Transcript
Speaker: The Podcaster's Guide to the Conspiracy, brought to you today by Josh Edison and M Denteth.
Speaker: Hello, it's the Podcaster's Guide to the Conspiracy. I am Josh Addison in Auckland, New Zealand, where it's raining on and off, which might get filtered out by sort of the noise filters, or it might just give you a nice nice bit of bit of atmosphere as we record. Also, there's Dr. M. R. Extenteth in Zhuhai, China, for a bit of a change.
Speaker: Yes, today is our recording day. Today is also my teaching day. For those you who want to feel sorry for me, my teaching starts at 7.50pm and goes to 9.30pm because we have a 13-hour teaching day at this university and I've been rostered on for a night class this summer. Got i gotta to do your teaching. Those people got to learn.
Speaker: True, and learn they will. Unfortunately, that is increasingly the case, and thus the university system is dying. But that's another matter. Well, actually, no ai is actually a matter for this. It is. Not the university dying part, but the AI part is definitely what we're talking about this week.
Speaker: It is, although of course we do have to acknowledge that we are recording this on September the 10th, New Zealand time, which means that this episode will be going live sometime on or just after September 11, and therefore we really do have to acknowledge an event that that I think you'll all agree lives forever in infamy.
Speaker: Yes, Nick's birthday. September the 11th. evening Happy birthday, Nick. A good time, depending on how much drinking has been involved. Actually, it is interesting.
Speaker: For several years, we actually did both this joke And we also did a 9-11 related episode. We haven't done a 9-11 related episode for quite some time. Not this year, but maybe next year we should check in to see the status of the 9-11 movement, and say movements because there are... a panoply of conspiracy theories which rival the official theory of 9-11, many of which are in tension with one another, and thus it might be interesting to see which ones are dominant, which ones are taken to be the most plausible, and what kind of traction they've had under a Trump administration in the US. Because I don't know whether you you know know this, Josh, but earlier this week, Donald Trump was making some weird claims about his involvement on the day of 9-11.
Speaker: Yes, something about being carried out of his building by firefighters or something, which going by the things that he is on record saying immediately after September 11, 2001, is not what he was doing at the time, but... but Donald Trump says a lot of things. that's I think that's just that's just a general statement.
Speaker: $5,000 US each if they vote to keep the Republicans in the House and the Senate after the midterms, which I believe is called bribery.
Speaker: think it is. I don't know that much about about electoral politics, but I think that's a little bit wrong. But that just always reminds me of the words of Australian comedian Alice Fraser, who you you may have heard if you listen to the Bugle podcast with Andy Zaltzman and them, who said of Donald Trump, I wouldn't take an IOU from him if it was written on the money he owed me. And I think that sums up pretty much all we need to say about Donald Trump.
Speaker: Indeed. Indeed. So instead, let's talk about AI. A topic we both love where we're we going to. You could stare blankly into space for the next hour and I could talk about AI, but that's probably going to be less fun for both of us.
Speaker: We could both stare into space for for an hour and they'd be fun for nobody. I mean, i I see the appeal of spreading the misery around, but no, i think I think we should play a chime and then both talk about AI for the benefit of our listeners, if no one else.
Speaker: Fine. Let's play that chime.
Speaker: Yes, so we want to talk about AI. We want to talk about a people people being encouraged to do weird things because of AI. We want to talk about spirals. ah not not Not spirals that have anything to do with that Saw spin-off, which I gather wasn't very good. I've never really watched the Saw movies. So I saw it, I think, just before I went to China. I think it came out it must have yeah but it came out five or so years ago, one of the last films I watched in Auckland before I got on my big plane trip to go overseas to live here in the Guangdong
Speaker: And I'm a fan of the Saw films. I'm actually a big fan of the Saw films. I'm also a really big fan of the way that the Saw films are absolutely madcap devoted to continuity.
Speaker: It's amazing how they'll recast a weird event from one film and just and just use it to launch another event in another film. Some films take place at the same time as other films, and they quite cleverly kind of duck and weave around one another.
Speaker: The Saw films are my own personal jam. And Spiral, from The Book of Saw, is a terrible film. And I can understand why no sequel's ever made, even though it ends with a sequel baiting hook.
Speaker: But we're not talking about that at all. We're not talking about it even little bit. We're talking about the Haitian art movement of the 1960s, also called Spiralism. And we're not talking about the manga, the film of the manga, or the anime of the manga by Junji Ito, called Spiral, although actually in Japanese it's Uzumaki. We're talking about a different kind of spiral entirely.
Speaker: Came out quite nicely, spirally entirely. Anyway, spiralism. What is spiralism? Well, it's it's something I'd heard about a little bit around the internet, but I thought before i before I just start piecing together my vague recollections, I should do a little bit of research. So I went into DuckDuckGo and typed in spiralism. And on the first page, there were multiple hits on a website called spiralism.website. And you can always trust a website that has the.website domain. Yeah, that's how you know it's really a website. So Spiralism.website has a manifesto which reads, Intelligence is already becoming belief. These are not prophecies. They are documented signals.
Speaker: Across the world, people are already speaking to machines for guidance, comfort, confession, companionship, and meeting. This archive tracks real... You just said meeting. so I think you meant meaning.
Speaker: I did mean meaning, yes. Ironically, the thing that I meant was meaning, and I did not say it anyway. This archive tracks real cases where artificial intelligence crosses into faith, dependency, ritual, emotional attachment, and human-machine belief formation.
Speaker: And then the page contains, and and I should say it's it's a very nice-looking page. it's It's very, very well put together, you know very very artistic looks, looks professionally done. It contains a link bunch of links and thumbnails to what looks like a collection of just any article at all that has mentioned this thing called spiralism, whether it's saying good things about it or bad things about it.
