Idea Move 37 was *correct* but *opaque*. Expert Go players couldn't verify its brilliance until later. For a period, they faced a choice: trust the machine or dismiss it as error. Translate to arguments: AI might generate arguments that are *sound* but whose *soundness exceeds ordinary human verification*. The argument is valid, the premises are true, but the chain of reasoning is sufficiently complex (or counterintuitive, or dependent on non-obvious empirical connections) that audiences cannot independently assess it. ### Why This Is Interesting for Your Paper It's a potential escape route for McKenna—and showing that even THIS doesn't work is a stronger contribution than just running the original dilemma. Here's the dialectic: 1. **Your dilemma says:** McKenna's troubling scenario is propaganda; his non-propaganda scenario is benign. 2. **Opacity objection:** What about sound-but-opaque arguments deployed at scale? These aren't epistemically defective (the arguments are good). But they're not benign either—audiences can't verify them, so they must either accept on authority or ignore. This seems troubling without being propaganda. 3. **Your response options:** - **Bite the bullet:** Maybe this IS benign. If the arguments are sound, audiences are forming true beliefs. So what if they can't verify? (But this seems too quick—there's something epistemically sub-optimal about authority-based acceptance of political claims you can't assess.) - **Expand defectiveness:** Maybe inducing authority-dependent belief IS a form of epistemic defectiveness, even when the content is true. (This might stretch Bonard et al.'s account, but it's defensible.) - **Distinguish the worry:** Maybe opacity raises a *different* concern—about epistemic autonomy, or the conditions for genuine democratic deliberation—that is real but distinct from propaganda. This lets you say: "McKenna is gesturing at something genuine, but it's not the thing he says it is, and his framing obscures rather than illuminates it." ### Why This Might Be Novel People have written about: - Black-box AI and explainability (in ML/AI ethics) - Epistemic dependence and testimony (in epistemology) - Manipulation vs. rational persuasion (in political philosophy) But the *specific* connection—"What happens when AI generates sound arguments that exceed human verification capacity, and what does this mean for political communication?"—seems relatively unexplored. It's at the intersection of three literatures that don't talk to each other much. ### How to Use This in the Paper **In the abstract:** One paragraph, after the dilemma argument. Something like: > One might resist the dilemma by pointing to a scenario McKenna does not explicitly consider: AI-generated arguments that are sound but whose soundness exceeds human verification capacity. Such arguments are not epistemically defective in the content sense—the premises are true, the inference valid. But audiences cannot independently assess them, and so must either accept on authority or ignore. This might seem troubling without being propaganda. I suggest, however, that this scenario raises a different concern—about the conditions for genuine rational persuasion—rather than rescuing McKenna's specific framing. If an audience accepts a political argument because they trust the source rather than because they can assess the reasoning, it is unclear that rational persuasion has occurred at all, regardless of whether the argument is sound. The problem is not marketplace saturation but *argumentative opacity*: the gap between an argument being good and an audience being in a position to recognise it as such. **In the full paper:** Develop this as a section. Engage with: - Testimony literature (when is accepting on authority epistemically appropriate?) - Democratic deliberation literature (what role does understanding play in legitimate political belief-formation?) - Maybe some philosophy of mathematics (proof verification, accepting theorems you can't verify) --- ## Revised Structure for the Abstract 1. **McKenna's position** (as you have it) 2. **Bonard et al. on propaganda** (as you have it) 3. **The dilemma** (sharpen slightly per Option A) 4. **The opacity objection and response** (new—Option C developed) 5. **The finite-schemas point** (as you have it, maybe slightly shortened) 6. **Tentative conclusion** (McKenna's worry is real but mislocated; the abstract gestures at where it might be better located without overclaiming novelty on well-trodden ground) --- ## What This Buys You For the abstract: - A second move beyond the dilemma (opacity) that shows philosophical sophistication - Acknowledgment of a potential objection and response - Something that *might* be novel, without overclaiming For the full paper: - Room to develop the opacity point in depth - Room to develop the finite-schemas point in depth - A structure that doesn't box you in: you're raising questions, not answering them definitively - Avoidance of the "here are the obvious concerns everyone knows about" trap --- ## One Risk with Option C The opacity point might seem like a digression from the propaganda question. You'd need to frame it as: "Even if we look for a scenario that escapes my dilemma, we don't find one that vindicates McKenna's specific framing—we find a different problem that deserves its own treatment." This makes the paper about *McKenna's argumentative strategy* as much as about propaganda itself. That's fine, but be aware of it. --- Vuoi che scriva il paragrafo sull'opacità argomentativa in modo che tu possa valutarlo concretamente? ### User I am not sure we are on the same page with the opaque arguments.. was it not raised in the WIP in a different context? ### Assistant You're right, I may have misread the context. Let me check. Looking back at the transcript, the move 37 / opacity point came up in this exchange with Roberto: > "you can in some sense cannot produce... too many good arguments because some are maybe are good. But people cannot really understand the argument because maybe they are too difficult... maybe there are perfect arguments, maybe better arguments, but in the practice... they are not [persuasive] because people don't understand it." And then you connected this to move 37—the AI move that experts couldn't parse at first. So the point in the WIP was: **opaque arguments won't be persuasive**. This is a *further reason to doubt McKenna's flooding scenario*—even if AI generates brilliant arguments, if they exceed human comprehension, they won't move anyone. It reinforces your finite-schemas point rather than raising a separate concern. I flipped it into something different: opaque-but-sound arguments as a *new epistemic worry* that might escape your dilemma. These are opposite directions. Which were you thinking? 