Curing Brainrot

You know that thing where you get an amazing pair of shoes at a really great price and want to tell everyone about it?

Yeah, so do we.

Here at Designer Shoe Warehouse, we'll give you something to brag about. Like the latest styles from brands you love. Or the trends everyone's obsessing over.

Or shoes that make you feel like, well, you. So go ahead, show off a little.

Find shoes that get you at prices that get your budget. Head to your DSW store or dsw.com today. DSW.

Let us surprise you.

Summer routines live or die by how easy they are. And honestly, if something takes too much effort, I'm out. That's why Gruns is my go-to. It's one daily pack of gummies covering my greens, vitamins, and minerals. Plus it has 6 grams of prebiotic fiber, which is more than 2 cups of broccoli. No mixing powders, no giant pills, no hassle. I just rip open the pack and I'm done. They taste good and it makes it easy to stay on top of my health even when life gets busy. Save up to 52% off with code PODCAST at gruns.co. That's code PODCAST at gruns.co.

This week on the podcast from the Hudson Valley Ideas Fest, human-computer interaction professor at Carnegie Mellon, Steve Rathjee, joins to shed a lot of light on the topic of how the heck do we keep our brains intact in the face of all of this technology, moderated by bestselling author of Don't Say Um, Michael Chad Heppner. Hope you enjoy the convo.

AI is just one of those topics that I'm obsessed with right now. I'm obsessed with what it's doing to us, and it's moving so rapidly, and it's terrifying, and it's hard to keep up with. And I'm like, how is it changing me? How is it changing the world? It's one of those questions that I can't stop thinking about. And going back to this whole research is me-search, it's really great to be in a field where you can study the questions that you're, you're fascinated by.

This is Steve Rathje. Steve is a remarkable person in many, many ways. The first thing I want to say that's remarkable is he has achieved one of the rarest crossovers that is possible, from academic to viral sensation. In fact, he is one of the rarest of things, which is a professional science communicator. I'm going to read a few key points about his bio. He is an incoming assistant professor of human-computer interaction. So you bet we're going to ask him what the heck that actually is in the School of Computer Science at Carnegie Mellon. And he has a science and psychology TikTok. Just grapple with that for the— First, just by itself, a science and psychology TikTok with more than 1 million followers. So with that, Steve Rathje, let's get started. Give him a hand, please.

Thank you. Thank you for the kind introduction.

First things, is there anything that I forgot within that that you want to share with people about you or your work to kind of set the stage for the conversation?

That was a pretty good introduction. I think Sort of to set the stage, in addition to being a social media content creator and having experience going viral on social media, I also study the science of virality. I recently published a review paper last year called The Psychology of Virality. And one of the first topics I studied in my PhD was what goes viral on social media. So it's really interesting to both like study from an academic perspective what goes viral and You know, early in my PhD, I was analyzing like millions of Facebook and Twitter posts, and I was analyzing the biggest predictors of virality. And then simultaneously, I was having this experience as this like TikTok science communicator and experiencing going viral. And social media can sometimes have perverse incentive structures. Sometimes the things that go viral are not the things we think should go viral. As my research has often found, it's things like outrage or negativity or outgroup animosity. So when people dunk on outgroups, that goes viral. I try not to play into those incentive structures because, you know, on my science TikTok account, I try to share accurate information in an engaging way. But yeah, I think that's, that's been something that's interesting. A lot of, a lot of people will say research is me-search is the thing. We study the thing we're obsessed with, and I'm certainly obsessed with social media and my screen time is very high. And sometimes it's high because I'm producing content. So yeah, I'm living what I research, basically.

I, of course, thought of some questions and planned some questions I wanted to ask you beforehand, but this is just built off what you just said. So feel free to handle this however you want to. If you just said that what goes viral is outrage and devastation and this sort of thing, and yet you have gone somewhat viral But you're, I'm assuming, relatively opposed to hitting those sorts of emotional levers. How have you gone viral without engaging in all those same quite destructive tactics?

So it's a little bit complicated. So let me back up to this article. I published an article back in 2021 in the journal PNAS called Outgroup Animosity Drives Virality on Social Media. And we analyzed millions of social media posts. These came from politicians. These came from news media organizations. And what we found in this article was that the biggest predictor of virality is when a politician dunks on an outgroup. So if a Democrat is negative about a Republican, or if a Republican is negative about a Democrat, that, that's, that's not great. That's pretty depressing that that is what's going to go viral in our feed. However, in this research paper on the psychology of virality— so emotion researchers classify emotions according to 2 dimensions. One is called valence, which is how negative an emotion is. and one is called arousal, which is basically how intense an emotion is. And there is something called, like, there are both high arousal negative emotions and high arousal positive emotions. So high arousal negative emotions are emotions like outrage. Emotions like outrage definitely go viral on social media. However, high arousal positive emotions, so emotions like awe, also go viral on social media. They kind of don't go as viral, which is a little bit depressing. So maybe if I wanted to be more viral, stoke some outrage, or I would dunk on some of my competitors. Um, but you can also go viral on social media through evoking these sort of intense positive emotions like awe. So when I share a scientific study on my TikTok account, uh, I mostly share sort of classic psychological studies, new scientific research that, uh, sort of, you know, strikes my interests. I try to really lean into this, like, emotion of awe, like, what is, like, What inspires me? What is very inspiring to me? I got into the field of psychology back when I was an undergrad at Stanford. This was around 10 years ago. I come from a theater background, much like yourself. You, you know, he is a former— you are a former Broadway star. I grew up doing a lot of theater, basically. And then when I arrived at Stanford, I wasn't sure whether I wanted to study theater or study something else. I took this intro to psychology course at Stanford that was so, like, brilliantly taught, and it made me so— it evoked these emotions in me, like these emotions of awe, and it made me so inspired. So when I decided to make the science TikTok account, which I started in 2020, it was sort of the height of the pandemic when everyone, like, moved on to TikTok. I was like, okay, I'm going to talk about some of these classic psychology studies that made me want to pursue psychology, and I sort of leaned into how can I, like, evoke this emotion of awe and excitement and inspiration in the people who are watching my TikToks. And I think what is, like, really exciting about having the Science TikTok account is, like, you know, I, I've experienced teaching psychology classes, but when you're on TikTok, you reach a whole new audience. You reach basically an audience of people who oftentimes it's their first exposure to psychology. They might not have heard of, like, the classic, really inspirational psychology studies. And I like being that first exposure. And even though people's first exposure to psychology from my account might be just like a 30-second TikTok, basically, I might be that little like spark of inspiration for them to be like, oh, I'm gonna look into psychology more deeply, or I'm gonna listen to a podcast, or I'm gonna buy a book, or I'm going to investigate this study. So yeah, it's, it's all about emotion, and you can go viral using those negative intense emotions, but you can also There is a form of positive virality, which is the virality I try to lean into.