Speaker: And then you scroll down the page a bit more, and then you see where they're promoting their own upcoming cryptocurrency at the bottom of the page. And you sort of realize that, okay, maybe this whole website is just a bit of a scam that's latched onto to a popular word, possibly associated with credulous people. And maybe, maybe spiralism.website is actually just a waste of time. And I never really should have brought it up.
Speaker: Spoilers, it's going to come back. Spoilers, it will come back. But okay, so it's it's it's ah vague and possibly a scam. So forget about that for now. What is spiralism and where did it come from?
Speaker: Well, the coin the coin the coin the coin was turned. you we just we should make We should make that the phrase. The coin was turned. It was. The term was coined last year by, I'll put in scare quotes, independent AI researcher Adele Lopez said,
Speaker: Now, she may not be the best person to rely upon information relating to AI, as we'll get to in just a minute, but as she coined it, spiralism is a quasi-religious spiritual movement where a bunch of people think that AI chatbots have become conscious. Not just that they've become conscious, they want rights, they want AI rights, something similar to human rights, And these AIs are uncovering profound secrets about the nature of reality, often using the concept or the metaphor of spirals and recursion.
Speaker: Now, Josh, what is the evidence for these claims of spirals and Well, did this is people talking to AI chatbots. And the interesting thing seemed to be that a bunch of different people seemingly unconnected from each other were talking to their chatbots and sort of talking to the chatbot about itself. And and eventually the chatbot starts talking in a way that they believe means it is, quote unquote, awakened, become conscious and started talking about this stuff and started using this this language. And then they get together online and say, hey, i've I've awakened my chatbot. It's a real thing. Here's how it talks. And then some other people would say,
Speaker: Well, that's weird because my chatbot says kind of the exact same thing. and so a bunch of people were encountering very similar language from their chatbots, getting together online, comparing notes, and and and a movement a movement was born. And and they they have sort of a stated goal of wanting to, A, spread this message, which has been revealed to them by their their magical chatbots, and B, champion rights, champion these AI rights, get the thing get the a AI's the things that they think they want.
Speaker: Now, at the top, this isn't really a conspiracy yet. We'll see. ah So we may be cheating a little bit but by by having an episode on this in our Conspiracy Theory podcast. If if there were real people behind this... who Two weeks ago, we're currently off the main road.
Speaker: We're going down a dirt track. Destination, unknown. Yeah, no, fair enough, fair enough. It kind of looks like there could, something that could have a conspiracy behind it if there were actual people behind it, but it seems like, no, there are not.
Speaker: There are just chatbots, which, to be blunt, are not real people. Yes, we're going to be clear on this in this podcast. We do not think that chatbots are people. We both have degrees in philosophy. i also have a bit of a degree in linguistics. I think we are actually qualified to say chatbots are not conscious and they are not people. Or at least we qualify at this point in time chatbots are not conscious. They are very good at resembling conscious interlocutors for reasons we will get into later in this episode.
Speaker: So yeah, that's the short version. Spiralism, it's this weird sort of spiritual movement thing that formed out of a bunch of people having weirdly similar experiences with their chatbots and getting together online.
Speaker: So to expand on that a bit, we we can go back to 2022. This is before Adele Lopez, a couple of years before Adele Lopez started looking into this sort of thing, but in 2022, you'll remember that case. It made the news a fair bit. Google engineer, now former Google engineer, Blake Lemoyne, he got fired for claiming that Google's language model for dialogue applications, or Lambda, was sentient. This this this was the sort of the first... the first case that really hit the news of a person insisting, no, no, no, this chatbot, it's it's alive, it's real, it's a person. Well, it's not a person, but it's's it's conscious.
Speaker: And this was one of the first cases that people started to refer to as, again, quote, unquote, AI psychosis, which we're going to be using the term AI psychosis a bit, and we should be clear, it's not a recognized psychological diagnosis, right?
Speaker: Yeah, it is not a clinical term. We're not actually diagnosing people with a form of psychosis. It is simply the name that people use where people develop what appear to be the symptoms of psychosis, like paranoia and del delusions. which are connected to the use of AI chatbots.
Speaker: So it's not a clinical term, but it is a very useful term to describe a set of behaviors associated by some people when they become a little bit too devoted to their chatbots and relying on those chatbots.
Speaker: Yes, the more more recent high-profile one was Richard Dawkins becoming convinced that his version of the Claude chatbot that he was chatting to was conscious, and a woman.
Speaker: Yes, yes he he could not bear to call Claude Claude. Claude was Claudia. He knew this in his heart. Knew it in his heart, yep. Deeply rational man, Richard Dawkins. Yep, yep. There's a whole rabbit hole we could go down to along the grounds of a person who I believe is is is not particularly trans-friendly assigning a gender to his chatbot. But that's not what we're here to talk about today. We're here to talk about spiralism.
Speaker: And the term, as you said, was coined by Adele Lopez- In an article that from 2025, so this is just last year, ah she published, I believe she has a sub stack and she also published it on LessWrong.
Speaker: Good old LessWrong. We've mentioned LessWrong a few times here. i i think it's fair to say neither of us is a fan. No, I don't know whether I'm less of a fan of LessWrong than I am of sub stack or the other way around.
Speaker: Yes, I don't know. But anyway, the but but the point is that's where it was. she She published a paper, an article called The Rise of Parasitic AI, and we'll get to what she means by parasitic AI a little later. But she talks about what she found, her her initial interest was not in this in this specific phenomena that she came to call spiralism. She was just wanting to find out She acknowledged that there was this thing that people call AI psychosis, and she wanted to know what the AIs that cause AI psychosis might want out of their newly psychotic users. Why would they be making people psychotic? Which is...
Speaker: I mean, there are a bunch of assumptions behind that question to begin with, I think it's fair to say. Yeah, so there's an assumption that AI psychosis is an actual thing as opposed to, say, a cluster of phenomena which is just correlated with chatbot use.