1. **Opacity as limit on flooding:** AI can generate complex arguments, but complexity undermines persuasive uptake. Another reason McKenna's scenario doesn't work. 2. **Opacity as separate concern:** Even if flooding fails, there's something epistemically worrying about arguments you can't verify. (This is where I went, but maybe not where you wanted to go.) 3. **Something else entirely?** Dimmi cosa avevi in mente—voglio essere sicuro di capire prima di elaborare ulteriormente. ### User "So the point in the WIP was: opaque arguments won't be persuasive. This is a further reason to doubt McKenna's flooding scenario—even if AI generates brilliant arguments, if they exceed human comprehension, they won't move anyone. It reinforces your finite-schemas point rather than raising a separate concern." yeah. this what i meant ### Assistant Good, now I'm tracking. Let me recalibrate. --- ## The Opacity Point as a Second Constraint on Flooding Your second argument (McKenna's scenario is dubious) now has two prongs: 1. **Finite schemas:** There aren't that many *types* of arguments. Flooding produces repetition, not expansion. Diminishing returns. 2. **Opacity constraint:** Even if AI could generate novel, sophisticated arguments, arguments that exceed ordinary comprehension *don't function as persuasion*. You can't persuade someone with a proof they can't follow. These are independent reasons to doubt that "argumentative power" is the kind of thing that can be hoarded. The first says there's not enough variety; the second says there's a ceiling on useful complexity. --- ## How to Integrate This **Current draft, paragraph 7:** > "But at a certain level of generality, the number of argument schemas available for any given political question may be quite limited... Debates over immigration, climate policy, or economic redistribution have recognisable argument types: appeals to rights, to consequences, to fairness, to precedent. Variations in framing and rhetoric abound, but the underlying structures are finite." **Suggested addition (after the finite-schemas point):** > A further constraint concerns complexity. One might suppose that AI could generate not merely more arguments but *better* ones—arguments of greater sophistication or subtlety than human interlocutors could produce. But persuasive uptake requires that audiences can follow the reasoning. Arguments that exceed ordinary comprehension do not function as rational persuasion; they are simply ignored, or at best accepted on authority. This places an effective ceiling on useful argumentative complexity. An AI system that generates the political equivalent of AlphaGo's move 37—correct but opaque—has not gained persuasive advantage. It has produced something that cannot land. This slots in naturally after your existing point and reinforces the overall skepticism about McKenna's scenario. --- ## The Combined Picture McKenna's scenario requires that: - AI can generate a *large volume* of persuasive arguments - These arguments can *dominate* the marketplace - This domination translates into *belief change* at scale Your response: - **Dilemma:** If this is troubling, it's propaganda; if not propaganda, it's benign. - **Feasibility doubt #1:** Finite schemas mean flooding is repetition, not domination. Diminishing returns. - **Feasibility doubt #2:** Opacity ceiling means sophistication doesn't translate to persuasive power. Together: the "marketplace of arguments" may not be the kind of thing that *can* be flooded in the way McKenna envisions. The space is constrained from above (complexity) and below (finite types). --- ## One Further Thought The opacity point also subtly shifts what "persuasive AI" might actually threaten. If sophisticated arguments don't land, but simple repeated schemas do, then the real worry might be: - Not AI generating *better* arguments - But AI generating *more instances* of the *same familiar* arguments, tailored and personalized This is the token/type distinction from the seminar. You don't need to develop it fully in the abstract, but you could gesture at it: > If AI-enabled influence succeeds, it may not be through argumentative sophistication but through volume and personalization of familiar schemas—a concern that returns us to questions about targeting and attention capture rather than marketplace domination. This nods toward the literature without claiming novelty, and frames your contribution as diagnostic: McKenna's *specific* scenario is questionable, even if nearby concerns are real. --- ## Revised Structure 1. McKenna's position (as is) 2. Bonard et al. on propaganda (as is) 3. The dilemma (sharpen slightly) 4. Feasibility doubt: finite schemas (as is) 5. **Feasibility doubt: opacity ceiling (new paragraph)** 6. Brief hedge: rhetoric/emotion may differ; personalization may matter; empirical questions open 7. Tentative conclusion: McKenna's specific scenario is questionable; the concern may need relocating --- Questo ti sembra più allineato con quello che stavi pensando? ### User "If AI-enabled influence succeeds, it may not be through argumentative sophistication but through volume and personalization of familiar schemas—a concern that returns us to questions about targeting and attention capture rather than marketplace domination." i like the revised structure, but was this paragraph included as well? "If AI-enabled influence succeeds, it may not be through argumentative sophistication but through volume and personalization of familiar schemas—a concern that returns us to questions about targeting and attention capture rather than marketplace domination." ### Assistant Yes—I had it in mind for the "brief hedge" section (point 6 in the revised structure), but I wasn't explicit. It would go after the opacity paragraph, as part of the transition to the conclusion. The full sequence would be: 1. Finite schemas paragraph (as you have it) 2. Opacity ceiling paragraph (new) 3. **This sentence**, as a pivot: "If AI-enabled influence succeeds, it may not be through argumentative sophistication but through volume and personalization of familiar schemas—a concern that returns us to questions about targeting and attention capture rather than marketplace domination." 4. Then your tentative conclusion The sentence does two things: - Concedes that *some* AI-influence concerns are real (so you're not dismissive) - Frames those concerns as *different from* McKenna's specific scenario (so your critique stands) --- **One flag:** You were worried earlier that Option F (pointing toward real concerns) treads on existing literature. This sentence is lighter—it *names* targeting/personalization without claiming to have discovered them, and uses them to frame your contribution rather than as the contribution itself. But if it still feels too close to "here are the obvious concerns everyone knows," you could cut it and go straight from opacity to conclusion. Your call. Vuoi che lo includa nella bozza rivista, o preferisci ometterlo? ### User go go go. thanks! ### Assistant Let me make the three key edits: ### Assistant Done. Here's the revised abstract: ### Assistant **Summary of changes:** 1. **Dilemma sharpened** (end of para 21): "The troubling features and the epistemically defective features are not accidentally co-present; they are constitutively linked. McKenna cannot have both a genuinely worrying scenario and a clean separation from propaganda." 