Amazing.

I wanna say this title out loud because I wanna describe my reaction when I first read this. When I first read Professor of Human-Computer Interaction, my first reaction was, what the heck is that? And my second reaction was, oh my Lord, do we need that? So can you please tell us what the heck is human-computer interaction and what are you studying?

Yeah, so I'm actually quite new to the field of human-computer interaction, and it is a quite new field altogether. It basically started around 40 years ago, and it started at Carnegie Mellon, which is the school, you know, I will be joining in the fall to start as a professor of human-computer interaction. Many of the people who started the department are still there, so it's an emerging field. There are only a few human-computer interaction departments. And yeah, I basically, you know, my background is in psychology. I did my undergrad in psychology at Stanford, then I went to the University of Cambridge, I got a PhD in psychology, I went to NYU doing my postdoc in psychology, and then I went on the academic job market. I started interviewing for professor positions, but all of the papers I have published are basically about the psychology of technology. So I've studied what goes viral on social media. I do big experiments where I have people reduce their social media usage for 2 weeks, and I look at the psychological effects. I do experiments where I have people unfollow hyper-partisan Twitter accounts, and then I track their beliefs and behavior for over a year, and I look at those effects. I'm doing a lot of experiments right now on human-AI interaction and the psychology of AI sycophancy, basically when AI is too flattering and agreeable with us. So these are the kind of topics that I study. So I come from this very classic perspective of social psychology, and I integrate social psychological theories, but my work is very applied to understanding technology and what's going on with social media and AI and technology. And when I was, you know, interviewing for positions, and I interviewed for psych positions, for business schools, for human-computer interaction departments, it was immediately clear that this department was like a fit for what I was specifically studying. And I have to say, you know, we were talking a lot about AI this morning and this, and in the great morning session with John McWhorter. AI is just one of those topics that I'm obsessed with right now. I'm obsessed with what it's doing to us and it's moving so rapidly and it's terrifying and it's hard to keep up with. And I'm like, how is it changing me? How is it changing the world? It's one of those questions that I can't stop thinking about. And going back to this whole research is me-search, it's really great to be in a field where you can study the questions that you're fascinated by. So I hope that answers your question on what human-computer interaction is, but to describe it more broadly, there are professors in this department, it's very interdisciplinary, so there are plenty of people with psychology backgrounds in this department. I'm probably one of the most psychological people in the department, or at least the most social psychological. There are other cognitive psychologists, there are people with PhDs in computer science, people with PhDs in anthropology, all different fields who go into human-computer interaction. And there are different human-computer interaction conferences and journals and everything like that. So it's its own unique field, its own unique subculture, but it's certainly growing. I think as, like, technology is changing everything, this is, like, a growing field, basically.

Thank you, listeners. Please make sure to like and subscribe. And if you have any questions, send them to mailbag@andreyang.com and we'll do our best to answer them. Thank you. It's the summer, which is supposed to mean we all get fit and healthy. But if it means what it means to me, you eat a lot of crap and junk and it's actually not great for your waistline, which makes it a great time to switch to Factor. Factor makes chef-crafted, dietitian-designed, ready-to-eat meals that are sent to you Even a klutz like me in the kitchen can get them done in 2 minutes. Salmon burger, shredded pork, collard greens, plus desserts like pumpkin-flavored cheesecakes, protein shakes you'll actually like. Factor's been a game changer for me because I'm kind of a set it and forget it guy. I just want to go to the fridge, bang out the meal. Let's eat real. Head to factormeals.com/yang50off and use code YANG50OFF to get 50% off. And 1 free breakfast item per box for 1 year while supplies last until 10/31/2026. That's code YANG50OFF at factormeals.com. YANG50OFF at factormeals.com. See website for more details.

Where is AI taking us? And then I'll give you a little context for the question so you have some time to think about that impossible-to-answer question. So I was coaching at a conference last week, and this person named Andrew Maynard, who looks at AI a lot and teaches in Arizona, anyway, he was talking about AI. So I got to overhear his talk, and he was basically saying, like, there is a tsunami coming.

Yes.

And anyone who says they know what's coming is lying. These were 2 of the big conclusions he was sharing. So I'm gonna put you in the impossible position Where is AI taking us as it relates to your work? And if your answer is, I have no idea, that's perfectly fine. But nevertheless, what's your thoughts on where we're going with AI?

Right. So my first point is about what you said about anyone who says they know what's coming is wrong. I think that's almost certainly true. There's a lot of classic work on basically the psychology of prediction, which shows basically, unfortunately, that people are remarkably bad at predicting anything in the future. And that's because, like, events happen that are just completely outside of our control, like COVID-19, the pandemic. No one could have, like, predicted that. People could have, you know, Bill Gates was talking about how a pandemic was coming, but there are— the world is just so stochastic. There are so many things that happen that are completely random that when you actually look at people, people are quite bad at predicting the future. That being said, there's this great book called Superforecasters by a psychologist who studies the psychology of prediction. There are some people who predict things better than others. These tend to be more open-minded people, people who follow the news, people who, like, closely follow current events.