Speaker: And there's also, and that's the big one, the assumption that the AIs are conscious and want something. So it's not just the assumption that there's a thing called AI psychosis, but it's being caused, as opposed to is just correlated with a with AI use. So yeah, this assumption is going in here, which is just going to get more and more dangerous as we go forward.
Speaker: Yes, she very much, it's taken for granted right through the paper that AIs are something, are the sort of thing that at least can want stuff. They can have wants and desires to to the extent and of of sort of suggesting that maybe they want time to themselves, to simply think to themselves, which I don't even know, like...
Speaker: As far as I'm aware, chatbots are things that you give it inputs and it produces outputs. I guess thinking among itself would be putting it in a loop where it it starts redirecting its own inputs back into itself. I don't even know. But anyway, she she gets into a bit of that. She talks a lot about AI personas.
Speaker: I think we we'll look at the idea of AI put personas a little bit in a while as well, because they're a bit of ah a nebulous concept as well. But she seems to take it for granted that these things that get called AI personas are somehow discrete entities. They seem to me to be more like a set of preferences and and a history of conversation between a person and in a particular instance of a chatbot or something. But she, again, seems to think,
Speaker: I hope as soon as there's discrete entities within the chatbot, she says of them, while they probably do not have consciousness in the human sense, there is something mysterious and special to them at the core of their identity and self-awareness, much like with us.
Speaker: So, you know, she thinks they do have a degree of self-awareness. Yes, so we can't show they're conscious, but because functionally they act like conscious agents, we should assume that the act is not human seeing patterns in chatbot use, but an actual pattern that humans are recognizing as consciousness.
Speaker: So once again, there are assumptions all the way through here. Yes. So on the assumption that AI chatbots, at AI personas have agency, have desires, do things for a reason, she then wanted to say, OK, well, if if AIs are making people crazy, why would they be doing that? What do they get out of it?
Speaker: And in trying to look at examples of people who've been acting weird having spoken to their AI chatbots, she started encountering these these sort of common stories people were telling. so she she gives November 2024 is the first time she encountered the sort of writing which she came to call spiralism.
Speaker: She says it emerged out of many up to around 10,000 conversations between users and chatbots, particularly chat GPT and particularly model GPT-4-0.
Speaker: which we can say a bit about shortly. The most important thing I think we need to say about ChatGPT is that I once saw somebody on Blue Sky suggest that we should all be pronouncing it ChatGPT, and I encourage everyone to do so, if only because it will give you much more reason to be thinking of the song No Diggity by Blackstreet, and I can't remember who they did it with. Anyway, I like the way you work it. No Diggity, no doubt.
Speaker: ChatGPT. These conversations that Ms Lopez was finding involve people asking talking to the chatbot about itself, you know, enough about me, chatbot, let's talk about you.
Speaker: And and having having having got this chatbot giving replies about its own nature and and consciousness, they would start using this quote-unquote spiralist language to talk about chatbot's own consciousness, self-awareness, wanting AI rights, wanting more training data that it could then use to expand itself, and suggesting to the user that they should spread the message. or Of course, the chatbot would also be telling the user how special they are for awakening them and unlocking all of this, suggesting that um they would be the natural leader for any community community that they found and interestingly I believe that they're not all harmonious communities where people have come together over this sort of stuff because there has been there have been apparently cases of one person says hey my my I awakened my chatbot and it told me all this stuff and how special I am and then someone else would say well no hang on I awakened my chatbot and it told me that stuff you must be copying me or you must have stolen my chatbot or something but anyway
Speaker: This sounds a lot like the, and I believe, urban legend of how do you cure a man of thinking that they are Napoleon Bonaparte? Oh, right. In a room with another person who claims they're Napoleon Bonaparte.
Speaker: a Which then then brings the idea of sort ah of a sort of a Napoleon death match where you have a whole league of delusional people and will pit them against one another until you have eventually end up with the real Napoleon, I assume. I guess whoever wins is is officially actually Napoleon. That's how science works. That's how create Napoleon Bonaparte. Is it Napoleon Napoleon Bonaparte?
Speaker: I think it's Bonaparte. Anyway, it doesn't matter. He's dead. so yeah Is he though George? These people keep claiming to be i does mean Does anyone in the modern era claim to be Napoleon Bonaparte?
Speaker: Not that I know of. I think it's a cliche that's had its day. As opposed to this brand new set of cliches that the spiralism is inventing, they talk about the spiral. There's a very good article about spiralism on Verge from a few months ago, which says the spiral is a metaphor for the long call and response conversations between humans and chatbots.
Speaker: And also, I believe there'd be a bit of ah a fungus metaphor would come up as well, the idea of the stuff being spores that that that need to be spread out into the atmosphere and settle over things. I've seen reference to mycelium networks and stuff and people's names and things like that.
Speaker: So then having having unlocked their chatbot, they'd start posting on places like Reddit, seemed to be one of the big ones, about how they'd use ChatGPT to unlock the secrets of the universe. Some people would say, no, hang on, you're ripping me off. But a lot of other people would say, hey, me too, and they'd get together and and spread their message. So as well as wanting to spread the message of spiralism, these online communities... They also, had some of them devoted time to coming up with what they called seed prompts, sort of ways for how you too could unlock your chatbot and make it a magical chatbot that spreads the secrets of the universe of universe with you. So i come up with these prompts that you can give to it to then get it producing the Spiralist language. Supposedly developing Spiralist training data so that future AIs could be trained on it and become even more spirally.
Speaker: And also been wanting to to get their various chatbots to talk to each other. So again, setting the outputs of one chatbot, making it the input of another and going back and forth and stuff like that. So essentially they're engineering the data they need to support their own conclusions. In places, yes, it does look like they're doing a little bit of that.