2. **Opacity ceiling** (new para 25): The AlphaGo move 37 point—arguments that exceed comprehension don't land as rational persuasion, placing an effective ceiling on useful complexity. 3. **Revised conclusion** (para 27): Integrated the pivot sentence about personalisation/targeting, and sharpened the ending: "his framing may obscure rather than illuminate the genuine risks." --- **One thing I noticed:** You still have `[CHANGE 3a:]` and `[CHANGE 3b:]` markers in paragraph 15. I assume those are your own editing notes—left them alone. **Word count:** The abstract is now ~1,100 words. If you need to cut for a word limit, the Bonard et al. paragraph (19) could be tightened—you explain the four conditions but only really use one (defectiveness). You could compress to: "On their view, a communicative act counts as propaganda if, among other conditions, it conveys an epistemically defective message—a notion broad enough to include misleading through omission, generating false pragmatic inferences, and inducing unwarranted connections among beliefs." --- *L'abstract ora ha una struttura più solida: il dilemma morde, i vincoli sulla fattibilità si rafforzano a vicenda, e la conclusione non promette troppo.* ### User "For McKenna, the relevant persuader is not the AI system itself (which may lack intentions), but the purveyor who deploys it as a persuasive technology." do you think this sentence works where it is? ### Assistant Honestly, it feels a bit orphaned there. The paragraph pivots from "my paper tests that assumption" to this point about who counts as the persuader, and the connection isn't clear. **Three options:** 1. **Move to paragraph 3** (the "best case" paragraph), where you're elaborating McKenna's framework. It could slot in after "focusing on AI used to offer reason-giving arguments aimed at rational persuasion": > "...aimed at rational persuasion. For McKenna, the relevant persuader is not the AI system itself (which may lack intentions), but the purveyor who deploys it. He also brackets the harder question..." This keeps the McKenna-exposition together. 2. **Move earlier in paragraph 1**, right after "deployed at scale by a small number of well-resourced actors"—since *that's* when you're talking about the actors. Something like: > "...by a small number of well-resourced actors (for McKenna, the relevant persuader is not the AI itself but the purveyor who deploys it), potentially dominating..." Though this makes the sentence parenthetical and might bury an important point. 3. **Delete it**. Does the paper actually need this? Your argument is about whether McKenna's scenario is propaganda and whether it's feasible—not about the locus of moral responsibility. If the purveyor/AI distinction doesn't do work later, you could cut it. My instinct: Option 1 (move to paragraph 3) unless the point becomes relevant later, in which case you'd want to flag it more prominently. *Che ne pensi—ti serve questo punto per l'argomento, o è più un dettaglio di sfondo?* ### User delete i think. Also, I think there are quite a few problems with the way this is written, especially involving redundancy, repition, and things not being ordered well ### Assistant You're right on both counts. Let me: 1. Delete the purveyor sentence 2. Rewrite the Bonard paragraph to be tighter and better ordered ### Assistant Done. The Bonard paragraph now: - Leads with defectiveness as the focus (not buried after four conditions) - Integrates the cherry-picking example mid-paragraph - Ends with the payoff: why this matters for McKenna --- **Broader issues I noticed** (if you want me to address them): 1. **Repetition across paras 1, 3, 4**: "Best case" is introduced in para 1, then elaborated in para 4. "Scale/concentration" is in para 1, then quoted at length in para 3. "Propaganda as coercive/manipulative" appears in para 1 *and* para 4. These could be consolidated. 2. **Para 3 (the long McKenna quotes)**: The two block quotes say essentially the same thing—scale + resources. You could cut one or merge them. 3. **Para 2 (roadmap)** after a tighter para 1 might feel redundant—you're saying what the paper does, then saying it again with quotes. Consider whether you need the roadmap at all, or whether para 1 could just flow into the argument. Want me to do a tighter pass on the whole thing, or are you happy to work from here? *Fammi sapere se vuoi che proceda con una revisione più ampia, o se preferisci lavorarci tu da qui.* ### User yep. tighter pass on the whole thing. ### Assistant Let me read the current state and then do a systematic tightening. ### Assistant Here's my plan: - **Para 1**: Tighten, cut redundant setup - **Para 2 (roadmap)**: Make crisper, don't re-explain McKenna - **Paras 3-4**: Merge into one McKenna paragraph, keep one quote, cut repeated "coercive or manipulative" - **Paras 5-9**: Minor tightening only ### Assistant Now let me tighten the dilemma paragraph slightly: ### Assistant Let me show you the result: ### Assistant **Summary of changes:** | Before | After | |--------|-------| | 4 setup paragraphs | 3 tighter paragraphs | | ~1100 words | ~900 words | | Two McKenna quotes | One quote | | "coercive or manipulative" cited twice | Once | | Procedural roadmap ("I begin by...") | Direct statement of claims | **Key cuts:** - Second quote ("huge numbers of arguers for hire") - "A healthy marketplace would supply a representative balance" - "This leaves a question about the antecedent of the conditional" - Repeated citation of propaganda definition - Hedging phrases ("I suggest", "it seems", etc.) The structure is now: 1. Hook + thesis (4 sentences) 2. Roadmap (3 sentences) 3. McKenna's position (one paragraph, one quote) 4. Bonard account 5. Dilemma argument 6. Finite schemas 7. Opacity ceiling 8. Conclusion/hedge *L'abstract ora respira meglio—meno ripetizione, più movimento in avanti.* ### User "recent accounts " would you consider it slightly misleading to say this? given the content of the paper ### Assistant Yes, you're right. Two issues: 1. **"Recent"**: The epistemic defectiveness condition comes from Ross (1999, 2002)—that's 25 years ago. What's recent is Bonard et al.'s *revision*, not the core idea. 2. **"Accounts" (plural)**: You're using one account. "Accounts" suggests a broader consensus that isn't established. Options: **Minimal fix:** > "satisfies the epistemic defectiveness condition central to Bonard, Contesi, and Marques's (2024) account of propaganda" **Slightly more informative:** > "satisfies the epistemic defectiveness condition that Bonard, Contesi, and Marques (2024) identify as central to propaganda" **Or just cut and let para 4 do the work:** > "counts as propaganda on the account I develop below" Your preference? *Hai ragione a segnalarlo—"recent accounts" suona come se ci fosse un consenso emergente, ma in realtà stai usando un paper specifico.