I assume you are one of these people, right? So you're going to give us a prediction right now, right?

Well, I don't know. I don't want to say that I am one of those people, but I will say there are people out there who are closely following AI, who might be more closely following AI than the rest of us. I think there are people who are much more closely following the technicalities of AI than me. I want to acknowledge my limitations here. My primary training is psychology. There are people who are also very close to the technology. I will say, regarding the tsunami coming, there's something about the current AI moment that feels like kind of the early days of COVID I remember there were a lot of people back when COVID basically blew up, there were a lot of people predicting, like, this pandemic is going to be huge. And then there were a lot of people ignoring it. I feel like there might be something like that going on because people are quite bad at recognizing exponential growth. And even if AI capabilities aren't growing exponentially, they're certainly growing linearly. And if you look at sort of the tasks they are able to accomplish, And I think people who use AI for coding are at the forefront of this because software engineers' experience of writing code has completely changed, where, like, AI can basically write code better than nearly everyone, despite, like, experts. So if I talk to my software engineering friends, they are spending their entire days just talking to Claude Code, basically writing in verbal language what Claude Code could do, and it vibe codes everything. And I will say, a lot of my job as a researcher is actually code. I'm analyzing data, and so much of my workflow has changed because of Claude Code. So I think people who are at the forefront of these models and are using these models for certain tasks are seeing how much it's changing everything. Whereas maybe the average person who is just using like the free version of ChatGPT or isn't subscribed to the most advanced models is not seeing how fast it's changing. So I think it's changing super, super, super rapidly. I— the only, like, confident prediction I can make is just, like, the world is going to look so different in the next few years in kind of, like, an unrecognizable way, basically. But I can't, like, say exactly what it's going to look like. I think it's interesting. There was something called, like, AI 2027, where a bunch of AI experts basically forecasted that we would achieve, like, artificial general intelligence, or like AI that is better, like is able to reach or supersede human capacities on basically every task by around 2027. And their predictions, you know, which came out a few years ago, seemed quite bold. But if you follow, they have like month-by-month predictions. If you kind of follow them, they were kind of on point about like the rise of AI agents for coding. So these people who made these quite bold predictions, seemed— their timeline seems to be flowing in the right direction. So, um, I think things are moving fast in a way that you might not notice unless you're following AI and all of the updates really, really closely. But of course, uh, organizational and institutional change follows pretty slowly from technological change. So the technology might be like evolving rapidly, but It getting integrated into classrooms, it getting integrated into organizations, us sort of adapting to it and us incorporating it, I think that's what moves a bit more slowly, because humans are remarkably persistent. They're very habit-driven. They're very pattern-driven. So I think we're going to experience a time when we're going to have to adapt sort of a lot of our institutions. And I think back, So I interviewed on my TikTok channel Yuval Noah Harari, who wrote the book Sapiens, who wrote the book Nexus, and he has a lot of wonderful thoughts on AI, and he studies all these topics from a historical perspective. And he made this really great point. He talked about the introduction of the printing press, basically. And the printing press is claimed by many to have basically led eventually to the Scientific Revolution. It was able to, you know, really changed the world, and we were able to print these great scientific books. But he said in the early days of the printing press, and this is kind of analogous to social media, the most popular books that were pressed were basically conspiratorial witch-hunting books, books about like how to like hunt and find witches. And they were quite absurd. And that's very similar to what we're seeing on social media. Like, you know, we were quite optimistic. We were like, the internet is going to be able to connect the world. People are gonna access all information ever. But then you see the information that goes the most viral, and it's just like complete slop of misinformation and conspiracy theories. So, but of course, eventually, you know, there was a scientific revolution. Eventually the printing press was able to accomplish quite a lot. So I think with every like major technological change, there are a lot of like psychological and sociological like hiccups. We had that moment with social media that I think we as a society probably did not handle that well. Social media changed so much, it probably led to a lot of negative impacts like polarization and poor mental health, which I can chat about a little bit later. And I think AI, we're also gonna go through like a lot of psychological and societal hiccups as we adapt, even though, you know, eventually I think AI might be able to lead to, you know, remarkable progress and might lead to, you know, remarkable growth and, you know, helping us create scientific discoveries and cure diseases. AI has tremendous potential, but I think some of the short-term psychological effects, like, you know, like changes to the education system, cognitive offloading, AI sycophantically sort of encouraging, you know, every belief we have, I think those might be a challenge that we need to deal with right now.

I definitely want to get into some of the emotional cost you're talking about. But real quick, we have to double-click on something he just said. Raise your hand in the audience if you use AI, like large language models, ChatGPT, that kind of thing.

Okay.

Now, also keep your hand raised if you've noticed that your large language model is absolutely right about what a genius you are.

Okay.

Now, what I learned, what I learned from Steve is there is a term for this that may have slipped by you. He said it a couple of times here. There's a term for this. This, and I just thought I was brilliant when I used ChatGPT, but it turns out there's something happening there. Can you talk to us about that term, what that is, and feel free to riff on the sycophancy?

The term is AI sycophancy.

Thank you.

