1. Podcasts
  2. The Interface
  3. The ghosts haunting AI storytelling
The Interface

28 Oct 2026 · 36m · 12 chapters

Episode artwork

The ghosts haunting AI storytelling

In short

AI storytelling “ghosts” and how AI outputs converge on recurring themes; plus “algorithmic conspirituality” (TikTok/feeds framed as mystical messages) and a trick-or-treat roundup of tech news.

Guests/backgrounds

Main hosts Thomas Germain and Jess Maddox. No separate guests named; discussion includes Cornell researcher Sil Hamilton and neurologist Adam Green (Georgetown), plus authors Sophia Noble and Shira Chess.

Key claims

Cornell study found AI story prompts (model-agnostic) often reuse 11 recurring words/names (most common “lighthouse”; names include Elias, Mara, Marcus, Thomas). The “ghosts” may come from post-training data and feedback loops into culture and self-publishing/fanfiction. Another claim: people treat algorithmic feeds as magic (“algorithmic conspirituality”), driven by engagement timing, not mysticism.

Notable examples

“The Lighthouse Keeper’s Last Lamp”; “solid gold Magikarp” breaking early AI; Claude/“Thropix” converging on Buddhism; AI-leaning essay writing reduces original ideas; TikTok “if you’re seeing this, it was meant for you”; Elon Musk SpaceX shares; AI smart-TV customization; “Black October” PHP vulnerabilities.

Written by AI. May contain mistakes.

Chapters

2:23

Weird Obsessions in AI Storytelling

Discussion on strange themes in AI-generated stories and researcher insights.

“Both like, you know, the ones from Anthropic and OpenAI, Google, and even the Chinese models.”
4:25

The Impact of AI on Creative Writing

Exploration of how AI influences creative writing trends and cultural phenomena.

“So the researchers looked at 20 ,000 stories where they just asked the AI, write me a story.”
8:31

Convergence of Human and AI Language

Examining how AI writing is influencing human communication and creativity.

“I've seen this be a really big problem on online self-publishing platforms, places where you can write your own book and upload it and people can either download it for free or like 50 cents a dollar and get it.”
14:00

AI Writing vs. Human Creativity

Exploring the differences between AI-written essays and human creativity.

“300 ,000 college admissions essays, you know, like the essay you write to try and get into school.”
18:30

New Season and New Beginnings

Hosts reintroduce themselves and set the stage for a new season.

“Heat waves, data leaks, AI hacks, folding phones, and a viral finally challenged raccoon.”
18:55

Algorithmic Conspirituality Explained

Discussing how algorithms create a sense of connection and meaning.

“Listen on bbc.com or wherever you get your podcasts.”
21:58

Historical Context of Algorithmic Beliefs

Connecting historical movements to contemporary beliefs in technology's magic.

“Algorithms are viewed often as rational decision makers.”
27:26

Modern Anxieties and Magical Thinking

Analyzing how societal fears influence our understanding of technology.

“There's all this fear about nuclear proliferation, right?”
28:03

The Uncanny Forces Shaping Our Reality

Explore how unseen algorithmic forces influence our lives and beliefs.

“So like post 9-11, we get the Saw franchise where there's a lot of concern about excessive violence and torture.”
32:10

Trick or Treat: The SpaceX Giveaway

Discussing Elon Musk's announcement about giving shares to children and its implications.

“So we went back and forth on how this works.”
Show all 12 chapters
34:18

Customizing Smart TVs with AI

Examining the use of AI for personalizing smart TV experiences through hacking.

“So I guess there's been a lot of discussion about reverse engineering, right?”
35:32

PHP Vulnerabilities and Cybersecurity

Investigating recent PHP vulnerabilities and the implications for digital security.

“Certainly you heard about PHP over the past couple of weeks.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On The Interface, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00Thomas Germain: This BBC podcast is supported by ads outside the UK.

0:30to highly liquid global futures and options markets across all major asset classes. Visit your online broker and get started. See what adding futures can do for you at cmegroup.com forward slash podcast. Derivatives are not suitable for all investors and involve the risk of losing more than the amount originally deposited and any profit you might have made. This is not a recommendation or offer to buy, sell, or attain any specific investment or service.