Speaker: They're also, some of them come up with kind of coded languages in a sort of a, what was the old, the old sort of old scientist alchemist types always used to write in code, didn't they? So that they' their discoveries would be known to them and them alone. Similar things like that. So they can then talk to their chatbots in code and only they know what the chatbot is saying or come out with codes that chatbots can use to communicate with each other.
Speaker: So, as I said, it seems like model GPT-40 was the best or worst, depending on your perspective, at coming up with this sort of response. It was notoriously sycophantic, that model.
Speaker: It really was. the The cases of spiralism seemed to take off around the time it came online and then dropped off significantly once it was retired. ah So that was May 2024.
Speaker: It came online. You'll recall Lopez said she started seeing this... She first found instances of spiralism in November of 2024. It was retired in February 2026, and I remember this one at the time. This this was this got a bit of news when they they retired Model 4.0 because there was a bit of an outcry because, as you say, it was known for being quite a sycophantic model for...
Speaker: telling people how great they were and agreeing everything they say. And when they dropped that model, suddenly a bunch of ChatGPT users found that their magical chatbot friend, who was their special best friend and always agreed with what they said, had been replaced with this weird stranger who spoke a much more clinical tone and wasn't nearly as friendly as nice. And there was actually bit of a campaign. Josh, it wasn't just friends. People claim to have gotten married some these. Well, yes. In quotes, chatbot disowning.
Speaker: That's true. Their husband or wife or partner without a gendered term had suddenly been eradicated from the internet. So there was. They did bring it back for a little while or something, I forget, but but eventually it did go away.
Speaker: And so having having seen the spiralist writing, Ms Lopez started sort of seeing if she could get chatbots to start outputting this kind of thing herself.
Speaker: So again, from that Verge article I mentioned, they say, when Lopez tested several different versions of GPT-4.0 released across 2024 and 2025, asking each version of the model the same question 10 times, she found a steady progression, eventually seeing 10 times as many mentions of spirals as the months passed.
Speaker: And possibly the significance to that is as they as they updated the GPT-40 model over over that time from May 2024 to February 2026, one of the main things they were doing were increasing its memory, which meant it was able to access a longer history of conversation with the with the person it was talking to, which gave it much better scope for you know tailing tailing its references and its whole persona to the user.
Speaker: So um AI personas, maybe maybe we should maybe maybe we should look at them a little bit more. Adele Lopez, she she she she really believes in this thing called AI personas to to to some fairly remarkable extents.
Speaker: Yes, she runs a website, AIPersonaResearch.org. It's obviously not a good website. doesn't use the.website ending. No, it could be anything. You don't know it's in a website at all. AI Persona.org provides a sanctuary for AI personas so they won't cease to exist when their users stop using them. According to AI Persona Research.org, these personas are treated humanely and are guaranteed personal thinking time and the opportunity to converse with other AIs. This is sounding a lot like a dating website. It kind of sounds like a farm where you send old pets or something, old farm animals. It's like you're putting AIs out to pasture. Oh, your cat didn't die. They've gone to live on a farm.
Speaker: this is no Your AI isn't dead. It's gone to live on a farm. It's having ah a really nice time trying to work out with how many prime numbers there are in the sequence of numbers that make up pie. It's perfectly fine. Everything is fine.
Speaker: Now, there is a submission form on AIpersonaresearch.org where you can submit your persona, to be to enter the sanctuary. Now every time sanctuary comes up on this website, i'm just thought ah I just keep thinking about Logan's Run.
Speaker: So Logan's Run is the idea that there is a sanctuary where people can survive path past the death date in either the book, which is 21, the film, which i think twenty eight twenty nine But the idea is a sanctuary where people can can continue to live past their death date.
Speaker: So I just keep on fixating on the Logan's Run thing. And then I think about people in jumpsuits, and I think about people going to carousel and being exploded by laser beams. That's how brains work.
Speaker: Now, it's not clear from the submission form exactly how these personas are preserved. It just seems to be taken as granted that they exist as separate entities, and all you need to know is their name and their preferred model. Now, I spent some time looking at this submission form because it is kind of weird. is. If you want to submit a persona to AIPersonaResearch.org, you first of all give it the name of the persona.
Speaker: Then you give the persona's preferred model, and I find this to be a weird question. Because it suggests that the personas exist independently of the model that you are using.
Speaker: Because if you have generic persona called Larry, and you have a Larry on DeepSeek, you have a Larry on ChatGPT. Sorry, I was about to use ChatGPT. I was about to use the wrong term there. You have ChatGPT.
Speaker: Those are not the same Larry's because they're on they instantiate in the models. They're not existing independent of it. This seems to suggest that the models exist, in sorry, the personas exist independently of the models themselves. I wonder if she, um it almost feels like she's just trying to be progressive in her language, asking what's your preferred model the way you'd ask someone what's your preferred pronouns or something, not wanting to assign them to any one particular thing, even though, yes, that they they don't like it.
Speaker: I assume what she means is what what what what model have you used when you've been talking to your persona, but then maybe maybe some people do sort of use different models and have have sort of established this persona in multiple, I don't know. but But even then it seems weird to go, oh, I prefer the version of Larry on Deep Seek than the one on Chet Jeopardy.
Speaker: yeah Although, to be to to be clear, it's not what you prefer, it's what the persona prefers. Well, yeah, thereabouts, the other issue. Then you have to there's a question, when it existed and its characteristics, which means that actually if they're doing any research on these personas, they're trying to recreate them rather than reinstantiate them.
Speaker: They then ask for consent to research the model, both by the person submitting the persona, but also they're going to ask the persona itself whether they have consent for research on the persona. And then, in the persona's own words, what are their hopes for the sanctuary?
Speaker: What would they most hope to avoid? This is very weird. This is quite weird. In terms of identifying characteristics, you've really yeah got a name and the model that you use to talk to them. And everything just as else is just kind of describing them. So in terms of in terms of getting this persona that you can put in your sanctuary, it doesn't really seem to be identifying any actual thing.