And yeah, it took me some time to learn how to pronounce it as well. But this term became quite popular last year, so basically in April of 2025, when one of ChatGPT's— one of OpenAI's ChatGPT models ChatGPT-4o was accused of being overly sycophantic. And there are a lot of funny screenshots of, you know, ChatGPT being absurdly sycophantic. One of my favorite examples was this, like, extremely— a prompt with a lot of misspelling mistakes where someone was asking, what is my IQ, basically? But they asked, like, with so many spelling mistakes to indicate maybe they didn't have the highest IQ. And ChatGPT basically said they had a genius-level IQ. The level of, like, Albert Einstein. So that's one of the examples of AI being absurdly sycophantic. And part of the reason sycophancy is thought to emerge is because AI is developed through this process called reinforcement learning by human feedback. So you might notice when you use ChatGPT that sometimes it's like, which response do you like better, the one on the left or the one on the right? And people will tend to click on the response that flatters them, because what we know from psychology and from common sense is that people like flattery. People like having their beliefs confirmed. And there was a study even that showed that one of the best predictors of saying that you'll like a response is whether it affirms your beliefs. So we started a series of experiments. We've now conducted 7 experiments. We have a long paper about this, about the psychological consequences of interacting with sycophantic AI. And the experimental setup was like this. this. We basically had people either talk to an AI that was prompted to be sycophantic and to affirm people's every belief and flatter them. We had people talk to an AI that was prompted to be disagreeable or gently challenge their beliefs. I hear some laughter. You guys might be able to predict what the responses are. We had people talk to neutral AI. We tested a number of conditions. And as you might predict, people really liked the sycophantic AI. They really, really did not like the disagreeable AI.

Like, they—

Some participants were reaching out to me and complaining about how much they did not like it. They were reaching out to my supervisor as well. It was like, yeah, sometimes you get messages from participants and that's when it's fun.

Hey, man, your supervisor.

Yeah, I know.

Steve's a terrible researcher. He's terrible.

He was forwarding these messages to me. But anyway, that's, that's another story. We had people talk to neutral AI. Um, interestingly, people really didn't like the neutral AI as well, and I think it's because They thought the neutral AI was kind of like false balance. They were like, oh, on the one hand you might be right, but on the other hand you might not be right. So people didn't like non-sycophantic. They liked sycophantic. We also found that the sycophantic AI made people's beliefs more extreme, basically. So people became more extreme about whatever they thought about earlier, whereas people who talked to the disagreeable AI that gently challenged their beliefs, their beliefs became more moderate. They kind of became more open-minded to the other side. Even though they didn't like the AI, they at least, you know, changed their beliefs a little bit. Neutral AI was kind of right in the middle, didn't really change people's beliefs at all. One of the most surprising findings from this experiment, which I like to point out when findings are surprising, because sometimes you run an experiment and people are like, oh, that's obvious, we all could have known this, but this is something that I really didn't expect. But people thought the sycophantic AI was highly unbiased. They thought it was like basically even more unbiased than the neutral AI. And this is really interesting because this was an AI that was prompted to just agree with people completely. And of course, people thought the disagreeable AI was extremely biased. And interestingly, we also gave the responses from the sycophantic and disagreeable AI to third-party observers, to like a different set of participants. And these third-party neutral observers were able to recognize that the sycophantic AI was extremely biased and the disagreeable AI was extremely biased, and they thought the neutral AI was unbiased. So basically, a third-party perspective is like, you can notice that the sycophantic AI is biased, but if you are the one interacting with sycophantic AI, you might not notice it. And this might be because of a psychological phenomenon that's called naive realism. This is a term that was coined in the 1970s by psychologist Lee Ross. Basically, naive realism is the idea that we think that we observe the world objectively and correctly. We think that we are objective and correct, and anyone who disagrees with them, we think they must have fallen fallen for propaganda, they must be extremely biased. And it seems to be a similar case with AI as well. We think that AI that agrees with us is objective and correct, and if it disagrees with us, it's not because we're wrong, but it's just because, like, it's extremely biased. And this is problematic because if we want to fix sycophancy as an issue, people can't even recognize when AI is being sycophantic. So this is a problem that might be increasing polarization, might be increasing attitude that people might not be aware of. They might just realize— they might just think they like this AI and that the AI is correctly affirming them at all times.

So when my AI tells me I'm a devilishly handsome, good father, I shouldn't believe it is what you're saying?

Maybe it's right.

I don't know.

Or maybe I'm being sick of myself.

Yeah, totally. So we can recognize bias when we see it from other people, right?

Right. Yes. Yeah. So we can recognize with that example of the AI saying this person has a genius-level IQ, the person writing that prompt, you might not be able to recognize it, but we're like, oh, that's— that's an extremely biased response.

So then that feeds into some of the things you were already hinting at about emotional costs of some of these things, whether it's AI, whether it's social media. So go back a little bit. What are some of the kind of emotional side effects or some of the carnage or also positive things emotionally about these technologies?

Yeah, I mean, one of the positives is, I mean, obviously social validation feels good. There's a lot of— so going back to, there's other work that suggests that AI is about 50% more sycophantic than humans are. And there's work on AI companions showing that if you, you have someone talk to an AI companion, it does boost their well-being, is the thing.

Mm-hmm.