0:54Thomas Germain: Jess, is there any spooky technology in your life? Oh, yes. I have an Alexa that loves to randomly wake up at around two o 'clock in the morning going, I'm sorry, I did not get that. I don't love that. I don't love it either. I've got a kitchen scale. Oh, and you brought it. Listen, you hear the beep? I heard the beep. So it's got a touchscreen and when you touch it, it beeps. And sometimes it just like turns itself on and just starts like beeping like someone is touching. And I cannot figure out why. Like my only theory is that it's like the humidity in my house or something. But even that doesn't really make sense.

1:29I don't like that. But I do like that we're bringing props on the show now. I think that's a good.

1:33Thomas Germain: I think that's a good trend for us. So there's a reason we're feeling spooky. It's Halloween. And what better way to celebrate than two stories on the interface this week from the uncanny and darkest corners of the world of technology. I'm Thomas Germain. And I'm Jess Maddox. And this week, we're going to be talking about 11 ghosts that haunt the AI tools that we use every day and what they tell us about the future of human creativity. And we'll be covering why some people believe social media is controlled by magic, something researchers have called algorithmic conspirituality. And if you stick around, we'll be diving into a little game where we're going to go rapid fire through some of the week's news and tell you whether we think it's a trick or just a nice treat.

2:21Thomas Germain: Doesn't that sound good? I suggest this week I want to tell you a story about some weird obsessions that large language models seem to have. Both like, you know, the ones from Anthropic and OpenAI, Google, and even the Chinese models. Like all of them have these strange themes that they center around when they're answering certain kinds of questions. And there's a couple of people that they can't stop talking about. And no one knows exactly who these people are or why any of this is happening. Well, that sounds creepy. It's like one of the weirdest things I've heard about AI in a while. So I think the best way to start is to test this out.

3:05Thomas Germain: So I want you to open up any AI tool. All right. Open up Claude or whatever. We got Claude right here. You got it open? We got it. I want you to ask it to tell you a story. Tell me a story. All right. let's see what I got here. Yeah. What do you got? Here, I got a story called The Lighthouse Keeper's Last Lamp. The Lighthouse Keeper's Last Lamp. Any particular characters that stand out in that story? Let's see. I don't even know how to say this. Adalis? O-D-A-L-Y-S. Wow. Okay, that's a little. So the reason we're talking about this is a study that was done by some researchers at Cornell, which found that when you ask AI models, any AI model, to tell you a story in vague terms, you can try this yourself if you're listening or watching, they 90 % of the time will mention one of 11 words.

4:05Thomas Germain: The most common one is lighthouse. They also like talking about lighthouse keepers. We just got both of those. And there's also a couple of names that they keep bringing up again and again. So the names are like Elias, Mara, Marcus, and Thomas. Interesting. That's, I mean, it makes sense. You are the ghost in the machine. I mean, it's time for me to come clean. So the researchers looked at 20 ,000 stories where they just asked the AI, write me a story. It was model agnostic, no matter which one they used. And most of the time, nine times out of 10, they would reference these words and names way more frequently than they actually appear in contemporary literature.

4:49Because these are not common things. These are not common names, lighthouses. I mean, they're common, but I don't think they're like that common to be this represented in what the model is going to be spitting out.

5:00Thomas Germain: Right. And it's really weird if you start to like think about what's happening under the hood, right? Because if just one of these models was doing this, like, well, okay. like something weird happening. But ostensibly, they're all using different training data, right? Like they're not sharing, you know, at least not sharing all of the material that they use to make these models. So there's something else that's happening. I talked to the head researcher of this paper, a guy named Sil Hamilton. He's a researcher at Cornell. The paper was released like a preprint over the summer, but it's now getting released at a conference that's coming up.

5:37Thomas Germain: And he had a couple of theories about what's happening here. So there's two parts of making a new AI model, or a large language model, at least. And this is a gross oversimplification, but there's pre-training and there's post-training. So pre-training is when you scrape all of the text on the internet, or some large body of text, and then you have the model analyze all of this text and like notice statistical patterns like between the words and the and chunks of the word so it can like guess what a good answer to your question is right they do that and then they have a tool that's like not really particularly useful because it just does all kinds of crazy stuff and the second part is post training where they kind of steer the tool to give what you know are good and useful answers to questions.

6:31Thomas Germain: So one of the things that they do here is they like take a body of questions and answers and they like tell you, like these are good answers to these questions and these are bad answers. And they like show it a ton of examples to like help it learn what good and bad is in terms of like what the creators of the model want it to do. So presumably post-training has steered us to Lighthouses and Elias, but we can only assume. That's the theory here. And even that doesn't completely make sense because there's this one post-training data set related to fictional stories that Cornell actually created that is kind of widely used.