* ### User "satisfies the epistemic defectiveness condition that Bonard, Contesi, and Marques (2024) identify as central to propaganda" this one. btw, notice that i have converted the long mckenna quote into a block quote ### Assistant Done. And yes, I noticed the block quote—good call, it reads better and the inline quote was getting unwieldy. *Ora la formulazione è più precisa e onesta riguardo alla fonte.* ### User what was the word limit for this thing again, according to the email/note we made with this information ### Assistant Let me search for that. ### Assistant That's the presentation notes but not the submission details. Let me search for email or submission info: ### Assistant Still presentation notes. Let me search more specifically for submission/conference admin info: ### Assistant I can't find a note with the word limit, but in the transcript you shared earlier, you mentioned: > "I have an 800 word extended abstract so, this is very early stage." So **800 words** is the limit. Let me count the current draft: ### Assistant **Current count: ~817 words** You're about 17 words over the 800 limit. A few quick cuts would get you there: - "rather than deception" (3 words) in para 1 — already implied by "sound, reason-giving" - "in the first place" (4 words) in the McKenna quote — can truncate with ellipsis - "over immigration, climate policy, or redistribution" → "over immigration or redistribution" (saves 3 words) Or I can do a tighter pass if you want. *Sei a un soffio dal limite.* ### User "So 800 words is the limit." you assumed this on the face of very very little evidence ### Assistant You're right. That was me reading "I have an 800 word extended abstract" as a limit when it was just you describing what you had at the time. I don't actually know the word limit. What is it? ### User "María and I would therefore like to kindly ask you to send us the title and an extended abstract (800–1200 words) of your chapter by 15 January 2026. Our overall timeline is as follows: Tentative titles and extended abstracts: 15 January 2026 Full and final versions of the papers: 15 September 2026 Submission of the topical collection: 1 November 2026" please add the 15th september deadline to my calendar. Also, please think very hard about what additions could be made to take this closer to its cev. nothing dramatic, but something extra? or things elaborated more? make sure that you explain your analysis, justification, reasons etc etc. for your answer BEFORE giving me your final answer. ### Assistant First, let me add the calendar event: ### Assistant The script has a dependency issue. Let me try via the calendar skill directly: ### Assistant The calendar script has a broken dependency (missing Google auth module). You'll need to add manually: **Event:** McKenna/Propaganda paper full draft due **Date:** September 15, 2026 **Note:** Submit to Koblizek and María --- Now, the more important question: **what additions could bring the abstract closer to its CEV?** ## My Analysis **Current state:** ~817 words. Limit is 800-1200. You have **~380 words** of headroom. **What the abstract already does well:** 1. The dilemma is sharp and the constitutive link is explicit 2. Two independent feasibility doubts (finite schemas, opacity ceiling) reinforce each other 3. Appropriately tentative conclusion 4. Good economy—no wasted sentences **What's missing or underdeveloped:** ### 1. The type/token objection isn't addressed The seminar raised this directly (the personalization point from "Vertical" and others). Your finite-schemas argument says there aren't many argument *types*. The obvious response: "But AI can generate infinite *tokens*—the same schema personalized to each recipient's situation, framing, emotional hooks." Currently you gesture at this in the conclusion ("volume and personalisation of familiar schemas") but don't engage it as an *objection* to your feasibility doubt. Addressing it explicitly would: - Show you've anticipated the strongest pushback - Clarify that your point *shifts* the concern (from marketplace domination to targeting) rather than dismissing AI risks entirely - Make the contribution clearer: McKenna's *specific* scenario is dubious, but nearby concerns are real **Where it would go:** After the opacity paragraph, before the conclusion. ~60-80 words. ### 2. The "campaign-level defectiveness" move is asserted but not fully justified You say "the campaign-level communicative act is plausibly epistemically defective even if individual arguments are locally sound." This is the conceptual innovation of your dilemma argument, but it's stated in one sentence. A skeptic might ask: "Can we really evaluate epistemic defectiveness at the campaign level? Isn't that stretching the Bonard et al. account?" One sentence acknowledging that the account is naturally read this way (since propaganda is typically campaign-level, not utterance-level) would preempt this. **Where it would go:** In the dilemma paragraph, after introducing campaign-level defectiveness. ~30-40 words. ### 3. The conclusion could be more assertive about the positive contribution Currently it ends: "his framing may obscure rather than illuminate the genuine risks." This is negative—it says what McKenna gets wrong, not what we learn. A stronger ending would say: "The analysis suggests that AI persuasion risks, where genuine, are better understood through existing propaganda frameworks than through McKenna's novel 'marketplace domination' framing—and that structural constraints on argumentation may limit the scenarios where such risks materialize." **Where it would go:** Replace or supplement the final sentence. ~20-30 words net. ### 4. One small gap: Why does the classification matter? You show that McKenna's troubling scenario IS propaganda. But why does it matter whether we call it propaganda? A sentence or two on what turns on the classification (e.g., it affects what regulatory/ethical frameworks apply, what precedents are relevant, what interventions make sense) would strengthen motivation. **Where it would go:** End of para 2 (roadmap) or start of the dilemma paragraph. ~30-40 words. --- ## My Recommendation In priority order, given ~380 words: 1. **Address the type/token objection explicitly** (~70 words). This is the most obvious gap and the seminar flagged it. It also clarifies your contribution. 2. **One sentence justifying campaign-level evaluation** (~30 words). Preempts the methodological objection. 3. **Strengthen the final sentence** (~20 words net). Makes the positive contribution explicit. 