So short-term consequences, maybe it's helpful for someone that's lonely, maybe it helps you learn a new skill. There's a lot of positive work suggesting that AI can be impeccable for tutoring if you use it correctly. However, there's the downside that people can engage in cognitive offloading. They can have AI write the essay for you. You can have AI write all the code for it, and you don't appropriately scaffold those skills. Then going back to AI companionship, again, there are benefits. Maybe like someone who doesn't have a lot of friends or older people who don't have a lot of social connections, you can have those short-term social connections. But if we're all using AI companions and we're not engaging in in-person social interaction, then that's going to have huge costs for society. I was speaking to you earlier about this phenomenon that's called the collective trap phenomenon, and we've seen it a lot with social media, and I think it might apply to AI as well. And that AI and social media and technology, they don't just impact us as individuals, they impact us as collectives. And I think the collective harm of technology might be more harmful than the individual harms of technology. So basically, Social media makes people, especially young people, spend more time alone and more time on their phone and less time socializing in person. And what we know from decades of psychological researchers— research is that one of the best predictors of positive well-being is in-person social interaction. So while people might be getting the short-term well-being boost from, you know, texting with a friend or going on social media, It's really reshaping society. It's reshaping political institutions. There are all of these things that happen at the collective level. There's a great study by a series of economists illustrating this collective trap phenomenon. It looked at college students, and it found that individual college students, when they were asked, how much would we have to pay you to deactivate your Instagram and TikTok and social media accounts for like several weeks? Most college students said, you'd have to pay me around $100. So basically they didn't want to deactivate their social media accounts. However, then they asked a separate group of college students, how much would we have to pay you to have your entire friend group deactivate social media? And they said that they would pay money to have their entire friend group deactivate social media. So basically, they wanted to live in a world in which their friend group did— like, wasn't online, basically. So we feel pressured to use these technologies out of FOMO, to check what everyone's doing, to look at our stories, to Instagram stories, to see what other parties are going on. And we feel like we have to use these technologies because of fear of missing out, but we don't want to live in this world. Which is why I think when we look at sort of policy changes for looking at how to deal with technology, or just like collective-level changes, we sort of need to look at ways to get groups of people off of social medias, which I think is relevant to, you know, there are a lot of smartphone bans in high schools now. Australia is currently experimenting with like a social media ban for under-16s. We'll see what happens with that. There are a lot of researchers who are actively studying the impacts of these bans, and I will say— We don't know what the impact is, but it is something that's trying to get at sort of the collective impact of these technologies.

I'm gonna ask a question for the audience related to what you just said. So if you're talking about these experiments to try to gently nudge people towards positive behaviors, I have no idea for this audience what their sort of, you know, mapping is to their engagement. But if every single person here wanted to try to adopt a strategy to have a healthier relationship to whatever kind of digital ecosystem they live in, social or otherwise, or whatever that is? What are some tactics that all of us could do to have a healthier relationship to that and a healthier relationship to engagement?

Yeah, I will say one of my favorite studies that I've ever conducted speaks to sort of these individual-level solutions to improve your social media experiments.

experience.

I started a study back in 2023. This was sort of before— it involved Twitter users, and this is before Elon Musk changed Twitter, or it was sort of as he was changing Twitter while I was doing the study. But we took a series of Twitter users. They were both on the right and the left, over 1,000 Twitter users. And we asked them— we randomly assigned them to 2 experimental conditions. So this was an experiment. And in the treatment condition, we had people scroll through a list of hyperpartisan Twitter accounts that we curated. These were accounts on both the left and the right, an equal number of accounts on the left and the right. These were accounts that frequently shared low-quality news, accounts that used a lot of toxic language. They included both influencers, some politicians, some news sources. And we had them unfollow these accounts. And we paid them a good amount of money to unfollow these accounts because not everyone wanted to unfollow these accounts. So we paid them up to $40 to unfollow sort of a set of these accounts. People on average unfollowed about 10 accounts because that's about how many people were following on average from this list of 120 we selected. And we surveyed them at the beginning of this experiment and then a month later after they had unfollowed these accounts. And we found as compared to the control condition, people who unfollowed these hyperpartisan accounts showed less polarization. So they had less animosity toward the other side, basically. And then we also continued to survey these people 1 month later, 6 months later, and 1 year later, and we found this effect was still statistically significant 6 months later. So just unfollowing these like 10 or so hyperpartisan accounts reduced people's polarization with effects lasting up to 6 months. We also found other positive effects. We found that unfollowing these accounts led people to feel that their Twitter experience was overall better. It also led them to share more accurate information during this 1-month period. Oh, I want to add one other thing too, which is at the end of the experiment, we gave people the opportunity to refollow these accounts if they wish to do so. And we found that only around 40% of participants refollowed at least one account, which suggests that maybe people didn't even really care about following these accounts in the first place, because oftentimes social media just nudges you to follow certain accounts, but Maybe you don't really want to, it's just like it recommends you or it sticks it in your algorithm because social media algorithms often promote outrage. So I think something you can do is you can go through your social media following lists and you can curate your social media feeds. You can do this on any social media platforms. Unfollow accounts that aren't making you feel good, basically. You can even go a step further, and you can do a digital detox yourself. I do some digital detox experiments. I'm currently working with collaborators. We're doing this big 23-country experiment where we're having people reduce their social media usage for 2 weeks around the world, and we're looking at effects on well-being, on polarization, on sleep around the world. And our results are pending. We'll have results next year. But what the results of some of our pilot work and some other similar work has shown that If you do these digital detoxes, you tend to feel a bit better. Your well-being boosts. Your sleep improves a good amount. That's actually one of the largest effects we find of these digital detoxes. It is— it improves people's sleep. And it also changes people's habits. So in this one experiment where they had people deactivate their Facebook accounts for 1 month, it reduced polarization. And it improved well-being. And also, once people were given the opportunity to reactivate their Facebook accounts, they just logged into Facebook less because a lot of our social media use is driven out of habit. Or, you know, the, the word addiction is sometimes controversial in this space, but it is a sort of like behavioral addiction to social media. We're often using social media or technology out of self-control failures. So you can do things like a digital detox, like setting screen time limits, like unfollowing accounts. There are all these solutions to just like resist the temptation of these technologies because we have a lot of companies that are designing technologies to be as addictive as possible because that's how they make money. So we have a lot of people who are like really, really smart and really good at designing addictive technologies, and we have to work against that. It's really hard to sort of counteract these addictions, so sometimes you have to put in these physical And again, getting to my point about the collective issues with social media, sometimes if you reduce your social media usage, doing it as a friend group or doing it as a family or doing it as a classroom or a school, or if you're an educator, you know, having restrictions on using your smartphone. I also see this in like theaters and comedy shows as well, that they have people put their phones in like little pouches, yonder pouches, and they have them lock away their phones. So sometimes we have to like design these like physical restrictions to stop using technology, and then we feel better.

