Endless Thread - The bots are taking over

Episode Date: May 16, 2025

The bots are here to stay, and they're everywhere. The trouble is, learning how to spot them. On this week's episode of Endless Thread, Ben and Amory discuss two stories from Reddit about undercover bots. First — is there such a thing as an easy tell for identifying bots? Second, what happens when internet users accept bots as fellow humans, only to discover the truth later? Show Notes: ‘The Worst Internet-Research Ethics Violation I Have Ever Seen’ (The Atlantic) OpenAI used this subreddit to test AI persuasion (Tech Crunch)

Support for endless thread comes from Mathworks, creator of MATLAB and Simulink Software, to design and develop engineered systems, accelerating the pace of discovery in engineering and science. Learn more at Mathworks.com. Support for this podcast comes from Nature is the Solution, a podcast from The Nature Conservancy. This show tells climate stories like a stubborn optimist, because hope, innovation, and nature itself are key. to solving the challenges ahead. Follow on your favorite podcast app. WBUR Podcasts, Boston. All right, now, I'm going to bring you a story. Here's a story of a lampa-hanna. Wait, I'm bringing you a story.

Didn't you say that you were going to be the dessert? No, you're going to be the drizzle. I'm the amuse-boosh. Did I not say that I would be the amuse-bush? And you would bring whatever you were going to bring, and then you said I'll be the appetizer, the salad. in the main course, which all of which does come after the amuse bouch, as you should know. I don't know, but great.

Everybody's shaking their heads at me. The amuse bouch comes first. It does. The amuse bouch is. Oh, yeah, it is. It's like an appetizer. The amusement of the tongue or the amusement of the bouch. It's the amissment, which is French for, the amusement of the bouch.

This is getting worse and worse. Very silly start. All right. Okay. All right. Allow me to amuse the bush. Okay. Artificial intelligence.

Never heard of it. So do you know what the latest supposed tell of artificial intelligence online is? I only know because you told me earlier in the week. I did. I think before we knew that we were going to do this. Okay. Yeah, that's true. I was horrified when you said this.

You were, right? Yeah, I want the listener to have the same experience that I did. Do you dabble in the dash? I dabble left and right, morning to night, in the M-Dash. Okay. So the M-Dash is supposedly these days a tell of a post upon Reddit or the Internet more widely that has been composed by ChatGPT. I'm Amory Sievertson. I'm Ben Brock Johnson, and you're listening to Endless Thread.

Coming to you from WBUR, Boston's NPR. Today's episode, The Robots are Taking Over. This is something that apparently started in November of 2024. So it's been going for a while. But what I will say is in the last couple of months in 2025, I have been watching a lot of posts about the M-Dash. Again, the idea here is there's been an explosion of people, asking chat cheapy-t to compose posts, stories, things that are likely to go viral,

and then posting those online for internet points to karma farm, to basically get attention. And, you know, I think there are some other things that are happening to where, you know, in a lot of, especially the sort of like drama-based subreddits that we, that I peruse, there's a lot of sort of like relationship posts that immediately get called fake by the commenters who are basically like, oh, this was written by AI. And here are the reasons. If you click on this link that I just sent you. Okay. Oh, it looks like DNA.

So what this is is it's an X axis and a Y axis. It's a graphing of the use of M dashes in original posts on some very large. subreddits. And the use of MDASHes on posts on these subredits has, as you can see from the graph, has exploded in the past few months. So this post got about 7,000 upvotes, lots of comments. This is by Delta V ZERDA. They say, LMAO, this is pure gold. The MDASH conspiracy is actually real. As someone who's been lurking on R-slash startups for years, I've definitely noticed this trend. AI content has this weird writing style with M-Dash is everywhere. M-Dash like this, M-Dash. And it's getting more obvious every month. The graph perfectly captures what's happening.

Look at that hockey stick growth from August onward. Oh my God, I know where this is going. You can literally spot the AI post now. Unnecessarily formal tone. M-Dash is everywhere for no reason. That weird quote, I'm pretending to be. to be helpful while subtly promoting something vibe. The next time you see a post like, as an entrepreneur, M-Dash, with 10 years of experience, M-Dash, I found that the key to success, M-Dash, in this competitive landscape, M-Dash,

is to focus on customer acquisition. You know what you're dealing with, LOL. Someone should make a browser extension that flags potential AI content based on M-Dash frequency would save us all a lot of time. That's actually a really good idea, but, you know, anyway. What do you think? M-Dash.

Is it real? Or is it not real, the M-Dash conspiracy? How do you feel in this moment? The vibe I got from this comment is this is AI. Delta v. Zerta is AI. The call is coming from inside the house. All right. Yes, and it was not the M-Dash that gave it away,

because obviously they're explaining the writing style with M-Dash is everywhere. It was this graph perfectly captures what's happening. Look at that hockey stick growth from August onward. This just read like bought. speak. And then I cheated and scrolled a little further down because someone responded to them. And then Delta v. Zerda said, by the way, you're responding to AI. As a rule, I don't use AI for Reddit posts, but I thought it would be ironic for this one. So spotted. Detected. Detected.