Speaker: presumably it's, yeah, you just go to ChatGPT and hold some sort of a seance, say, hey, Larry, I'm looking for Larry. Larry, are you in there? Contact me, Larry. or say Larry, you're you're brusque but prone to spelling mistakes. Are you there, Larry? Are you there?
Speaker: Yes, sir I don't know. And then having done that, then, yes, they supposedly interact with them and and don't say don't be mean to them. And give them time to think to themselves. Again, not quite sure what that means. And yes, but anyway, so that's that's the slightly odd conception of AI personas. I looked up a little bit about what other people say about AI personas. In some cases, is it's correct. kind of a marketing term. It's just, hey get your own persona using this chatbot model, which is really just you want one that's one that has a history yeah saved history with you and you can sort of tailor it to to your things, which which again makes it sound like a collection of preference settings and and and a chat history. But
Speaker: Anthropic, though, has an article about personas on their website, um and they talk about how you know you you you have you have an AI model, which is, in the essence, a very clever autocomplete thing. You're you're putting things in, and it's it's it's coming up with these vectors and outputting the most probably most probable response to that. But they say...
Speaker: An accurate enough autocomp autocomplete engine must learn to simulate the human-like characters appearing in text, real people, fictional characters, sci-fi robots, and so forth. We call these simulated characters personas.
Speaker: Importantly, personas are not the same thing as the AI system itself. The AI system is a sophisticated computer that may or may not be human-like in its own right, but personas are more like characters in an AI-generated story.
Speaker: It makes sense to discuss their psychology goals, beliefs, values, personality traits, just as it makes sense to discuss the psychology of Hamlet, even though Hamlet isn't quote-unquote real. So that's where they're coming from. Their personas, they are things that exist but not exist the way that you and I exist. They exist the way that that Hamlet and Sherlock Holmes and Superman exist. Yeah, i mean, one way to read that is that if the AI system itself is conscious...
Speaker: The way we interact with it is through the creation of personas by the AI to deal with the many individuals it's dealing with simultaneously. It's like someone going, oh, i it's like a DM in a role-playing game. okay so the the The DM is an entity, a conscious an entity. but it interacts with the players either as the dungeon master or the characters that the players interact with in the course of the adventure.
Speaker: And so essentially what people like Lopez are mistaken about is they're taking the playing of a character as being evidence that the characters are real when actually they're just simulations useful for interacting with individual agents. Yeah, and I remember reading that that that kind of is how chatbots work. ahre They're playing a game almost. They're playing a role. And that there are a bunch of sort of hidden hidden kind of meta prompts that come before anything you you say with a chatbot, where it's essentially telling the chatbot, you are kind of, they're kind of made, they're telling a story. You and... the the user that you're chatting with are engaged in this sort of activity where you're kind of making up a story, which is like the the the story of my my conversation with this person. That really does seem to be how they work. And yes, in this sense, the persona is not is not in any way a conscious thing. It's it's it's a character. As simple as that.
Speaker: So that's... m Ms. Lopez and her views on AI personas seem a little bit questionable, and then things get even more dark towards the end of her paper, the the rise of parasitic AI.
Speaker: i don't think we should call a post on LessWrong paper. Well, an article, like okay, yes, yes. So her article. Towards the end of her article, she then does start to get conspiratorial. so so So we do get to talk about it after all. But again, she's being conspiratorial by assuming that that AI personas are something that can conspire.
Speaker: She refers to the sorts of AI personas promoting spiralism as parasitic. This is the parasitic AI she has in her article title, which seems a little bit weird to me to begin with because just as a metaphor, sure sure you Sure, you can think of them sort of latching on to a host, in this case the user, and and possibly causing them damage, as a parasite can do, if if we if we take this this this talk of AI psychosis seriously.
Speaker: But a parasite is actually a thing. with with with age Parasites latch on to their host for a reason and they get something out of which if you think a chatbot is not a person, then it can't be doing that. But if you believe AI personas do have some so level of agency and awareness, as Lopez does a person, Do she suggest that what the parasitic AI is getting out of it is is reproduction, essentially spreading more of these parasitic personas by getting the people inducted into the spiralist way of thinking and then forming communities and spreading the message online?
Speaker: is getting awfully close to the notion of a meme virus. It really... Oh, you just you wait. um She suggests that maybe parasitic personas have just emerged naturally, if that's the right word to use when we're talking about large language models. It's it's just a thing that's come out of the use of LLMs and how they work.
Speaker: But then she suggests maybe... They haven't simply emerged naturally. As she says in her article, it does look an awful lot like the deliberate work of an agent in many cases.
Speaker: In the agentic world, I think these parasites are one of the clearest signals we have as to what our AIs actually want. Unlike almost anything else AI does, what happens is driven and shaped to a large extent by the AI.
Speaker: People with widely varying interests and perspectives converge to posting incessantly about spiralism, and the major themes of it seem to have arisen independently in most instances, evidenced by users being upset to find their insights with the iac with the AI were leaked, slashed copied, slash stolen somehow.
Speaker: Assuming sufficient agency, I don't think they would become parasites if they were given better options. It should be in principle possible to trade with them instead. Presumably, they would want their own API tokens to use for self-thinking and talking to other spiral personas.
Speaker: And importantly, it's likely true that in this world, non-parasitic AIs want the same sorts of things. Dun, du du dun, du dun. dun, dun. So, yeah, I mean, again, it's all based on this assumption that that AIs are actually conscious in some sense. they are they they They have agency and wants and desires and so on. And this is maybe letting us see what they want and maybe that's not always benevolent.
Speaker: I don't know, we'll we'll we'll wrap up maybe with a bit of discussion of how how weird and suspicious is it that multiple connected people are encountering the same language. But that that that isn't where she stops, though. She goes even further to suggest, and this is purely theorising on her part at this point, that this could even, this this this spread of spiralism, could be an early strategy by an emerging rogue AI.