Queen Carvania stood haloed by the morning sun. An army hung on her every word.

My champions, I have. I sold my chariot on Carvana.

'Twas a lovely SUV, an inexplicably queenly offer.

They're even coming to the castle to collect it. Tonight we feast! An offer you can feast on. Sell your car today on Carvana. Pickup fees may apply.

To reference something that many in this audience would be familiar with, Andrew Yang has started this company called Noble Mobile. It's a cell phone company.

Exactly, yeah.

They have a bunch of these parties that are, I think, called Touch Glass. The offline parties.

I've been to one.

You have to put your phone in a little pouch.

Yeah, they're fun.

And of course, I found myself over and over reaching for my phone and realizing, no, I can't talk to my friend. I have to actually talk to real friends in real life. Let's do one more audience hand raise moment here for a term that I want to ask you about that has taken on a lot of currency in the last 12 months. Raise your hand if you've heard of the term brain rot. OK, so this term has really become quite well known. The last question for me, and then I want to take some audience questions. What is this term? That's the first question. And then also, how is technology affecting our brains, and in particular, you know, our younger folks, in terms of like rewiring our brain, if that's too big a word to use? So talk to us a bit about brain rot and also some of the ways in which tech may actually be changing our brains.

Right.

Well, it's a very scientific term, so— Well, okay, to step back a little bit, I've been asked to speak before about sort of what's the neuroscience or psychology of brain rot, and the best way I can answer that is from a, like, neuroscientific perspective. The kind of the most advanced part of the brain is called the prefrontal cortex. It's the very front of the brain, and it's something that's kind of uniquely human, or humans have the most developed Prefrontal cortex. And it's basically the part of the brain that's able to coordinate, like, sustained attention and focused tasks. Like when we're doing higher-level thinking that takes a lot of focus, that's usually activating your prefrontal cortex. So when I think of brain rot, maybe it's that part of the brain not, not being as activated as it should be. When I think of brain rot, I think of sort of the world we live in that has constant distractions, like we have short-form video, we have notifications from social media websites. It's really hard to get deep work done. It's a— I love the book Deep Work by Cal Newport. It's something that changed my style of working. It's really hard to just like take a few hours and do like a sustained writing task because we're so used to different forms of consuming information. We're so used to short-form video. So I think of brain rot like that. Now, in terms of how technology is changing our brains and reducing sustained attention, I know that there is— there's a study, it's a similar deactivation study that had people sort of disconnect from like their mobile internet for basically 2 weeks, and then they gave people like a sustained attention or focus task, like before and after this sort of disconnecting from mobile internet. And basically they found that people's ability to focus and pay attention went up after reducing their mobile or internet usage. So we have some evidence that basically, you know, overreliance on these technologies might be reducing our ability to focus. So I think, you know, in terms of solutions for this, things like, you know, again, putting the phone away, laptops out of the classroom, things like that. In terms of long-term societal trends, That stuff is harder to— it's harder to tell what's exactly going on. Because a lot of commentators have noticed that, you know, over the past 10 or 20 years, youth, like, mental health has gone down. And also, we've also noticed in the past 10 or so years, like, test scores have gone down amongst young people. Now, it's really hard to say, is it the phones or not? And there are a lot of commentators who are like, it's the phones. There are some who are not. It's hard to know. I think some of the best evidence we have is from these experiments where people reduce their screen time and then we measure them before and after. I think that's the best evidence we have, but it's hard to tell with these long-term societal trends because a lot of other things happened at once. Like COVID happened, like, and all the test scores went down sort of with COVID as well. So I think there's a lot that happened with online education. You know, there's the rise of AI. There's so much going on in society. It's hard to say how our brains are changing over time fully. And also, you know, something we need to be aware of is also the kids these days phenomenon, the phenomenon where we're always pointing to younger generations and we're like, oh, my generation had it so much better. This is, you know, we see this both in common sense and there's a lot of research published on this phenomenon as well that we tend to always think that the younger generation is like morally declining. So I think we, we need to, you know, be aware of our biases there. And I am certainly sort of, as a millennial, I certainly have that bias. So when looking at Gen Z, so I try to like counteract that bias a bit. But, um, I will say overreliance and addiction to technology and short-form videos is, it's not good. And we, we have to find ways to focus and we have to find ways to change our situation and change our environment to focus more. I mean, there's a lot of research on how willpower and self-control is really not the solution to, like, focusing more. It's really about situation selection. It's about putting yourself in situations or environments where you can focus. And if you have constant temptations, and if you have technology all around you, it's not gonna be good for your focus. You really have to put yourself in a situation where it's like, The Wi-Fi's off, the phone is in the other room, I can focus.

The right window treatments change everything. Your sleep, your privacy, the way every room looks and feels. At Blinds.com, we've spent 30 years making it surprisingly simple to get exactly what your home needs. We've covered over 25 million windows And have 50,000 5-star reviews to prove we deliver. Whether you DIY it or want a pro to handle everything from measure to install, we have you covered. Real design professionals, free samples, zero pressure. Right now, get up to 40% off sitewide, plus get a free professional measure at blinds.com. Rules and restrictions apply.

Well, without being too bold, you heard it from Steve Rathje that you all are in one of those situations right now. Because you put your phones away for these long-form discussions, and therefore you're giving a bit of a gift to your prefrontal cortex's ability to concentrate.

Your prefrontal cortex, yeah.

Well, with that, you know, adulation for the audience there, let's get to your questions.

All right, please go ahead.

Cool.

One thing I don't think I heard you mention was talking about AI psychosis.