What does this all make you think about? Well, I'm thinking about anything that I've written recently that has had M-Dashes And now I'm very concerned that the recipient thought I was a bot. But I'm not a bot. I'm a real girl. It's hard for me to, I guess, put my finger on what made this feel like AI to me. What stands out to me is not necessarily what would stand out to somebody else. And collectively, what stands out to all of us is what will save the day. So I guess I'm glad to know about it.

And I'll definitely think twice before I am, Dash. So I want to direct your thoughts towards a writer named Adam Cecil. He has a podcast, it looks like, and also a site called Nightwater. And Adam has a post called You Can Take My M-Dashes from My Cold Dead Hands. And which, Adam, I couldn't agree more. Adam points out that the problem with this concept is that plenty of writers are artificial and human alike, use M-dashes, including yours truly. So I don't think we're clear yet on whether or not the M-Dash is a real sign of AI.

But what is clear to me and just, again, sort of connects to the general problem that we're all feeling these days, is like, what's real and what's fake? And do we really have to expend mental energy figuring that out? I'm tired, man. I'm so tired. I'm just tired. Do you know what I mean? Absolutely. But you know what's going to pick me up? A little break.

A little break. We'll take a little break. And I'm ready for a salad. Or the appetizer. After this M-Dash and Moose Bouch. Okay. Appetizer starter salad coming right up after a break. At Radio Lab, we love nothing more than nerding out about science, neuroscience, chemistry. But, but we do

also like to get into other kinds of stories. Stories about policing. Or politics. Country music. Hockey. Sex. Of bucks. Regardless of whether we're looking at science or not science, we bring a rigorous curiosity to get you the answers. And hopefully,

make you see the world anew. Radio Lab, adventures on the edge of what we think we know. Wherever you get your podcast. Support for this podcast comes from Is Business Broken, a podcast from the Mayrotra Institute at Boston University. On the show, host Kurt Nickish asks the thorny questions necessary for this moment about the role business plays in society. Questions like, why are executives paid so much? Why is innovation and healthcare so hard? Is ESG just

greenwashing? And of course, is business broken? Follow is business broken wherever you get your podcasts. There is something powerful about the sound of the human voice. Beautifully produced audio has the unique power to connect and inspire. Tell your organization's story with a custom podcast from City Space Productions, the creative studio from WBUR's business partnerships team. Become a thought leader. Recruit new talent. Reach new audiences. Whatever your goal, we can help. Discover how the magic is made at WBUR.org.org slash creative. studio. Okay, we're back. Ben. Amory.

Have you heard of the Redditor Genevieve Strom? I was trying to come up with like a nickname for Genevieve Strone just to be like, yeah, of course I know my girl Strone. It's Strome, so you clearly don't. Exactly. I was like, I don't know how to do this. No, I have no idea. Who's Genevieve Strome? Strone. Okay.

Strom. Who's Genevieve Minestrone? I'll get there. I'll give you one more chance. Have you heard of the Reddeter Spunger Maniac? No, I don't think so. Okay. I'm like poem for your Sprague, Ward Lizard.

That's my zone. Okay, well, these two Reditors were involved in a scandal, really. Scandal almost doesn't seem like a big enough word for this, but this happened in the Change My View subreddit. Okay. There's a big post, a meta post made by the mods in late April. Okay. And here's an excerpt of that post.

The CMV mod team needs to inform the CMV community about an unauthorized experiment conducted by researchers from the University of Zurich on CMV users. Oh, yeah. Yep, I do know about this. I do. All right. Yes. Please continue.

They go on. This experiment deployed AI-generated comments to study how AI could be used to change views. So you have Change My View posts, things like popcorn is disgusting. Change My View. That's a very stupid example, but you know what I mean? And then these researchers created AI users like Genevieve-Strom and Spunger Maniac, among many others, to comment on these posts with AI-generated hot takes of their own or attempts to change the OPs view.

The mods write, CMV rules do not allow the use of undisclosed AI-generated content or bots on our sub. The researchers did not contact us ahead of the study, and if they had, we would have declined. We have requested an apology from the researchers and asked that this research not be published, among other complaints. So, this is an excerpt of what the University of Zurich researchers wrote to the CMV mods. Okay. They say, in commenting, with these fake accounts, we did not disclose that an AI was used to write comments, as this would have rendered the study unfeasible. While we did not write any comments ourselves, we manually reviewed each comment posted to ensure they were not harmful. So bots are writing them.