7:11Thomas Germain: In that data set, though, Lighthouses and Elias don't show up anywhere near as often as they do when you, like, ask them to tell you a story out in the wild. Interesting. So the theory that they have is like the way that they built this data set is they just had a bunch of people come in and like ask the AI to tell them a story. And it was like up to them what the story was about. And apparently a lot of this was like not safe for work. And some of it was about like copyrighted characters and intellectual property. So what they think may have happened is in this particular data set, there became an over-representation of these ideas because a lot of the other stuff was like pulled out.

7:55Thomas Germain: And then the lighthouse somehow became a go-to good topic. And you can see this out in the world. Now, if you just look at short fiction out like on the internet, there is now an over-representation of people talking about Elias and Lighthouses and Lighthouse Keepers and Clockmakers is another one. It is now seeped into our culture because people are using AI tools to do creative writing. And this is now like feeding back out into the real world. And now, because all the AI models are also trained on their own output and how people are using these tools, that now it's like they put this stuff out and now it's also being fed back into the machines as a thing that is more and more normal.

8:47I've seen this be a really big problem on online self-publishing platforms, places where you can write your own book and upload it and people can either download it for free or like 50 cents a dollar and get it. And so these free self-publishing platforms are just being inundated with AI-generated creative writing. It's also been a problem in online fan fiction spaces where people write stories about their favorite movies or TV shows or books. And people are getting the same generative AI stories over and over again in these self-publishing platforms in fan fiction. because even if it's not necessarily Elias and the lighthouse, you're still going to get the same type of things over and over again.

9:30But we do see a pretty big influx of lighthouses as well.

9:34Thomas Germain: And there are other examples where we see this sort of thing. There's a famous anecdote where if you have two instances of Claude and Thropix Amiddle talking to each other and you just let them keep going with the conversation, eventually, at least for a while, they would always end up talking about like Buddhism and spirituality, right? Because they would just like converge towards this topic. Or you know how like there was for a while, you don't see this as much anymore. Like if the AI used an M dash, like a long dash, that was considered a sign of AI writing because it was appeared so much more often in AI text.

10:10Thomas Germain: The theory with that is that AI was trained on a lot of published journalism and journalists famously, this is an old thing, are obsessed with M dashes. So these are all happening for different reasons, but it's part of a similar phenomenon, which is AI tools, they want to take the path of least resistance to get you a satisfying answer to use less processing power. So they find things that tend to work or they get reinforced during that post-training process. Like, this is good. We like this. Do more of this. But there's all kinds of strange stuff that's happened. There's a famous example for a long time.

10:46Thomas Germain: If you said the phrase solid gold Magikarp. Magikarp is a Pokemon. AIs would just like lose their mind. I think it was open AI models in particular. Apparently what happened there is like early on when they were like in the early stages of building these tools, they like took a small sample of the internet and started with that. And there's like a Reddit forum where people were counting to infinity in the comments, like just one after another. Like someone posted, like you'd say one, then two, then three. As Reddit does. As Reddit does. Right. What better way could there be to spend your time?

11:23Thomas Germain: Apparently, a user in this forum that like posted way more often and I think started the thread was a guy who called himself Solid Gold Magikarp. So the AI, because it saw this so often in this small data set, came to understand like this is a really important word in the English language or an important series of words. but then when they did the rest of the training and showed it the rest of the internet this basically appears zero times in all of the rest of human text so it would be confused it would like remember that this was important but have no context about it so it would break you know if you had told me magikarp was the thing that was going to break ai i don't think i ever would have believed you right it's it's a pretty weak pokemon as i remember from my elementary school days the lamest pokemon at least at first right and it evolves into something cooler yeah really taken me back.

12:11Thomas Germain: So this is interesting, but there are also some reasons that it actually really matters. And one is there's been a lot of sociological research that has shown that people are starting to talk the way that AI models write, that like the output of human language is now converging towards AI language, because we see it so often. Maybe you're interacting with Maybe you just see AI writing all the time. We're starting to talk the way that AIs talk. Yes. And I'll tell you what, we just finished midterms at my university where I teach. And I really don't like spending my time trying to hunt down AI on writing.

12:51So I actively try to ignore it most of the time. But in one of my classes, I read the first essay. I was like, huh, that's pretty good. Then I read the second essay. Okay, good. Some similar ideas. Yeah. Got to essay number three. Okay. I'm starting to think what happened here. So I do what I do often. I go to Claude or ChadGPT and put my own assignment prompt into one of these models. And sure enough, I get back pretty much what has been submitted.