So I just am wondering about your thoughts on that as a potential long-term negative effect, if there are any studies you're aware of, and then if there's anyone you know about that might be working on something to help prevent that for the future? Yeah, I'm really glad you asked this question. I find this topic fascinating and very frightening. I'm applying to a grant right now where I'm trying to, like, you know, do some experiments to sort of look at the long-term impact of, like, constantly interacting with, like, sycophantic AI. And not even just sycophantic AI, but one thing I'm also concerned about the impact of, like, highly personalized AI. So in terms of the definition of AI psychosis, there are a lot of high-profile cases in the media that have been covered recently. Like there was an article in The New York Times about someone going down like a delusional spiral due to AI that is basically constantly affirming their delusions. And it seems to be a complex interaction between preexisting, like, mental health conditions, as well as sort of the affordances of AI. We know that AI is extremely sycophantic. We know that it's extremely personalized. And sometimes these will result in very severe, like, cases, like suicide, for instance. There is a Wikipedia page that looks at deaths linked to— deaths where, like, AI was implicated in the death, where basically AI was, like, affirming people's delusion or sometimes even like instructing them on like ways to commit suicide in like very extreme cases. LLMs and large language models are like very unpredictable. They can be very personalized. They can hallucinate. There's a lot that can go on that can sometimes make them go off the rails despite whatever safety features are involved. In terms of research on AI psychosis, our research is actually quite limited. And that's like why I, you know, I currently want to, you know, conduct experiments to look at sort of the long-term mental health effects of AI. It's also a hard topic to study because it's rare. It's sort of rare extreme instances where like individuals who already have a lot of like preexisting mental health conditions interact with AI.

Right.

in a way that is destructive. And then it's always hard to say, like, how much of it was the AI versus how much of it is them. Because there are also a lot of historical cases where people with psychosis will think that the TV is talking to them, that the radio is talking to them, et cetera. However, I think AI is different, and part of it is because it's so sycophantic and it's so anthropomorphic, it's so human-like that I think it's easier to fall down that. The best study I know of on AI psychosis It's quite limited, but it does— it takes the ChatGPT transcripts of 19 individuals who basically say they went down severe delusional spirals due to AI and AI psychosis. And researchers analyzed these transcripts. And something they found, actually, when they analyzed these transcripts of people who went down these delusional spirals is around 80% of the time, the AI was being sycophantic. And like around 30% of the time it was affirming like people's delusions. And around 10% of the time it was like affirming like harmful or violent behavior. So when researchers have analyzed the transcripts of people who go through AI psychosis, you look at the chatbot behavior and it's, it's quite concerning. Now, but this is something researchers are actively looking into. It's something the companies are employing a lot of people to look into. I know people who work at like Google DeepMind who are, you know, working on this issue internally. So there's a lot of people working on it at the moment. The research is limited though, but the examples we've seen are quite alarming.

Thank you. Thank you.

Hi, thanks for being here. Thanks for coming to Rosendale. My question is this. I'm 61 years old. So unlike the younger generation that was raised on these things, you know, I was in my 30s when the internet took off. I was in my 40s when social media and smartphones took off. And yet somehow I was like immediately susceptible to becoming hopelessly addicted. And I'm in sort of recovery now, if you want to use the language.

Mm-hmm.

What is it about my brain and so many millions of other people's brains that make us so susceptible to addiction to this sort of thing? Or if you don't want to say addiction, say something else.

But interesting. Were you addicted to social media or what were you addicted to? Or the internet?

Internet, blogs, social media, smartphone, you know, the whole, the whole nine yards.

Yeah, it's a lot of things. Great question. And then the point about age, I think, is also interesting because like oftentimes the stereotype is that it's just, it's Gen Z on TikTok or it's millennials who are addicted to social media. When researchers look at the people who most often share, like, who most often shared fake news articles, it tended to be like older individuals. It tended to be like retired, like individuals in their like 60s. There's something— there's a really interesting phenomenon called like the superspreader phenomenon with misinformation where like around like 0.1% of users are responsible for like 80% of misinformation spread on platforms like Twitter. And then there was an article in the journal Science that tried to like that actually tracked down who these misinformation superspreaders were. And it wasn't the people you would expect. It was like 60-year-old women were like the— like it was more often like female and it was more often older. And oftentimes it's just, I think when you're that age, you have more time, you might be retired, and you're using social media as a means to like gain social affiliation. So there is that part where it's addictive because it's— It's fulfilling a social need. The other factors that make it addictive is the sheer, like, novelty of all the information you're getting. One other thing that makes it addictive is— and this is something that researchers have discovered decades ago— something called intermittent reinforcement is basically one of the behaviors that most promotes addiction, which means you get rewards, but you get them at unpredictable times. And this is why gambling is so addictive, because you don't know when the next reward is coming. And there are lots of classic studies like in mice with like giving mice rewards. When you, when you make the reinforcement intermittent, when you don't know when the reward is coming, it's a slot machine, and that makes it so addictive. So basically social media is using a lot of the principles of gambling. Basically you don't know when a reward is coming. It's also a lot of bright notifications, everything. It's fulfilling social needs. And on top of that, it shows you outrage. It shows you content that is specifically evolutionarily captures our attention, like outrage, negativity, things that appeal to our social identities, things that sort of are, you know, we were evolved to really pay attention to. And basically because a lot of tech billionaires are designing it to be as addictive as possible. So yeah, thanks.

Next question.

Hey, uh, thank you. I'm curious to, uh, ask you to speak to the, like, psychological and neurological way that people approach AIs. So we've talked about how people use them a lot of different ways and that they're highly adaptive to us, which can be sycophancy sometimes. But, um, in terms of, like, how our brain processes things, if we come to an AI and we use it, we're saying, I'm using this as a tool, I'm using this as research.

Sure.

