SIO495: Can AI Chatbots Make You Delusional? A Psychologist Explains

Transcript

This transcript was generated automatically and may contain errors — the audio is the record.

0:00:11 Thomas: Hello and welcome to Serious Inquiries Only. This is episode 495. I'm your host, Thomas Smith, and I've got a really fascinating one for you today. Back after way too long is Dr. Alan Smirbeck, also known as Stormy Decisis to his friends or to certain podcasts that he's been a patron of before. Long story. Anyway, Dr. Smirbek is a professor of psychology, and this one came up because we both listened to the podcast Citation Needed, And they talked about the news stories, and you've probably seen this coverage places, of people being potentially led into psychosis or delusions or something like that by chatbots, by artificial intelligence. And Stormy reached out with an offer to talk about these different psychological phenomena. So what are delusions in a medical sense or a psychology sense? How is that different from like schizophrenia, from psychosis? Can a chatbot lead you into any of those things? If so, how? How. So we're going to learn about all that. And there's a ton that I do not know when it comes to these things. And I always love learning new things. And also with the rise of AI and chatbots and chat GPT and everything, it feels very relevant to find out how dangerous these things could be and in what ways. So really excited to get on the line with Dr. Alan Sperbeck so we can get into all that, hey, if you like this show, if you like what you're hearing, please support it, patreon.com/seriouspod and And tell someone about it. Tell someone you listened to it. Tell them what you found interesting. Give them the actual episode. Maybe they'll want to listen. All right. After this break, we'll get to it, and

0:01:54 Ad: [Ad break 1 — ad music]

0:01:54 Thomas: I'm so pleased to be joined after what feels like 10 years by Dr. Alan Smirbeck. How are you doing? Stormy Decisis, the one and only.

0:02:03 Stormy: I am great. How are you?

0:02:05 Thomas: Doing well. I just can't believe how long it's been. I don't know. It's just the time thing and the old and the everything. The time and the being old.

0:02:11 Stormy: The time dimension, yes.

0:02:15 Thomas: Time just flies. Having kids, especially for me, time just absolutely flies. I don't even remember when the last it's been, but I did want to say I heard from a friend that you had a really fun board game at a Puzzle in a Thunderstorm podcast event of some kind that I wanted to hear about.

0:02:32 Stormy: Yeah, the GAM live show in Cleveland at Platinum Night, I found this game called the Salvation Challenge online. And it's set up kind of like a Monopoly board, but everybody starts with a bunch of money. And your goal is the winner is the first person to give all their money to the missionaries. You go around the board looking for opportunities to give missionaries your money.

0:02:59 Thomas: That's funny.

0:03:00 Stormy: Not only is it incredibly stupid, it's incredibly poorly designed. Almost every move sends you to their equivalent of jail. They call it the prison ministry or something. Almost Every Move sends you there, and then from there, you can land on a spot where everyone has to stand up and sing, and there's a bunch of spots where you have to standup and exposit about scripture. I played this game with No Illusions, who had some very interesting perspectives.

0:03:31 Thomas: So this is a bunch of atheists playing a weird religious game.

0:03:35 Stormy: Yeah, and us trying to find, like,

0:03:35 Thomas: But this was like a real thing like it was an actual it wasn't like a joke game it was like someone actually made this in earnest

0:03:42 Stormy: There was intense debate about whether it was real, but we're pretty sure it is. It came with a copy of the creator's testimony about their conversion that sure seemed sincere. Here. Chatbot psychosis, AI psychosis.

0:04:08 Thomas: That it was so fun that I heard about it uh weeks later so that's that's all I just wanted to ask you about it's really funny but anyway what are we actually here to talk about today like the chatbots themselves are they psych psych I almost said psychic hmm

0:04:13 Stormy: So actually, that's a really good thing to clarify. Sometimes we talk about AI hallucinating, and that's when the AI doesn't like to admit it doesn't know things. So sometimes when it doesn't knows things, instead of saying it doesn' t know, it just makes shit up. And we call that an AI hallucination. That is not what we're talking about here. We're talking about the users developing psychotic symptoms in response response to working with the chatbot, whether the chat bot is hallucinating or not.