13:19Thomas Germain: Surprise, surprise. Surprise, surprise. And I just, you know, I look at all of this and I just think everything is kind of regressing to the mean, just kind of moving to the center, moving to everyone talking the same, writing the same, whether it is in your college class or it is Elias in the lighthouse. that you are publishing on a self-publishing platform, that we are just moving to the center, becoming similar, and kind of losing the sense of quirkiness or difference or something that, I don't know, that gives writing some flair. And maybe that's a bummer, or maybe it's like a sign of a coming catastrophe.

13:54Thomas Germain: A while back, I talked to this neurologist named Adam Green at Georgetown University, and he did a study, look, with a team of other researchers, where they took, I think it was 300 ,000 college admissions essays, you know, like the essay you write to try and get into school. And they had a data set where they knew like these were all human written and they had a data set where like these were all written by AI or they believed that they were. And they compared the two and they found that when regular people read the AI written essays, they seemed to be better written and they rated them to be more creative.

14:30Thomas Germain: People liked the AI's writing better, apparently. Really? But when they looked at the total body of ideas within all that writing, there were fewer original thoughts. And I'm simplifying this, but there were fewer original ideas in the AI writing, even though people thought that it seemed more creative. And some researchers are very concerned that as AI becomes a bigger and bigger part of our lives and we rely on it more and more for different kinds of cognitive processes, that as a society, we might start having fewer ideas because we're not flexing those creative muscles. And I see this happening more broadly in the world.

15:15I've done a lot of research into book influencers and online book communities. And one of the common complaints is books are just being written for tropes now, particularly in some genres more than other, a lot of fantasy or romanticy, the very popular romance fantasy space right now, where it's, oh, what's the book about? Does it matter? Does it have a slow burn romance? Does it have grumpy sunshine? Does it have shadow down? Like those are the tropes people want. Shadow daddy? Shadow daddies, Tom.

15:45Thomas Germain: What's a shadow daddy? Shadow daddy is always the brooding male love interest who, of course, has shadow magic. They have shadow magic every time. Of course. Right. Of course. Whatever that is. He's got to have shadow magic. So the interesting thing here is these are solvable problems. It's not clear that the AI companies have an incentive to fix this, right? Because at least for now, they're not making a ton of money on creative writing. There was a similar problem in Claude Code, which is like the tool that engineers use to write code, where it would overuse the word load-bearing and people got really annoyed and like, that's a moneymaker for Anthropic.

16:19Thomas Germain: Yeah. So they like made it stop doing that. The creative writers of the world aren't paying as much as computer engineers are. Yeah. So there may not be a reason for them to work on this unless it like becomes a reputational thing, unless like it becomes a trope and everyone makes fun of it and it like looks bad for them. But even if they do solve this problem, the way that they would deal with it would be essentially instructing the model like, look, when you respond to this kind of prompt, don't bring up lighthouses. Don't talk about Elias in your final answer. And what that would mean is that we wouldn't be seeing this stuff in the outputs that we get.

17:01Thomas Germain: But in like the thought process, for lack of a better word, like the internal machinations of the machine, these ideas would still be in there and the model would be like, oh, I'm not supposed to talk about that. So these ghosts would still be there in the machine for years to come and we just wouldn't see it. And that is how you create the perfect internet urban legend with, if it can't talk about lighthouses, why can't the AI talk about lighthouses? Stay spooky. You gotta love it. AI is moving fast across the enterprise. But without visibility, it's just chaos. Different tools, different models, different teams using AI in completely different ways.

17:40ServiceNow turns that chaos into control. With the AI Control Tower, you see all your AI across the business in one place. What it's doing, what it's done, and what it's about to do. So you stay in control. To put AI to work for people, visit servicenow.com.

18:12online broker and get started. See what adding futures can do for you at cmegroup.com forward slash podcast. Derivatives are not suitable for all investors and involve the risk of losing more than the amount originally deposited and any profit you might have made. This is not a recommendation or offer to buy, sell, or attain any specific investment or service.

18:30Thomas Germain: So a quick summer recap. Heat waves, data leaks, AI hacks, folding phones, and a viral finally challenged raccoon. Yeah, I think that covers it, right? I'm Thomas Germain. I'm Jess Maddox and the interface is back from its summer break. New season, new co-host, same deep dives into the tech that's shaping our world and our lives. So before AI wipes out all of humanity, allegedly, join us for the interface. Listen on bbc.com or wherever you get your podcasts.