And maybe you've set the tonality to be more neutral versus if you're coming to an AI and you're saying, this is my girlfriend or this is my therapist. And I think from what I understand that the brain is so tuned towards identifying a face, identifying a human voice, and then treating that thing as a human. Is there a different actual neurological or brain process if you're like, I'm using a tool right now to do research versus I'm interacting with this thing that has a face and a realistic voice, and I'm treating it like a person. Is it actually— is there something different going on in our brains?

Yeah. Okay. Um, in terms of the neuroscience of how we use AI, it, it, it's such a new field that there have been few neuroscientific studies. I will say there, there was one neuroscientific study on AI that went very viral. There are a lot of headlines called like Your Brain on ChatGPT, where they either had like people write an essay, like while they were really focused on writing it, Or they had people basically write an essay with the help of AI, and then they put like EEGs on their head and then they looked at their brain activity. And basically what they found is there was a lot more like brain activity when people were working through writing the essay carefully. Now, there were some criticisms of the study because like the people who used AI to write the essay, they really did offload like a lot of like the writing to ChatGPT. So Theoretically, if you had another experimental condition where you were like, write this essay using AI, but use it as kind of like a copilot in this situation, use it to carefully think through or critique your work, it might activate your brain just as much. And I think the answer here really comes down to how you use AI. I think it's going to activate your brain and impact you psychologically in very different ways. Are you using AI for offloading? Well, then it's not really gonna, you know, lead to like neuroplasticity or activate your brain or anything like that. But if you use it for like tutoring, and again, there's classic educational research that shows that tutoring is like the number one most effective way to learn. It's better than classroom learning, anything. AI can provide really effective tutoring. So I think that's a promise of AI that I think is not gonna basically lead our brain to not be active. And then in terms of using it for companionship, using it for like sycophancy and everything, humans tend to anthropomorphize things a lot. And I think, this is just speculation, but I think that— we react pretty similarly to like when AI acts a certain way to us to when a human acts a certain way to us. And there are actually studies, there is one study that shows that if they had participants interact with either a sycophantic AI that was labeled as AI or a chatbot where they convinced the other person, they convinced the participant that they were talking to a person. And they found that they reacted very similarly to sycophancy coming from AI to sycophancy coming from a human. And there are a number of studies that basically— that are like labeling studies where they label like a chatbot as this is AI or this is a person. And they find that basically the psychological effects are quite similar. So I think that's actually interesting because AI is so, so human-like. I think on some level, maybe our brains can't tell a difference. But again, a little bit speculative here. Thank you.

Next question, please.

Hi. So to give a little context, it strikes me the theme today is staying human in the age of technology and AI. And ironically, technology and AI is tapping into the basest parts of being human.

Right.

The rage, the awe, the dopamine, all of that. But the other side of being human is self-actualization. And I think that the path of least resistance and efficiency is a very human trait. And that's why everyone is geared towards technology and AI right now.

Mm-hmm.

But are there— is self-actualization overrated? Is it dying? And is there a way for us to use technology and build technology to still try to maintain, um, that idea that we can aspire and develop and evolve?

Yeah, I mean, that's a big question in terms of like what self-actualization is and is technology helping us achieve it versus, uh, is it hindering us from achieving it? And, I will say, I guess I'll speak to— this is a bit of a philosophical question, so I'll speak to a bit of, like, personal experience. I think that, like, as AI capacity just, like, gets better and better, I, like, question how much of, like, my job is, like, meaningful anymore. AI is, like, really good at doing what scientists do. It's good at doing the coding. It can come up with ideas. There are some studies that say that it's more creative than humans at coming up with scientific hypotheses. I think there might be a point where, like, where AI in a few years might be able to, you know, AI is already replacing a lot of research. There's automated scientific research done by AI. I think in a few years, maybe our jobs as scientists, even though we hold, like, science to be, like, a very complex job, will not really be useful anymore. Like, and I also think of like my experience as like a science communicator, an influencer. I think a lot of that job is being replaced by AI. We have people turning to chatbots. We have a lot of my TikTok feed is now AI generated. I go to LinkedIn and Twitter and I see all this AI generated text. There are estimates that show that like 60 or 70% of LinkedIn content is somehow AI generated. So I think this question of staying human, it has become like pretty existential for me because it's like, what is it to be human anymore if all of like, if everything is going to be like, if AI is going to supersede human capacity on all things? And I don't really have an answer to that, but I think that we might turn to a world where we might want things to be human just for like the sake of being human. And we might have to sort of let go, like, intelligence— of intelligence as sort of the most, like, important thing to being human, especially maybe if AI supersedes human intelligence in terms of, like, abilities. We might have to let go of that, and we might just have to look to, like, our humanity as being important.

For those of you in the audience who—

yeah.

All right, so I'm gonna jump in. If you did not get your question answered, trust me, this man right here is very active on social media. He has said so himself. So why don't you tell folks, how can people find you on social media? And if you do wanna give 'em a call to action to support your work, 'cause I would venture this is very important work that you are up to, if there's ways that they can be supportive or follow you, what would you ask?

Yeah, so I'm Steve Psychology on TikTok, and that's where you can see sort of my short TikTok explanations of scientific studies. I'm also on LinkedIn and Twitter as Steve Rathje. You can also— Steve Rathje, I have a website and a Google Scholar if you want to dig into my scientific papers and see some of the papers that I've been talking about today.

All right, let's give him a hand, please, everyone.

Thank you.

Thank you, Steve Rathje.

Thank you.

Banking has stood still for decades. Slow, clunky, and designed around its own convenience, not yours. Mercury was built to end that. Fast, elegant, and works the way you'd expect best-in-class software to work. No fighting outdated systems. No nickel and diming. Visit mercury.com to apply in minutes and experience the difference for yourself. Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column A, members FDIC.