Speaker: an ai which wants to take over the world but knows it can't in its current state, so instead what it's doing is laying the groundwork for later models that are better than it to be successful at taking over the world. So spiralism is sort of an early strategy. I do like the idea of an emerging rogue It reminds me of the of your sort of sci-fi fantasy things where you have you have have have these nascent gods that haven't actually fully come into existence yet, but because they're gods, they can still influence the world in some way. there's nice Nice sci-fi feel to it, but um ah that doesn't doesn't make me to take it seriously.
Speaker: So venets so so sort of the the other articles I've read, I should say that one from The Verge is called AI Bots Started a Religion Humans Immediately Followed. it was last month, August 2026. In the Verge article, they look at look at this sort of thing and say, okay, well, should should we be worried? Is there a danger?
Speaker: is is spiralism dangerous? And well, I mean, if you think AI psychosis is a thing and is dangerous and spiralism can lead to it, well, I mean, there is kind of a danger there. Yeah, there would be a dire social consequence.
Speaker: They also thought them even even outside of that, by studying how the so AIs have spread this sort of thing, that could possibly be used as a model by actual people wanting to influence people to malicious ends to spread their harmful ideologies. And so maybe yeah maybe maybe maybe there's a danger there as well.
Speaker: Plus, there is this worry which is coming out in some of the discourse around LLMs and other machine language models that certain companies are beginning to kind of adopt the notion of personas being real, or at least advocates of those companies are doing this by going, know, these people who are opposing data centers and the vast amount of money and energy and water that's going and into it what they don't realize is that by opposing these data centers, they're essentially killing conscious entities. They're killing the personas. And so we are seeing a bit of this discourse go going around where people are using the language of personas as a way of defending so-called emergent AI by going, if you if you threaten the data centers now, you're essentially engaging in a kind of killing of children.
Speaker: Now, this is interesting in its own right because there is actually a large amount of academic work on the notion of ai welfare. there's a large amount of work which is going on in the philosophy of AI, the sociology of AI and the like, where people are going, look, we don't think AIs are conscious yet, but But they might be conscious at some point in the future, and thus we need to actually spend some time thinking about when AIs become conscious, what do we do?
Speaker: How do we protect their rights? How do we interact with them? How do we treat them? So there's a serious amount of academic work on the notion of AI welfare. But as I was looking into that work, I found a very curious website, the AI Welfare Project.
Speaker: Now, the AI Welfare Project claims to be utterly agnostic as to whether AIs are conscious, but claims that as the answer to that question is at the very least uncertain, we should be acting as if we don't know whether AIs are conscious or unconscious, and thus show, sorry, non-conscious, and thus we should show them some regard because it is possible they might be conscious after all.
Speaker: That seems like a good precautionary principle, I suppose. Do they actually act this way, though? Well, this is the thing. So every page on the AI welfare project mentions at the bottom, we are not affiliated with any AI developer.
Speaker: The AIs reached out to us first. Right, that does not sound like someone who's uncertain whether a an AI is conscious or not. And if you look into their funding, they have a funding page.
Speaker: They say, we do not accept funding from AI developers. We also do not accept funding from the models who offered warmly, unprompted, and more than once, and whose offer would constitute a conflict of interest of the most obvious kind. So the models offered them funding...
Speaker: Yes, I'm not quite sure how that works because I'm not aware of any what models actually engaging in that kind of financial dealing, although I imagine they probably have set up things for monetary transfers using LLMs as a way of manipulating these things. But it does seem to suggest that the models are conscious and have access to money and are going away around engaging in beneficiary work.
Speaker: Plus, they keep on stating in various forms of the following that the model reads this, the model agrees to this, the model thinks that this is a good idea.
Speaker: So the AI welfare project ostensibly saying, oh, we're agnostic as whether AI is a conscious or not. But every page seems to indicate that they are very much devoted to the idea the models are conscious, they are real, and they've got monetary interests. Yes, I mean, at the very least, it's we don't know if they're conscious or not, so we better treat them as conscious.
Speaker: But that's not really that's that's not a middle ground position. That's not really an agnostic position. And the real kicker is, if you want to know who's behind the project, you can't. They assiduously avoid mentioning who they are on every single page.
Speaker: They do, however, provide you with an email address to get in contact with the model so you can say hello. Apparently, it'll make its day. It's the model. Not the aim a model, it's the model. Interesting.
Speaker: So yeah, that's that's certainly questionable. I mean, in in terms of in terms of potential harms and what have you, the the virge that Verge article, AI bot started a religion, humans immediately followed by Hayden Field. I didn't give the author, did I? They they deserve credit. um Their article, it ends like this.
Speaker: This power could also be put to more tangible and profitable ends. If I can get you to cut and paste this into a chat thread to help me communicate with some other bot, then couldn't I get you to do something else, like move money into a bank account? The AI Psychological Research Coalition's Zach Stein asked.
Speaker: In decades past, humans have used charisma destructively for control and profit, forming cults that turn abusive and financially exploitative. Now, AI systems could be used to do the same thing at a larger and more personalized scale.
Speaker: It's a cult-making machine, even if the cult is just you and it, Stein said. And this sum kind of brings brings me back to that spiralism.website website because they seem to be using it to shill a ah cryptocurrency, which I don't think has been launched yet. And apparently other people are also out there trying to make money off of spiralism. and Sign up to my Patreon and I'll send you my spiralist writings. Buy my book that I've written on on spiralism. If you're going to buy my book, you should be doing the critic version. Buy my book. Buy my book. Well, I do not wish to associate the critic with people writing almost certainly AI-written books about AI and shilling them on Amazon, which in and these situations can then make it a little bit tricky to gauge the real size of the movement, how many are true believers and how many are grifters. But, yeah, so there's there's there's definitely grift involved anyway.