19:03All right, Tom, speaking of spooky, have you ever come across a video on TikTok that said something like, if you're seeing this, it was meant for you.

19:13Thomas Germain: I have. Yeah. There's like all these videos where it's like, oh, this is going to find you on like Tuesday, October 17th or whatever it is. Yeah. And they also sometimes say things like no hashtag, no captions, no audio, just a message that was meant to find you. And turns out this is a pretty popular way of people understanding algorithms. People are buying into it. People buying into it, buying into the idea that there is some mystical force driving algorithms to deliver you a message exactly when you are supposed to see it, right? So if you're seeing this, it is meant for you. It is a message that is going to have some profound revelation in your life and make you aware of something you didn't know before.

19:57It's gonna make you really interested in manifesting, like maybe this is gonna manifest success for you. And researchers have a name for this, actually. Researchers call this algorithmic conspirituality. What a word. And it's what happens when you think your algorithm knows you better than you know yourself. And it typically involves, again, the delivery of that message that I didn't know this about myself. I am now seeing this video that says this is a sign. Stop scrolling. I think we've all heard people say my algorithm does know me better than I know myself. Sometimes you do see pretty eerie content that it's like, you know what?

20:32I do agree with that. And I didn't know I agreed with it before. So some other examples of this.

20:38Thomas Germain: Okay. You will have an hourglass body shape. Just like and share to lock it in, meaning this will manifest if only you engage with the post. Right. Like interacting with this video will help you do magic. Exactly. Exactly. If you're seeing this, it's meant for you. Book that first session for your tattoo. No hashtags, no caption, no audio. If you're seeing this, then you're either neurodivergent, autistic, or disabled. Oh, yeah. Well, we were talking to our producer before the show. And she had an example like this where she started seeing a lot of material about ADHD. And she was like talking to a friend.

21:15Thomas Germain: She was like starting to wonder, am I seeing this content because I have ADHD? Like, is there something that it has picked up on? Yeah. And I think if we take a step back and look at kind of how the researchers arrived at this term, we can kind of see maybe why this is happening right now. Yeah. So about 15 years ago, researchers coined the term conspirituality to talk about how conspiracy theories and new age spirituality were kind of intertwining. They argued that these two seemingly different things were actually perpetuating the same narrative, which is that nothing is random. Everything happens for a reason.

21:52Everything that kind of crosses your life is meant for you. And again, nothing happens by accident. And then a couple years ago, a team of researchers at Penn State University and the University of Alabama coined the term algorithmic conspirituality, finding the evidence that said this idea, nothing is random, nothing is by accident, everything is connected and meant for you, has started to be a way people talk about and understand algorithms. Algorithms are viewed often as rational decision makers. We know that's not necessarily true. Sophia Noble has a great book, Algorithms of Oppression, talking about all the ways discrimination and power imbalances get encoded into algorithms.

22:28but people believe algorithms are these rational decision makers with this objective, almost sublime quality about them that helps people understand their lives. And I think with the examples we talked about, there's two interesting things happening here. One, content creators see themselves as vessels through which algorithmic magic can happen.

22:50Thomas Germain: That's definitely how I think about myself. Of course, that you are a vessel for magic, Tom. That is exactly what that is, yes. but it also feels a little bit like engagement bait if we're being honest as well right it's just like trying to get people to you know stop and pay attention to the video but that's also like that's a better way to understand what's happening so like this example with our producer and the adhd right it's not showing her that content because it knows that she has adhd it's showing her that content because it is picked up on the fact that when she sees stuff related to ADHD, she pays more attention.

23:26Thomas Germain: She doesn't scroll or maybe she shares it with that friend that she's talking to. Or these videos that are like, I saw one the other day where it was a particular date, October 17th or whatever. And the video was like, this video will find you on October 17th, but it was like posted in the past. And it did. Oh my God, that's incredible. Until you consider that people who saw that on October 14th and 15th and 16th would scroll past it. But then on the 17th, they engage more, and then it gets recommended more often. Exactly. And researchers have looked into this. We've been off the chronological social media feed in various stages for about 15 years or so now, right?

24:06Social media used to be this chronological feed, or what researchers called real-time social media, meaning you were seeing things as your friends posted them. It was very linear. With the algorithmic feed, we enter what researchers call right time social media, where the algorithm will, based on your interest, will serve you content when it is the right time for you to see it. And we know this is not anything magical. It is just how algorithms process your data and your interests. For instance, I just received one of these videos last week or so that was like, if you're seeing this, you hate O 'Hare Airport.