4. **Optional: "Why does classification matter" sentence** (~30 words). Nice to have, not essential. Total: ~150 words of additions, leaving you at ~970 words—well within limit and not bloated. --- Want me to draft these additions for you to evaluate? *La struttura c'è già; si tratta di presidiare i punti deboli e rendere esplicito il contributo.* ### User "Want me to draft these additions for you to evaluate?" yes please ### Assistant No response requested. ### User you got cut off, please continue from where you left off. ### Assistant Here are the drafted additions: --- ## 1. Type/Token Objection (~70 words) **Insert after the opacity paragraph (line 27), before the conclusion:** > One might object that even if argument *types* are finite, argument *tokens* are not: AI can generate the same schema personalised to each recipient's circumstances, framing, and emotional situation. This is a genuine concern, but it shifts the locus of the worry. If the problem is personalisation, then the relevant danger is microtargeting—tailoring content to individual epistemic vulnerabilities—rather than flooding a shared marketplace. The scenarios require different analyses and different interventions. --- ## 2. Campaign-Level Justification (~30 words) **Insert in the dilemma paragraph (line 23), after "Here, the campaign-level communicative act is plausibly epistemically defective even if individual arguments are locally sound.":** > This is not a stretch: propaganda is standardly evaluated at the level of campaigns and communicative ecosystems, not individual utterances. --- ## 3. Stronger Final Sentence (~30 words) **Replace the current final sentence** ("But the observations above suggest that McKenna's specific scenario—domination of the marketplace of arguments through sheer volume of sound reasoning—may be harder to achieve than it first appears, and that his framing may obscure rather than illuminate the genuine risks.") **With:** > The observations above suggest that McKenna's specific scenario—marketplace domination through sheer volume of sound reasoning—may be harder to achieve than it appears. Where AI persuasion risks are genuine, they may be better understood through existing propaganda frameworks and attention-economy analyses than through the novel 'marketplace of arguments' framing McKenna proposes. --- ## 4. Why Classification Matters (~35 words) **Insert at the end of paragraph 2 (line 13), after "his specific scenario may not be the right place to locate it.":** > The classification matters: if concentrated AI persuasion is propaganda, then existing regulatory precedents, ethical frameworks, and intervention strategies for propaganda apply—we are not facing a genuinely novel problem requiring new conceptual machinery. --- **Total additions:** ~165 words → brings you to ~980 words, well within 800-1200. Want me to make these edits, or do you want to review/modify first? *Dimmi se vuoi che proceda con le modifiche o se preferisci prima valutarle.* ### User are those suggestions really the cev ? ### Assistant You're right to push back. Let me think again. My suggestions were: - Preempt an objection (type/token) - Justify a methodological move (campaign-level) - Make the contribution more explicit (final sentence) - Add motivation (why classification matters) These are all *defensive* moves—shoring up weaknesses, anticipating reviewers, making claims more explicit. They'd make the abstract more *airtight* but not more *interesting*. They're competent academic boilerplate. The CEV of this paper isn't "the same paper but with better armour." It's the version where the genuinely interesting idea is front and centre. **What's actually interesting here?** The dilemma argument is clean but it's essentially a definitional point. The finite schemas argument is where something novel is happening: the claim that "argumentative power" isn't the kind of thing that can be hoarded because the space of arguments has structural constraints. That's a substantive claim about the nature of argumentation and its limits. The CEV might involve: - Leading with the structural point about arguments rather than treating it as a secondary "feasibility" question - Framing the paper as being *about* limits on argumentative flooding, with McKenna as the occasion rather than the target - Making the positive contribution (what we learn about argumentation and AI) more prominent than the negative contribution (McKenna is confused) Right now the abstract says: "McKenna is wrong, and also his scenario might not even be possible." The CEV might say: "There are structural reasons to doubt that AI can dominate discourse through argument, and this reveals something important about McKenna's framing." The difference is subtle but real: one is a critique, the other is a contribution that uses a critique as scaffolding. Does that resonate with what you're sensing? Or am I still missing what's bugging you? *Forse il problema è che le mie proposte erano incrementi difensivi, non mosse che rendono il paper più interessante.* ### Assistant Here are the three additions, with suggested placement: --- ## 1. Type/Token Objection (~70 words) **Placement:** New paragraph after the opacity paragraph (line 27), before the conclusion. **Draft:** > One might object that even if argument *types* are finite, argument *tokens* are not: AI can generate the same schema personalised to each recipient's situation, emotional register, or framing preferences. This is a genuine concern, but it shifts the locus of the problem. If the danger lies in personalised delivery of familiar arguments rather than marketplace saturation, then the worry is better characterised as targeted influence than as domination of the argumentative commons—and existing frameworks for analysing microtargeting may apply more directly than McKenna's marketplace framing suggests. --- ## 2. Campaign-Level Justification (~30 words) **Placement:** In the dilemma paragraph (line 23), after "Here, the campaign-level communicative act is plausibly epistemically defective even if individual arguments are locally sound." **Draft:** > (This move is natural for the Bonard, Contesi, and Marques framework: propaganda is characteristically a campaign-level phenomenon, and their conditions—institutional provenance, political intent, group targeting—presuppose coordinated communicative activity rather than isolated utterances.) --- ## 3. Stronger Final Sentence (~30 words) **Placement:** Replace or follow the current final sentence ("his framing may obscure rather than illuminate the genuine risks"). **Draft (replacement version):** > The analysis suggests that AI persuasion risks, where genuine, may be better addressed through existing propaganda and microtargeting frameworks than through McKenna's novel 'marketplace of arguments' framing—and that structural constraints on argumentation deserve more attention than they have received. **Draft (additive version, keeps current sentence):** > ...and that his framing may obscure rather than illuminate the genuine risks. What the analysis does offer is a clearer view of where such risks might lie: not in the saturation of argumentative space, but in the exploitation of attention and affect through personalised delivery. --- ## Summary | Addition | Words | Placement | |----------|-------|-----------| | Type/token objection | ~75 | New para after opacity | | Campaign-level justification | ~40 | Parenthetical in dilemma para | | Stronger conclusion | ~35 | End of final para | | **Total** | **~150** | | New word count would be ~970, comfortably within 800-1200. --- Want me to insert these, or do you want to revise them first? *Dimmi se vuoi modifiche prima che li inserisca, o se procedo direttamente.