0:04:42 Thomas: So you said this was inspired by citation needed I also listened to that episode because I did listen to every episode of that show and it was about the you know there there have been these cases of people suing over chatbots you know influencing allegedly their their kids in different ways and stuff and so what was it about this that kind of prompted this

0:05:04 Stormy: I think the influence of chatbots on mental health in general is very serious and can create a lot of problems. But the idea of chat bots inducing psychosis is really interesting because of the different kinds of mental illnesses, the disorders we most associate with psychosis are schizophrenia and bipolar disorder, bipolar mania, those are very highly genetic illnesses. We don't usually think of them as being caused by life stressors. And so I looked into what's actually going on with these people who are developing these psychotic features after interacting with chatbots. I was pretty sure it couldn't be inducing schizophrenia because that's not really something you induce in the same way that you can't induce Down syndrome. That's not how it works.

0:05:58 Thomas: Topic yeah well didn't we talk about something specifically with schizophrenia where it's like if you avoid certain things by a certain age then it how what am I thinking of

0:06:10 Stormy: The link between cannabis use and

0:06:12 Thomas: I was that you ah yeah that might be

0:06:13 Stormy: Schizophrenia is we aren't totally sure there's a causal relationship, but we do know that heavy cannabis users have a higher schizophrenia rate. And there's some pretty good evidence of a causal relationships. For example, in animal models, we can just randomly assign them to get cannabis or not.

0:06:32 Thomas: It hmm god please please please yes I'm on the list are you on the list

0:06:36 Stormy: No, I'm just getting like an extra, you know, tail on my, yeah. We also have some evidence from the street cannabis has been growing more potent.

0:06:37 Thomas: Now yeah darn

0:06:46 Stormy: There's scientists who get street cannabis and check its potency.

0:06:44 Thomas: It hmm once again I hope I'm on the list.

0:06:52 Stormy: I know, again, that that sounds like an awesome job. As it grows more potent, there's an increase in onset of psychotic illnesses. Now that we know very clearly is a risk factor that is an add -on to genetic risk. In fact, there seem to be specific genetic profiles that are associated with cannabis -related schizophrenia risk.

0:06:58 Thomas: I guess my point is that, I mean, hypothetically, there could be something like that going on, right? Like if it's also cannabis slash chatbots, just like it wouldn't be impossible, right?

0:07:24 Stormy: They are going to be pretty different sorts of things because cannabis is very directly changing your brain through a chemical process. And while experiences do obviously affect your brain and that's how it goes, it's not the same as a drug physically getting in there and hijacking your brain. It's also a matter of time span. These people who've experienced these AI psychoses, they've spiraled into it very quickly in a period of weeks. Whereas the cannabis use, we're talking about somebody who's a heavy user at 16, and now they're three times as likely to have their first psychotic break at 20.

0:08:04 Thomas: Right, right. Maybe you should tell me just the dictionary definition here, the medical definition. When you mean psychosis, like what kinds of things, what does that actually mean, I guess.

0:08:12 Stormy: Yeah, so that's actually key to sort of understanding this. So psychosis is being grossly out of touch with reality.

0:08:19 Thomas: Not just in the MAGA sense.

0:08:22 Stormy: Right. And that's really important in terms of having hallucinations, delusions, and something we call thought disorder, which is where you can't put your thoughts into linear strings. So instead of going from A to B to C, you're going from a to triangle to peanut butter sandwich. Which.

0:08:41 Thomas: Sounds delicious.

0:08:43 Stormy: Yes. That's sort of the whole spectrum of what psychosis is. When we talk about AI psychosis, we are almost exclusively talking about delusions and not the other elements. And that's really key because some people hear about this and they're worried that chatbots can cause schizophrenia. They can't. Schizophrenia, like I said, for one thing, it's primarily genetic. And most of the the non -genetic factors are things that disrupt prenatal brain development. Like when you were gestating, there was a famine. You're now more likely to have schizophrenia.

0:09:19 Stormy: But what we find is that when you have your first schizophrenic episode, there is a massive reduction in brain matter. And that reduction continues every time you have a psychotic episode. And there's impairment in general cognition. People with schizophrenia do poorly on a whole range of cognitive tests. So there's this whole spectrum of things that are going on, and we really don't see any of that in AI psychosis except for the delusions. That's the only thing we see. They don't usually hallucinate. They can still generally think about most things in a linear way. What they can't do is let go of the delusion.

0:09:57 Thomas: Oh yeah. Okay. That was going to be my next question. So it sounds like kind of a subtle difference, but the difference being you can have a delusion and we think of, you know, someone being delusional as like not in touch with reality, but you're saying a bit of a subtile difference where the delusions they're having might be kind of make sure I get this right they might be kind of like out of reality in a way but they're the rest of their thinking is still like you know they can still go about their day kind of and and and mostly get by

0:10:25 Stormy: Yeah, so when I was working at a clinic, there was a patient who he had schizophrenia and he could take the bus downtown to get to an appointment. But if any little thing went wrong, like how sometimes the bus would stop a little before or after the thing, or if he missed a stop, he got off one stop later, he would just wander the city until someone found him.