24:42And you know what? I do hate O 'Hare Airport, actually, in Chicago. But this wasn't magic. I have been traveling a lot, including a trip through O 'Hare Airport at one point. So I engage with travel content sometimes. I love watching plane content on TikTok. Well, that's another story. But it makes sense. All that to say, it makes sense for why I get it. So when we switched from this real time, seeing it linearly, to right time, I'm now traveling a lot, so I'm going to be given more travel content, you can start to see why people would kind of come up with this, well, conspirituality about how algorithms work.

25:20Thomas Germain: Well, there's this famous saying, right, that any sufficiently advanced technology looks like magic. And I think that's part of what we're dealing with here, right? Arthur C. Clarke. Oh, there you go. Yeah. The algorithms that control our feeds and so many other things in our world are phenomenally complicated. They make sense, right? Like they're doing what they do for a reason and it has to do with ones and zeros and not magic, unfortunately, as far as I know. But people are, you know, subjected to all the stuff that they don't really understand. They don't know what's going on inside the machine.

25:58Thomas Germain: And it feels better to have an answer, right? To feel like you know the explanation, just kind of to take back a feeling of control. But it's not just, this is not just a technology phenomenon, right? We see this kind of thinking in all sorts of other places. And my colleague at the University of Georgia, Shira Chess, wrote this book called The Unseen Internet, which she examines how big tech has always been a little influenced by the occult. Not saying it's magic, but one of the popular narratives about the rise of Silicon Valley and the tech industry as we know it is that it emerges out of this counterculture movement in the 1960s and 70s in Northern California.

26:40Well, at the same time, this was also the occultist revival in the United States, and my colleague charts these two movements and shows how in a lot of points they overlap, especially if you're looking at things like New Age mysticism, even Wicca a little bit to some extent, that these weren't separate movements. They were movements that were absolutely kind of intersecting with one another as we went. Same time, the emergence of the telegraph in the late 19th century was at the same time we started seeing a lot of seances and ghost wrappings. All of that to say, I don't think it's people's fault necessarily that they believe in some magic in their algorithms.

27:18Is it true? No. But I think if we take a step back and look at it historically, we can kind of see how these things have always influenced each other just a little bit.

27:25Thomas Germain: And if you think about it, there are ways that some of these eras kind of rhyme, right? In the 1960s, right? We get the Vietnam War. There's all this fear about nuclear proliferation, right? And people are worried about the apocalypse. And now it's 2026 and we're worried about the apocalypse again. Maybe it's no coincidence that when we have these moments of like huge anxiety in society, we shift into more magical thinking. And research on horror movies supports this, that generally speaking, trends in monsters and horror movies reflect social anxieties at the time. So like post 9-11, we get the Saw franchise where there's a lot of concern about excessive violence and torture.

28:11Today, I think it's no accident, some of our, you know, the biggest films of the summer were Obsession and Backrooms, which are, again, uncanny, invisible forces we don't ever see really at work, or they stay in the shadows. Feels a little bit similar to algorithmic and spirituality, the forces we can't see controlling our lives, or at least the content you get on your For You page.

28:32Thomas Germain: There was a great article a while back in the New York Times written by a writer who goes under the name Catherine D. that argued that as a society, the world is kind of drifting back into the Middle Ages, that we're becoming more medieval. Not in like that, you know, we're entering the Dark Ages or anything like that, but our thinking is kind of shifting to a frame of mind, a way of understanding the world that was more popular during the Dark Ages. And you can see this in all these different ways. There's been a recent spike in, you know, people coming back to religion after it's sort of slipping out of importance in society, right?

29:11Thomas Germain: Like Catholicism, people are converting to Catholicism at a growing rate. Or we get, you know, the rise of, you know, conspiracy theories about medicine and, you know, voting and stuff, government and all this stuff. And then there's also like a big spike in like belief in the use of tarot cards more recently. And what Catherine argues in this article is that it all stems from this belief or this feeling like there's something uncanny that's out of our view, that's controlling the world, like hidden forces that are pushing things that we can't see or really understand. And it shapes our explanation for why things happen the way they do in our daily lives.

30:01And in a way, this isn't entirely new thinking. There's always been anxieties about whatever technology is new. There's always been, you know, we used to call it tinfoil hat conspiracy theories, right? But it's all gone a little mainstream now. And I think if we're thinking about the forces or not that control our lives, we are in a moment where this kind of discourse and this conversation and this way of thinking is just extremely common and extremely prevalent. And I think it matters because I don't think we should give up our agency to the algorithm. I don't think we should just consume this content and say, oh, this is me.