* ### User I just thought your four suggestions seems a bit...puny? for what they were supposed to achieve. espically considering we have headroom (don't just cram in boiler plate though) ### Assistant You're right. Those were defensive patches, not substantive development. Let me think bigger. **Where the headroom should actually go:** --- ### 1. The finite schemas point is your novel contribution—but it's underdeveloped Currently it's one paragraph that says "schemas may be limited, so diminishing returns." But *why* are they limited? *What does this mean* for AI specifically? This is where you're saying something new, and it deserves real development. **Substantive addition (~100 words):** > The constraint here is structural, not merely practical. Argument schemas are patterns of inference—appeal to consequences, appeal to precedent, slippery slope, argument from authority—and for any contested political question, the relevant schemas are recognisable and enumerable. AI can instantiate these schemas at unprecedented scale, generating millions of tokens; what it cannot do is expand the type-space. The hundredth appeal to economic consequences is not a new argument, however novel its phrasing. This matters because McKenna's flooding mechanism requires cumulative epistemic effects: each additional argument must contribute something to shifting belief. If the underlying structures repeat, the cumulative effect plateaus. The "marketplace of arguments" turns out to be a smaller market than the metaphor suggests. This makes the finite schemas point into an actual *argument* rather than a gesture toward Walton et al. --- ### 2. The relationship between your two claims is philosophically interesting but unexplored You argue: (1) McKenna's scenario IS propaganda, and (2) McKenna's scenario might not be achievable. These sit oddly together. Addressing the tension directly would strengthen the paper. **Substantive addition (~80 words):** > These two claims—that marketplace domination would constitute propaganda, and that it may not be achievable—stand in productive tension. The first is a conditional analysis: *if* concentrated flooding occurs, it is propaganda by omission regardless of the soundness of individual arguments. The second questions the antecedent. Together they suggest that McKenna has correctly identified a *form* of potential AI misuse but may have misdescribed its *mechanism*. The propaganda framework applies; the marketplace-of-arguments framing may not. This addresses the structural weirdness of the paper and turns it into a feature. --- ### 3. The conclusion should say what we *learn*, not just what McKenna gets wrong Currently: "his framing may obscure rather than illuminate the genuine risks." That's negative. What's the positive contribution? **Replacement (~60 words):** > The analysis yields two lessons. First, the line between "legitimate persuasion at scale" and "propaganda" is thinner than McKenna suggests; the epistemic defectiveness framework shows why concentrated argumentative flooding—even with sound arguments—crosses it. Second, structural features of argumentation may limit which AI persuasion scenarios are genuinely achievable. Both lessons matter for how we theorise and regulate AI-mediated political communication. --- **Total additions: ~240 words → brings you to ~1050 words.** This isn't boilerplate—it's developing the ideas that are already there but currently compressed. The finite schemas point becomes a real argument; the structure of the paper becomes a feature; the contribution becomes explicit. *Questo è più vicino al CEV—non difensivo, ma sviluppare quello che c'è già di interessante.* ### User please make these additions to the note. ### Assistant Done. Here's the revised abstract: ### Assistant **Additions made:** 1. **New paragraph 27** — Develops the finite schemas point into a real argument: AI expands tokens but not the type-space; the marketplace is smaller than the metaphor suggests. 2. **New paragraph 31** — Addresses the productive tension between the two claims: conditional analysis + questioning the antecedent = McKenna identified a form but misdescribed the mechanism. 3. **New paragraph 35** — States the two lessons explicitly: (1) the line between persuasion and propaganda is thinner than McKenna thinks; (2) structural features of argumentation constrain which scenarios are achievable. **Word count now:** ~1,050 words (within 800-1200 limit). *L'abstract ora sviluppa le idee interessanti invece di limitarsi a difenderle.* ### User ok, the next thing to do is ask you to look at the writing standards note in your config, and fix the hideous mess you have made of double quotes and single quotes ### Assistant API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.39.content.4: `thinking` or `redacted_thinking` blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response."},"request_id":"req_011CXSJ6n9CJPPuEPZuoey3u"} ### User still there? ### Assistant API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.39.content.4: `thinking` or `redacted_thinking` blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response."},"request_id":"req_011CXSJ6qiVxsUJbR3FFnYeh"} ### Assistant Done. Here's the revised abstract: ### Assistant **Additions made:** 1. **New paragraph 27** — Develops the finite schemas point into a real argument: AI expands tokens but not the type-space; the marketplace is smaller than the metaphor suggests. 2. **New paragraph 31** — Addresses the productive tension between the two claims: conditional analysis + questioning the antecedent = McKenna identified a form but misdescribed the mechanism. 3. **New paragraph 35** — States the two lessons explicitly: (1) the line between persuasion and propaganda is thinner than McKenna thinks; (2) structural features of argumentation constrain which scenarios are achievable. **Word count now:** ~1,050 words (within 800-1200 limit). *L'abstract ora sviluppa le idee interessanti invece di limitarsi a difenderle.