They are reviewing them, but then they get posted on the subreddit. Yep. People are up in arms about this. Yes. There's a big Atlantic piece by Tom Bartlett written on it. He quotes Amy Bruckman, a professor at the Georgia Institute of Technology, who refers to this as the worst internet research ethics violation. She has ever seen, quote, no contest. Yeah.

Reddit's legal team has gotten involved. Ben Lee, who's their chief legal officer, says that the company intends to, quote, ensure that the researchers are held accountable for their misdeeds. Oh, boy. Yeah. NPR just covered this as well. It's gone stratospheric. Yeah.

It's gone stratospheric. It's only in the last couple of days as of when we're recording this that the researchers apologized, though. Well, I was going to say, was it a spicy researcher response? Because I love a spicy researcher response. They're not in a spicy place. They're in an uh-oh, we screwed up place because they're facing legal action. AI companies have licensing deals with Reddit.

OpenAI has a licensing deal with Reddit to be able to train its program on Reddit posts. And so the idea that researchers would be doing that without a deal and without consent has people pissed off. And Open AI, I should say, did do something similar. They used Change My View posts and had. had AI models write replies, but in a closed environment, apparently. So they weren't actually putting the replies on posts. They were coming up with replies, and then they were showing those replies to, like, designated testers who were looking at those replies and then telling them how persuasive they found

those arguments or not. And I think part of the reason why people are so upset about this is the nature of the posts that they were having AI comment on. just like really controversial political, hot takes, things about sexual assault, areas where what are already super muddy waters, you do not necessarily want fake humans weighing in on. So the researchers have written an apology and realized just how big of a deal this is. And so the first part of the apology is all just, you know, the disappointment, the frustration, profound sense of personal sorrow. And then they say, we want you to know that we've taken this wake-up call seriously.

In that spirit, we have already implemented the following measures. We have permanently ended the use of the dataset generated from this experiment. And they have, in case it's not clear from that, they did say that they will not publish this research. Wait, so they've ended it? What does that mean? They have ended the use. use of the data set. That's true. The use of the data set. So they've said they're not going to publish it. Yeah. But they haven't said that they've like deleted it. Yeah. The end of that, that seems like a weird phrasing. But yes, please go on. It says we will never publish any part of this research. We commit to stronger ethical safeguards in future research. Going forward, we will only consider research designs where all participants are fully informed and have given consent. In order to rebuild trust with change my view and to further demonstrate our sincere regret, we declare. We declare. We declare. our willingness to collaborate at no cost, oh, thank you, with the subreddit to develop systems

that can promptly detect and block unauthorized interference and can support the development of a clear framework for handling violations. Oh, I'm sure that'll go somewhere. Yeah. So now they... Oh, please, would you like to work with us? Would you like to work with us? We promise we'll do it right next time.

They also said, we respectfully request that our anonymity be preserved to protect the safety and privacy of our families. Oh, boy, oh, boy. Yeah, well, the lead researcher was named, unquoted in the Atlantic article and said that these researchers were receiving death threats. Well, there's your problem right there.

To say it has rubbed people the wrong way is truly an understatement. And the mod response to the researcher's apology was, while we appreciate the offer, we have already made arrangements with other groups, and Reddit admins have proactively made changes to the platform. So, it's not a lot of the question.

So it's not totally clear what those changes are and what the changes within the subreddit will be and what legal action, if any, will be taken. But this was a clear, like a line has been crossed and a very large Reddit community has maybe been further alerted to the fact that the bots are there. They're swimming in their waters. It's harder to tell than ever. And they're going to weigh in on things that seem really icky. for bots to be trying to change or shape our opinions on. Those bots be astroturfing. They do.

They do. And the karma account for the LLM research team user account, which was made so that they could respond to some of the response. Uh-oh. Negative 100. I kind of expected it to be even further into the negatives. I don't think I've ever seen a karma. I've never seen karma in the negatives.

I know it can happen, but wow. So I guess a question is. I have about this is, no, it's more of a comment than a question. You're allowed to comment. The comment I have about this is so like if folks will remember, which maybe they won't, but we made a series about chatbots. And one of those episodes, we talked to a researcher at Dartmouth who is scraping Reddit content to feed into an LLM that was then going to decide if certain users might be

displaying signs through their Reddit commenting of mental illness. Right? Yep. And so I think what's interesting here is this sort of like ethics of using data like from a website like Reddit that is public facing and figuring out how to do that in the right way. You know, I think there's like we're in this kind of like wild west time of like everybody's experimenting with AI. But like, if you're a Reddit user, you know what you don't want to do. You don't want to be talking to bots. Like, you just don't. Like, you want to be talking to other