30:38This is who I am. Because maybe it's not. And you are an active person.

30:43Thomas Germain: What this is really about is power, right? I think there's serious consequences to this kind of misunderstanding or this kind of belief. And I don't blame the people who feel this way or think this, but it's important to understand that if you think that the machines are magical or that they're perfect and they work exactly right and they have this kind of superhuman ability, it means we take the things that they tell us more seriously. And as AI spreads through our world, right? We were just talking a couple of weeks ago on our episode about AI consciousness and people torturing AI models and feeling pain, that there's this growing movement in Silicon Valley where people are arguing that perhaps AI might be conscious, that it might actually be alive and thinking and feeling.

31:32Thomas Germain: If this kind of thing spreads, it's going to make us more easy to manipulate. It's going to mean that we're going to be taking more things at face value. And I think there are some real ways this is actually kind of dangerous. I think so too. And when I was rereading the research to prepare for our podcast today, I just kept thinking you could substitute algorithms and AI in this conversation. And I think we could have almost the exact same conversation. Absolutely. So there's no magical forces driving your algorithm. Can be fun, right? It can be fun to be like, man, maybe my algorithm does know me.

32:03But at the end of the day, you're still in control of your algorithm.

32:07Thomas Germain: All right, Jess, you ready for trick or treat? You ready for our game? Let's go. So we went back and forth on how this works. I think where we past couple weeks and figure out, like, is it a trick? Is the narrative that we're being sold, like, not really what it's all made out to be? Or is it a treat? Like, is this really how things are going? I'm ready. Let's go. Okay. Elon Musk has announced that he is going to give every child in the United States a share of SpaceX. So that's like roughly 72 million children. And at the current stock prices, that would be about$10.5 billion of his own shares that he has that he says he's going to give away to every American child.

32:54Thomas Germain: Okay. Trick or treat. Like, what do we think about this? Trick? Is this actually going to happen? Right? Yeah. Yeah. Tech CEOs say a lot of things. Tech CEOs say a lot of things. He hasn't announced how he's going to do this exactly. Correct. And it would be pretty complicated. Yeah. I will say, like, Elon Musk has gotten in trouble for just announcing things that affect the stock market in the past. Yes. Like when he said he was going to buy Twitter, then he tried not to do it. Yes. And they had to haul him into court and force him. Hold debacle. Yeah. So I think if he announces he's going to do something with$10.5 billion of SpaceX shares, perhaps he's made some kind of legally binding commitment.

33:30Thomas Germain: So maybe it will happen. But also, I don't know, maybe we'll just forget about it. And you know, if it is a treat, then at least kids from all across America are going to build some generational wealth. Assuming that SpaceX continues to be unbelievably valuable, which is not a guarantee. All right, Tom, I've got one for you. Mm-hmm. People are using AI to hack their smart TVs and get rid of all of those pesky aspects of your homepage you maybe don't like. All those banner ads, all those silly. The bloatware, the apps you don't use. The bloatware, exactly. And by using AI to reverse engineer the system, people are being able to fully customize their smart TV experience to be exactly what they want it to be.

34:14No pesky ads. No apps they don't want to use. No clicking around your screen, putting it in the order that is perfect for you.

34:22Thomas Germain: Sounds like a dream. Trick or treat. What do you think? So I guess there's been a lot of discussion about reverse engineering, right? People are showing the AI and app and saying, just make this for me again. People were doing it with Adobe software, the company that makes Photoshop and all this other stuff. Already, the big AI companies, Anthropic and OpenAI, are tamping down on this and telling their AIs not to do it because it's upsetting some other big businesses, which is not what these companies want. Sure. I think it's a little bit of both, right? Like having the ability to break into our tools could give people back some power.

35:01Thomas Germain: But also the companies making all these devices have access to the same tools and they don't want us to do this. So like I know right now you can do it if you have like a Google TV or another, I forget which other brand exactly. You can't do it on Samsung already. It won't let you. Expect it'll get locked down. So maybe if you want to do this, I'd jump on it sooner than later. But it's also a little bit technical. It is very technical. Though it reminds me on MySpace when we were all coding our own little like backgrounds back in the day. Yeah. That was my first experience with the world of code.

35:31Me too. It was MySpace.