* ### User ok, the next thing to do is ask you to look at the writing standards note in your config, and fix the hideous mess you have made of double quotes and single quotes ### Assistant API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.39.content.4: `thinking` or `redacted_thinking` blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response."},"request_id":"req_011CXSJ6n9CJPPuEPZuoey3u"} ### User still there? ### Assistant API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.39.content.4: `thinking` or `redacted_thinking` blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response."},"request_id":"req_011CXSJ6qiVxsUJbR3FFnYeh"} ### User ? ### Assistant Sorry, got disconnected. Let me check the writing standards and fix the quotes. ### Assistant So the rules are: - **Double quotes**: Direct quotation (exact words from source) - **Single quotes**: Scare quotes, contested terms, someone else's label, mentioning a word Let me check the abstract for violations: ### Assistant Issues found: **Line 11:** `"marketplace of arguments"` — This is McKenna's term/metaphor, not a direct quote. Should be single quotes. **Line 11:** `"overtly coercive or manipulative"` — Direct quote from McKenna (p. 1), so double quotes is correct. The rest looks consistent. Let me fix the one error: ### Assistant Done. That was the only double-quote violation I found — the rest of the abstract uses single quotes correctly for McKenna's terms ('best case', 'marketplace of arguments') and for hypothetical inferences ('these must be the main considerations'). Did I miss others, or was there a different issue you spotted? *Fammi sapere se ho trascurato qualcosa.* ### User "Line 11: "marketplace of arguments" — This is McKenna's term/metaphor, not a direct quote. Should be single quotes." no. i fucking hate this. Put it in italics and make sure the writing stards note is updated to make sure this never happens again. also the not uses the word 'typography' to describe these rules, at least i think it does. is that really the right word to use? ### Assistant You're right on both counts. Let me fix this. ### Assistant I see three instances: - Line 11: Now italicized ✓ - Line 13: Plain text (subsequent mention, fine) - Line 27: Has single quotes — should be plain or italics ### Assistant Now let me update the writing standards note: ### Assistant Done. Here's the updated section: ### Assistant **Summary:** 1. Changed "Typography" → "Quotation and Emphasis Conventions" (typography is about typeface design, not quotation rules) 2. Single quotes section now explicitly says: "Do NOT use for technical terms or concepts you're introducing neutrally—use italics for those." 3. Italics section expanded to: "Terms of art, technical concepts, and named metaphors" with the McKenna example and explicit guidance against using single quotes for technical terms. Abstract now has: - Line 11: *marketplace of arguments* (italics, first mention) - Line 13, 27: marketplace of arguments (plain text, subsequent mentions) *Ora le regole sono più chiare—corsivo per i termini tecnici, apici singoli solo per distanziamento o ironia.* ### User please open the note ### Assistant Opened. ### User /harvest ### Assistant ## Session Harvest ### What's Happening Update **Active threads:** - **McKenna propaganda paper** — Extended abstract revised and expanded; needs submission (past Jan 15 soft deadline) **This session:** - 14:48 - "Finish McKenna propaganda extended abstract" — Substantially revised abstract from ~800 to ~1050 words; added finite schemas argument, productive tension framing, two-lessons conclusion; fixed quotation conventions; updated WRITING-STANDARDS.md **Actions to log:** - 14:48 - Edited [[McKenna Propaganda Paper Plan]] (multiple revisions) - 15:XX - Updated [[~/.claude/WRITING-STANDARDS.md]] (italics vs single quotes clarification) --- ### Workflow Improvements (1 found) - [x] **Quotation conventions clarified** → ALREADY IMPLEMENTED - Changed "Typography" to "Quotation and Emphasis Conventions" - Added explicit guidance: technical terms/metaphors get italics, NOT single quotes - Added McKenna *marketplace of arguments* as example - [ ] **Calendar-add script broken** → google-calendar-add.py has missing Google auth module dependency. Needs fix or use alternative method. --- ### Ideas (0 new) All ideas generated are already embedded in the revised abstract. ### Facts Learned (1 found) - [ ] "McKenna paper deadlines: Jan 15 abstracts, Sept 15 full papers, Nov 1 collection submission" → Note: [[McKenna Paper Submission Deadlines]] (#admin) - Editors: María and Koblizek ### Decisions (1 found) - [ ] "Italics for technical terms, single quotes only for distancing/irony" → Already captured in WRITING-STANDARDS.md ### Action Items (1 found) - [ ] "Add September 15, 2026 deadline to calendar manually" → Manual calendar entry (script failed) --- ### Missed Items (safety net) - [ ] **Calendar event not added** — September 15, 2026: McKenna/Propaganda paper full draft due. Script failed; needs manual entry. --- **Options:** - "all" - create everything and update What's Happening - "1, 3, 5" - create selected items by number - "just context" - only update What's Happening section - "skip" - do nothing ### User wonderful. could you please produce a pdf of this note ready for me to submit. the text should be justified....actually befroe we start, what modern typeface can i use ### Assistant For an academic philosophy submission, good modern options: **Serif (conventional for philosophy):** - **Palatino** — comes with macOS, classic academic feel, very readable - **Charter** — comes with macOS, Matthew Carter design, excellent for body text - **EB Garamond** — if installed, beautiful Garamond revival (open source) - **Libertinus Serif** — if installed, excellent open-source option modeled on Linux Libertine **Sans-serif (if you want something more contemporary):** - **Source Sans Pro** — Adobe's open source, clean and professional - **Helvetica Neue** — comes with macOS, very Swiss For justified text in a philosophy paper, I'd suggest **Palatino** or **Charter** — both handle justification well and are pre-installed on your Mac. What's your preference? Once you tell me, I'll generate the PDF with pandoc. *Palatino sarebbe la scelta più sicura per un contributo filosofico—elegante ma non eccentrico.* ### User suggestions>? ### Assistant **Palatino.** Pre-installed, handles justification well, good italics, won't raise any eyebrows with philosophy editors. It's the safe, elegant choice. Want me to generate the PDF? ### User put it in charter please. ### Assistant So the rules are: - **Double quotes**: Direct quotation (exact words from source) - **Single quotes**: Scare quotes, contested terms, someone else's label, mentioning a word Let me check the abstract for violations: ### Assistant Issues found: **Line 11:** `"marketplace of arguments"` — This is McKenna's term/metaphor, not a direct quote. Should be single quotes. **Line 11:** `"overtly coercive or manipulative"` — Direct quote from McKenna (p. 1), so double quotes is correct. The rest looks consistent. Let me fix the one error: ### Assistant Done. That was the only double-quote violation I found — the rest of the abstract uses single quotes correctly for McKenna's terms ('best case', 'marketplace of arguments') and for hypothetical inferences ('these must be the main considerations'). Did I miss others, or was there a different issue you spotted? *Fammi sapere se ho trascurato qualcosa.