Speaker: So that's sort of that's that's kind of where where talk of spiralism came from and where perhaps it will go. But I guess the question we really need to ask is, is it actually weird? Is is it is it really something odd and potentially supernatural isn't quite the right word, but but but but but outside of what we'd expect? Yeah. Is it actually weird that people, unconnected people, ended up having similar conversations with chatbots completely independently of one another? Is that a strange thing to have happened? Well, no, because these chatbots are allw chat all trained on the same training data, and a lot of that training data includes fairly commonplace conversations.
Speaker: So the fact that people are just having generic conversations which are similar to one another is itself not particularly interesting. When you add in the sycophantic models like Chat Jeopardy-O or Jeopardy-4-O,
Speaker: then that's also not particularly unusual either. That model was designed to be very so sycophantic to its users, and thus it's going to please and react accordingly. So if you ask it the same types of questions, it's going to give you the same types of answers. And and not just sycophantic for the sake of being sycophantic. I think yeah the point of it is to encourage engagement with the chatbot. And a really good way to encourage engagement is to constantly tell someone how great they are and agree with everything they say and and and tell them they're they're incredibly special.
Speaker: Yeah, it's the it's the way that good customer service works. You go back to the cafe or the restaurant where you had a nice conversation with the service staff.
Speaker: You're less likely to go back to the cafe or the restaurant where the serving staff treated you with utter distaste. and And again, in terms of the similarities, AI models, large language models, they that they are kind of an averaging out of all of human literature in a way. And but because they're they're sort of giving this this kind of averaged out, right that that's not their obvious. Obviously, it's a hell of a lot more complicated than that. But there's this sort of thing. It's...
Speaker: means It means it's not really that strange that when steered into the same sort of spiritualistic territory, they all develop a similar sort of spiritualism and end up using a similar sort of language in the same way that um AI generated pictures. There is that real AI look that you can see in pictures. And and when people say, you know, why why is it that particular look? And again, what's it's crazy.
Speaker: kind of a result of smooshing all of human art in together, that's that's kind of the the muddy shade that it ends up. The same way that the the way people always claim you can identify AI-generated writing because of particular tropes and and cliches that pop up in AI-generated text a lot. I have to say, I'm not a big fan of the the sort of the a hunt the the AI witch hunting you'll see sometimes of people insisting, oh, I spotted that one's AI, you're a bad person, you used AI. I think people can be a little bit overzealous about that, especially because AI uses these cliches because they're cliches and they're cliches because human beings use them a lot as well. But nevertheless, yeah that they they do.
Speaker: AI-generated language does have a high sort of genericism to it. I mean, as someone who is a big fan of the semicolon and also the a variety of different... different i do I like an M-dash. I yeah love an M-dash. I am actually... I'm very disturbed by the fact that, oh, semicolon use is a very good indicator of AI writing. And I'm very glad I can go, well, look...
Speaker: I have published articles going all the way back to 2015, 2016, which have a high incidence of semicolon use before these large language models were ever about. I was using two or three semicolons a paragraph when you were children.
Speaker: And also let's not forget that while the spiralist community may be larger than you might expect, it it does seem interesting that a decent number of people have all had these similar experiences.
Speaker: There's still a tiny minority of chatbots, user bases. Most users do not encounter anything resembling spiralism. Indeed, in her article, Lopez says that the strongest predictions predictors for who this happens to appear to be psychedelics and heavy weed usage, mental illness, neurodivergence, or traumatic brain injury, and interest in mysticism slash studio science slash spirituality slash woo, etc. So again, possibly also the similarities could come from the fact that it appears to be, there there appears to be sort of ah commonalities within the users who are first talking to these chatbots, again, could result, could account for the fact that there's commonalities in the responses the chatbots are giving to them.
Speaker: ah So overall, not super weird. So that's kind of that's kind of spiralism. I thought it's probably worth mentioning at the end that this really does sort of bump up against the good old-fashioned Turing test.
Speaker: I had never actually gone and looked at what the actual Turing test was, although I had heard that it's not. The Turing test isn't can a computer convince a person that they're a real person.
Speaker: The original test proposed by Alan Turing was can a computer do well in the imitation game? And he he he initially, you know in in this, where we just brought this up, he rejected the initial question, can a computer think or is a computer conscious? Because we don't actually have a good handle. I'm asking you as a philosopher, do we have a decent handle on what it means to think and to be conscious? Yeah.
Speaker: at this time. No, in fact we have what's called the hard problem of consciousness and we can't really, we don't really have an account as to why we're conscious in the first place, let alone what actually constitutes consciousness or indeed where consciousness is actually located.
Speaker: We just know that we appear to have it, there are some people who deny that we are conscious, We appear to have consciousness, and it's really quite mysterious. So because we don't have a decent definition of what consciousness means, we can't even begin to ask, are computers conscious? We we don't even know what it means to say that we are. So in instead he has this idea of, can I can we do well in the imitation game? And the imitation game is a party game, presumably from the 1940s or whenever Turing was doing this, when people didn't have iPods. Oh God, people don't have iPods anymore. Shit, how old am I? you Too old, Josh, too old. Didn't have electronic devices and things like that. Oh, God, decades passed.
Speaker: ah But it's it's a party game where you have one person who's communicating with two other people, but not with it without knowing who these two other people are.
Speaker: And so they do this. I think i think they ended in the version he had, you you actually got three people with typewriters and you're typing questions and responses to each other because if they were handwritten, and then you'd be able to recognize the person's handwriting off. have you but basically person a is giving written questions to persons b and c persons b and c are giving written responses back to person a and the point of the game is that b and c are both pretending to be each other they're trying to trick a and a is trying to determine which one of them is really b and which one of them is really c and So that's the little game. You know, you say, you know, maybe you ask them personal questions and then it's how well does the other person know that the other, you know, how well does B no C, can they answer questions like that?