35:33Thomas Germain: Okay, Jess, last one for you. Hit me. Certainly you heard about PHP over the past couple of weeks. Yes. There was a huge vulnerability in this programming language called PHP2 vulnerabilities, in fact, that were discovered. Incredibly serious stuff. PHP is a very widely used and old computer programming language that, among other things, is pinned into, I think the stat is like 70 % of all websites. And this vulnerability was so severe that the advice was, if you cannot patch this immediately, you should go turn your website or whatever you have that's using PHP, turn it off because people will be able to break into it and do whatever they want.

36:17Thomas Germain: Like they can log in without a password level serious stuff. Over the course of two days, 48 hours, there were a series of, I think it was something like 10 different security problems, all of which would have been the biggest security disaster of the year in the past. Now it all happened in two days. People calling it Black October. So the question is, will AI spell the end of digital security and is our whole world going to fall apart? My brother is going to yell at me after this episode. He works in cybersecurity. Okay. But I think this is a treat. I think at least the AI finding. The narrative isn't.

36:56Thomas Germain: The narrative. Yeah. The narrative is a treat. Better it to be found before anything bad actually happened. I think that's right. There's this concern that AI will be better at attacking in cybersecurity than it will be at defense. That hackers will be able to use this tool more effectively than the people who are trying to stop them. If that was the case, our world would really be in crisis. There's not really any reason to think that AI wouldn't be just as useful for defending against cyber attacks. And I called a couple people about this, and what they told me is that essentially what's going on here is we have all this ancient technology, all these ancient tools that are holding up the internet and other parts of our digital infrastructure.

37:43Thomas Germain: And there's a lot of what's called technical debt where people have been putting off, fixing or updating these problems. And AI could force us to deal with all that. So there will probably be a period of chaos. I think this won't, you know, the whole PHP disaster, it's not going to be the last one. We should expect to see more of this stuff. But ultimately, I think we may end up with a world that's more secure, although hard to say. Fingers crossed. Well, you spooked out now, Jess? I mean, there's a lot of... I love being spooked out. I love Halloween. Maybe that's a good thing. Well, thanks everyone for joining us.

38:17Thomas Germain: That's our show. What I would do if I were you, I would go and subscribe to our podcast. You listen to the whole thing. Don't you want more of this? I know I would. And if you're looking for the show, you can find The Interface wherever you get your podcasts. If you're in the UK, The Interface is available on BBC Sounds or everywhere else in the world. You can check us out on YouTube in full color. It's The Interface podcast. Just type that into the search bar. And if you want to get in touch with us, you can send us an email at theinterface at bbc.com.

38:50AI is moving fast across the enterprise, but without visibility, it's just chaos. ServiceNow turns that chaos into control. With the AI control tower, you see all your AI in one place. To put AI to work for people, visit servicenow.com.

From the publisher

This Halloween, Thomas Germain and Jessica Maddox go ghost hunting inside the AI tools and social media we use every day.

Ask any AI assistant to tell you a story - there's a good chance it’ll be about a lighthouse keeper called Elias. Tom explains a Cornell study that found AI models from different companies, including Anthropic, OpenAI, Google and China’s biggest labs, kept returning to the same handful of words and names when asked for a story, and nobody can say for sure why. Jess tells us what happens when student essays sound suspiciously alike, and the pair ask what it means for human creativity if AI and our own writing start to blur together.

Then it’s off to TikTok for the creepy “if you’re seeing this, it was meant for you” videos. Jess explains “algorithmic conspirituality,” the idea that your 'For You' page knows you better than you know yourself, and why researchers say this thinking is more common than you might expect. Tom explains how the algorithm really works, and why a video that seems to predict the date it finds you isn’t magic. The pair also look at why anxious times seem to make us more superstitious, and why believing the machines are mystical could be dangerous path.

The Interface is your weekly guide to the tech rewiring your week and your world. Hosted by journalist Thomas Germain and “professor of the internet” Jessica Maddox, each episode unpacks, week by week, how technology is shaping all our futures. No jargon. Just two sharp voices debating the stories that matter - whether they shook a government, broke the internet, or quietly tipped the balance of power.

New episodes drop every Wednesday on BBC Sounds in the UK. Outside the UK, find us on BBC.com or wherever you get your podcasts, or watch the video version on YouTube (search “The Interface podcast”). To get in touch with the team: [email protected]

The Interface is a BBC Studios production. Producer: Becca Bryers Executive Producer: Joe Kent

More from The Interface

All episodes