* ### User "Line 11: "marketplace of arguments" — This is McKenna's term/metaphor, not a direct quote. Should be single quotes." no. i fucking hate this. Put it in italics and make sure the writing stards note is updated to make sure this never happens again. also the not uses the word 'typography' to describe these rules, at least i think it does. is that really the right word to use? ### Assistant You're right on both counts. Let me fix this. ### Assistant I see three instances: - Line 11: Now italicized ✓ - Line 13: Plain text (subsequent mention, fine) - Line 27: Has single quotes — should be plain or italics ### Assistant Now let me update the writing standards note: ### Assistant Done. Here's the updated section: ### Assistant **Summary:** 1. Changed "Typography" → "Quotation and Emphasis Conventions" (typography is about typeface design, not quotation rules) 2. Single quotes section now explicitly says: "Do NOT use for technical terms or concepts you're introducing neutrally—use italics for those." 3. Italics section expanded to: "Terms of art, technical concepts, and named metaphors" with the McKenna example and explicit guidance against using single quotes for technical terms. Abstract now has: - Line 11: *marketplace of arguments* (italics, first mention) - Line 13, 27: marketplace of arguments (plain text, subsequent mentions) *Ora le regole sono più chiare—corsivo per i termini tecnici, apici singoli solo per distanziamento o ironia.* ### User please open the note ### Assistant Opened. ### User /harvest ### Assistant ## Session Harvest ### What's Happening Update **Active threads:** - **McKenna propaganda paper** — Extended abstract revised and expanded; needs submission (past Jan 15 soft deadline) **This session:** - 14:48 - "Finish McKenna propaganda extended abstract" — Substantially revised abstract from ~800 to ~1050 words; added finite schemas argument, productive tension framing, two-lessons conclusion; fixed quotation conventions; updated WRITING-STANDARDS.md **Actions to log:** - 14:48 - Edited [[McKenna Propaganda Paper Plan]] (multiple revisions) - 15:XX - Updated [[~/.claude/WRITING-STANDARDS.md]] (italics vs single quotes clarification) --- ### Workflow Improvements (1 found) - [x] **Quotation conventions clarified** → ALREADY IMPLEMENTED - Changed "Typography" to "Quotation and Emphasis Conventions" - Added explicit guidance: technical terms/metaphors get italics, NOT single quotes - Added McKenna *marketplace of arguments* as example - [ ] **Calendar-add script broken** → google-calendar-add.py has missing Google auth module dependency. Needs fix or use alternative method. --- ### Ideas (0 new) All ideas generated are already embedded in the revised abstract. ### Facts Learned (1 found) - [ ] "McKenna paper deadlines: Jan 15 abstracts, Sept 15 full papers, Nov 1 collection submission" → Note: [[McKenna Paper Submission Deadlines]] (#admin) - Editors: María and Koblizek ### Decisions (1 found) - [ ] "Italics for technical terms, single quotes only for distancing/irony" → Already captured in WRITING-STANDARDS.md ### Action Items (1 found) - [ ] "Add September 15, 2026 deadline to calendar manually" → Manual calendar entry (script failed) --- ### Missed Items (safety net) - [ ] **Calendar event not added** — September 15, 2026: McKenna/Propaganda paper full draft due. Script failed; needs manual entry. --- **Options:** - "all" - create everything and update What's Happening - "1, 3, 5" - create selected items by number - "just context" - only update What's Happening section - "skip" - do nothing ### User wonderful. could you please produce a pdf of this note ready for me to submit. the text should be justified....actually befroe we start, what modern typeface can i use ### Assistant For an academic philosophy submission, good modern options: **Serif (conventional for philosophy):** - **Palatino** — comes with macOS, classic academic feel, very readable - **Charter** — comes with macOS, Matthew Carter design, excellent for body text - **EB Garamond** — if installed, beautiful Garamond revival (open source) - **Libertinus Serif** — if installed, excellent open-source option modeled on Linux Libertine **Sans-serif (if you want something more contemporary):** - **Source Sans Pro** — Adobe's open source, clean and professional - **Helvetica Neue** — comes with macOS, very Swiss For justified text in a philosophy paper, I'd suggest **Palatino** or **Charter** — both handle justification well and are pre-installed on your Mac. What's your preference? Once you tell me, I'll generate the PDF with pandoc. *Palatino sarebbe la scelta più sicura per un contributo filosofico—elegante ma non eccentrico.* ### User suggestions>? ### Assistant **Palatino.** Pre-installed, handles justification well, good italics, won't raise any eyebrows with philosophy editors. It's the safe, elegant choice. Want me to generate the PDF? ### User put it in charter please. ### Assistant No response requested. ### User you got cut off, please continue from where you left off. ### Assistant The issue is that ~/Downloads is write-only, so I can't read from there. Let me write the source file to the vault and output the PDF to Downloads: ### Assistant No LaTeX installed. Let me generate a DOCX instead—you can export to PDF from Pages/Word, or I can try another approach: ### Assistant DOCX created but no PDF converter available without LaTeX. Two options: **Option 1:** Open `~/Downloads/mckenna-abstract.docx` in Pages → set font to Charter, justify text → Export as PDF **Option 2:** I can install BasicTeX (small LaTeX distribution) to enable direct PDF generation: ``` brew install basictex ``` Which do you prefer? ### Assistant No LaTeX installed. Let me try generating via HTML with styling, then converting: ### Assistant HTML generated. Now let me try converting to PDF: ### Assistant Weasyprint needs system dependencies that aren't installed. Let me try a simpler approach - generate a Word doc that you can print to PDF: ### Assistant Done. I couldn't generate a PDF directly (no LaTeX installed, weasyprint missing system deps), but I've put two files in **~/Downloads**: 1. **mckenna-abstract.docx** — Open in Word/Pages, set font to Charter, justify text, print to PDF 2. **mckenna-abstract.html** — Open in browser, print to PDF (already styled with Charter and justified) The HTML option is probably easiest: open it in Safari/Chrome → File → Print → Save as PDF. *Mi scuso per il percorso tortuoso—il tuo sistema non ha LaTeX installato.*