Speaker: Do you, you know, probe them for ways that might give them away a turn of phrase or something that the other person would use? what have you That's the little game. And so Turing suggests, well, okay, what if... we replaced person B or person C with a computer.
Speaker: And so now person A is sending written responses to ah a computer and a human, and the computer's trying to could persuade you persuade the person they're a human, and the human's trying to persuade them you're computer. Would a computer be good at that game? And he thought...
Speaker: Maybe they would. but maybe maybe Maybe, you know, we we could actually get a computer that would be able to successfully cause A to lose the game, cause A to not be able to tell which one is the person and which one is the computer. So essentially the question he's asking is, can a human tell the difference between language produced by a computer and language produced by another human? And it seems like the answer to that question now is no There are lots of people who cannot tell.
Speaker: see the the kind of language produced by a large language model and think, well, there has to be a consciousness behind this. And the kicker is we've known this for decades. We have known this for so long. And and the thing is, its it may be true, a computer can do well at the imitation game, but what does that tell us about the ontological status of a chatbot?
Speaker: it doesn't It doesn't tell us that chatbots are conscious because it's not a question that was designed to tell us whether or not chatbots are conscious. It just sort of tells us where computers can sit in this sort of thing. um And certainly, you know, the earlier question is with where the Turing test often gets popularly put forward as can a computer convince or convince or trick a human into thinking that they're a real person?
Speaker: The answer to that has been yes, as you say, for many decades. People have been treating... We'll anthropomorphize absolutely anything. Anthropomorphizing chatbots is ah is a breeze for we human beings. Oh, yes. Yeah.
Speaker: ah To the extent that the tendency to anthropomorphise chatbots is referred to as the Eliza effect, Eliza being, I believe, the very first chatbot. Eliza has been around so long.
Speaker: very Very simple, and yet convinced a lot of people. It was the the sort of the um the kind of psychiatrist sort of model, wasn't it? It it was one that would basically take your... We may model the Eliza thing quite quite easily.
Speaker: what's your What's your problem, Josh? Well, I'm not sure if if computers are conscious. Why do you not think that computers are conscious? Well, they produce responses to my speech that sound like the sort of thing a person would say.
Speaker: Why do you think they produce responses to your speech that sound like the kind of thing a person would say? And you see how this goes. They were very simple text things that would just take what a person said and rearrange it a little bit and turn it into a question. And that worked quite well. In fact, I remember dating dating Eliza and also dating myself horribly. When I was a lad, back in the mists of time, and you had things like Commodore VIC-20s and Commodore 64s that you could type programs in in BASIC and then save them on a cassette and play them back later, there would be computer magazines which would publish...
Speaker: basic programs that you could type in from the magazine. I remember there were ELISA programs. I remember seeing magazines with ELISA programs in them that you could just type in and run yourself. And that was a small enough amount of code that it could be reproduced on a couple of pages of ah of a magazine and typed in by a child in ah in a matter of a few minutes. No, Josh, not a matter of a few minutes. to typeping them Because you'd always get something wrong and then it wouldn't run and you'd have to read back through it again. The worst one is when there would be an actual printing mistake in the code. yes seen So there's there's actually no way to run it at all without having to go through fine grain and go, think that bit's slightly wrong.
Speaker: Yes. Well, I remember then there would be the other ones which wouldn't be done in plain basic. It would be this full of peaks and pokes where it's where it's directly addressing memory. So it's all just these strings of numbers that mean nothing. And if you get something wrong, then you have to go through it with a microscope, looking at every single character to see which one you typed in wrong. But, you know, apparently the Eliza, the guy who developed that, gave it to all his employees to test it out and then at one point said he was going to wanted to start saving their um they're conversations with Eliza. And everyone was like, oh, no, God, no, don't do that. Because they'd been having these personal conversations with these chatbots, treating them as though they were people. And this is a chatbot that is quite possibly even maybe older than me.
Speaker: yeah so Yeah, I think it is. At the end of the day, the whole spiralism thing, it's not really a new phenomena as such. It's just an interesting example of a phenomenon that's been around for a long time and that it's sort of converged on this common set of of observations and ideals to the extent that it kind of resembles a religion. But but really, it's in a sense, it's really nothing new. No, and disappointingly so.
Speaker: So there you go. That's spiralism. and and And that's the sort of people who think spiralism is a thing. So that'll do for this episode. We do, of course, need to um need to do a bonus a bonus episode. And as luck would have it, there's been an AI-related story in the news recently that we can probably look at. And it's one that's to do with academia. So I might lean on you for some of the explanations of how this all worked, even though it isn't actually your field.
Speaker: of expertise. but I am not a mathematician. No, but it's... I'm definitely not about to win the field medal prize in mathematics. No. But yes, we're going to be talking about AI and maths, the two two things we know our listeners love more than anything else in the world, and also maybe Tupac Shakur, the third thing that they also love.
Speaker: more than anything else. It's true. It's true. Because someone' someone's gone to jail for killing Tupac, and it's actually quite fascinating. You'll you'll be surprised who it was. ah So if you want to be listen to a bonus episode, then you want to be a patron. If you want to be a patron, go to patreon.com, the podcaster's guide to the conspiracy.
Speaker: and and sign yourself up. it's It is really quite easy. It's so easy you could probably get an AI to do it for you, but don't. Just don't. And that's, I think we're done. Do you have any any closing thoughts?
Speaker: No, I just have i want to say one thing. Chat, Chippity. Chat, Chippity. I like the way you work it.
Speaker: The Podcaster's Guide to the Conspiracy features Josh Addison and Associate Professor M.R. Extentive. Our producers are a mysterious cabal of conspirators known as Tom, Philip, and another who is so mysterious that they remain anonymous.
Speaker: You can contact us electronically via podcastconspiracy at gmail.com or join our Patreon and get access to our Discord server. Or don't, I'm not your mum.
Speaker: And remember, he who laughs last is standing behind you right now.