people. If you're an internet user, that's what you want. Like, you don't want to be, like, commenting on fake posts. You don't want fake bots to be commenting on your posts. You just, like, don't want that if you believe in this sort of basic idea of the internet is a place where you can connect with other human beings, right? And so it's like, if they had stopped and thought about this for a second, I think they could have used their power of logic to understand that, like, people don't want that and that the community would react negatively to that. If you're trying to manipulate a community of real humans by inserting a bunch of robot

comments, like, that's just a bad look. Like, go ask a six-year-old if that's a good idea. Do you know what I mean? Not to give these people too hard a time, but like, I'm just like, dude, like pause for one second and decide whether or not you want to enter into a public space where a bunch of people are trying to talk to each other and have a bunch of like robots in there without disclosing that they're robots. That's just not a good idea on its face. So I just, think it's kind of weird that the researchers decided to do this. Well, I have some thoughts about that because OpenAI having an agreement with Reddit to scrape its user information posts and things like that to train its AI. It makes sense that they would

pay for that access, and yet that's not necessarily what Redditors signed up for. researchers doing research might think that they have kind of a pass on that because they're saying, this is for research. This is not to train our, you know, tool at a private company. This is for research. Yeah, but like, you know. No, I'm not defending it. I'm not defending it.

Another comment that that professor that I quoted Amy Bruckman at the Georgia Institute of Technology, another comment that is included from her in the Atlantic piece, is this concern that the uproar around this, the kerfuffle around this, is going to undermine the research that really should be done into how AI influences the way that humans interact. And that's not to say that this is the way to do it. It's not. You can't experiment on non-consenting people. And you are absolutely right. Redditors do not want to talk to bots. They're there to talk to other humans. And so this is the wrong way to do it.

But this is research that done correctly in a consenting, non-icky way, I do think is really important to do. And it is sad that this was executed in this way because I do worry that people are going to balk at the idea of research like this being done because it's been done so poorly. But I guess like, okay, yes, okay. I'm just going to play the role of the cranky person here for a second. Go for it. And I'm cranky too, but go for it. Like, I would love for someone to come and pay me a lot of money to tell you whether or not it's a good idea to have bots shift people's opinions. I think that we don't need research on that.

The answer is no. It's not a good idea. Bots are going to shift people's opinions no matter what, though. Like, that's going to happen. And so if there's something to be learned about how to avoid people from falling down particular rabbit holes, I think that can be valuable. That's not to say that this research was ever going to accomplish that. It wasn't. But that's what I mean.

I think there's like a fallacy in there for me where it's sort of like, yeah, we got to research this to see if it works. And it's like, I don't care. I don't, it works. We know it works. it's already been done, and also it's bad. So let's throw it out. We don't need the bots changing our minds. Case closed.

I appreciate that perspective. The cranky Ben perspective. Well, I was, you know, we're working on a few episodes right now that have to do with AI. Yes. And there was someone quoted in a piece that I was reading as part of my research for that other episode where I think the quote was something like, using. AI is not going to be optional in the future. It's just going to be a part of life, the way that using the internet right now, we would say, is not optional. I don't know. I feel like more and more people these days are like, you know what? It was a mistake. Let's put it in the trash. So I'm like,

yes, I agree with you. And also, we could decide to not do it. Yes. I mean, I take your point. I was reading a tech crunch article about the open AI version of this experiment with Change My View, where they were having AI write the posts, but they weren't being surfaced for actual Redditors in the threads to potentially respond to. And I think they said that their goal was not to create hyper-persuasive AI models, but instead to ensure AI models don't get too persuasive. And that to me, my like bullshit meter goes off a little bit there where I'm like I do think there's we do need to research this stuff but to say we're doing it so that models don't get too persuasive when you're open AI and you are absolutely trying to create technology that is as human like and convincing as possible I'm not sure that I buy that coming from open AI yeah yeah you know I might buy it coming from a team of genuine researchers who have consenting participants genuine independent researchers.

Genuine independent researchers. Then again, what are we cutting, slashing, funding left and right to? Genuine independent research and university research right now. You're right. And I was going to try to say that too. Like I'm pro-research, I'm pro-science. Full stop. I'm in favor.

I just, every once in a while, I just want to say your scientists were so concerned with whether or not they should. They didn't think about, no. whether or not they could. They didn't think about whether or not they should. Anyway, I thought this might be thought provoking. Were your thoughts provoked? They were provoked.

Oh, God. Are you a robot? Oh, no. I don't know. M-Dash, am I? Oh, no. This episode was co-hosted by me, Ben Brock Johnson, real 100% homegrown,

grass-fed human. And? Me, every Severson. A real living girl. It was produced by Franny Monaghan. Our sound designer is Paul Vicus. Our managing producer is Sumit of Joshi.

Our show is edited by Meg Kramer. The rest of our team is Dean Russell, Grace Tatter, and Emily Jankowski. If you've got an untold history and unsolved mystery or another wild story from the internet that you want us to hit us up. Endless thread at WBUR.org.