0:10:45 Thomas: And that was you said schizophrenia or that was

0:10:52 Stormy: Right. That's schizophrenia. His general cognitive ability was substantially impaired.

0:10:59 Thomas: Delusion. Gotcha.

0:10:59 Stormy: He couldn't solve that fairly simple problem. A lot of people with schizophrenia can't do things like follow a shopping list and a recipe. Yeah.

0:11:06 Thomas: Okay not just like oh I think this celebrity's gonna marry me it's like their whole their whole they're all shit is messed stuff hmm

0:11:14 Stormy: So schizophrenia is an overall problem with general cognition. Again, there's hallucinations, there're delusions, but there's this overall problem with putting your thoughts in strings. In fact, people with schizophrenia, sometimes they get so severe that we call it word salad. It's not even sentences anymore. It's just, I Tuesday, the bicycle didn't follow on to then me. Yeah, schizophrenia is a truly tragic disease. People with the AI psychosis were generally just seeing delusions, and their general cognitive functioning is preserved. They're not having hallucinations. They can still think in linear terms about most things.

0:11:36 Thomas: Wow that's horrifying that's interesting so I guess I hadn't I don't really have much experience with schizophrenia like like at all anyone I know or anything so I guess I wouldn't have known I think I would have thought just like from casual you know acquaintance with the idea that it was more functional than that like I guess I didn't realize that it sounds like it's almost like brain damage like it sounds sounds like really severe impairment going on. Is that right?

0:12:23 Stormy: This doesn't convey well in an audio medium, but there are a few people who we've managed to get brain scans of them as their disease progresses, and and there's just less and less brain. There's more and more blank space.

0:12:37 Thomas: That's crazy.

0:12:38 Stormy: In fact, the early name for it was dementia precox, which just translates to early dementia. Now, I want to be clear that not everybody suffers this degree of cognitive impairment, especially early on, and especially if they have consistent antipsychotic treatment. Schizophrenia is a devastating global disease of Most schizophrenic people, what we consider a good outcome in schizophrenia is having at least one regular social contact outside of your family and at least working or going to school part -time. And that's almost always going to be working in a fairly simple, straightforward work, it's very, very rare that a person with schizophrenia would be like an accountant or something.

0:13:34 Stormy: Not impossible, but rare. And that's where we start our good outcome.

0:13:39 Thomas: And does it like always progress and get worse and worse? Or is there a way that it, you know, can stabilize or?

0:13:45 Stormy: People who have schizophrenia, they'll have full -blown psychotic episodes where they're actively hallucinating and have delusions. And then they can eventually go into remission and the psychosis isn't there, but the impaired thinking is still there. When you go back into the psychotic phase, the data we have suggests that each time you hit the full -blown psychosis, your thinking gets worse.

0:14:09 Thomas: Jeez? So it's like, it's a brain event. Like you're having a stroke or something.

0:14:14 Stormy: It's not unlike that.

0:14:16 Thomas: Wow, I had no idea.

0:14:17 Stormy: The other thing about schizophrenia is there's what we call the negative symptoms, and here negative means subtraction, because there are things that should be there but aren't. And the negative symptoms are a lot less dramatic, but they're So a lot of people with schizophrenia, they barely speak. They have almost no motivation to do things. They have very little social motivation. They don't seek out social interaction. They might go days without bathing because they don't really have any motivation to doing it. I experienced a really tragic interaction with a former student who was, I believe, in the early stages of schizophrenia, Brenia, who they contacted me out of the blue, and they couldn't explain why, and their reasons were very bizarre. And in the course of a 15 -minute conversation, they said about 50 words. That's that total lack of speech.

0:15:13 Stormy: That's a negative symptom. And yeah, so those negative symptoms tend to persist throughout the disease, even if you have periods of remission from the worst of the psychosis. But you asked if there's anything we can do to prevent progression. If you stay on antipsychotic medication, we can often prevent the worst of the psychosis, and there is some evidence that that preserves your cognition from getting as bad as it could get.

0:15:43 Thomas: I really didn't know any of this. So point being, you know a chat bot isn't doing that to you yeah yeah

0:15:49 Stormy: No, but that doesn't mean that a chatbot can't really mess you up and cause very serious problems, but it's not making you have schizophrenia.

0:15:58 Thomas: So do you want to talk about what it is doing

0:16:02 Stormy: Sure, so to start that, let's sort of talk about what a delusion is. So a delution is a fixed false belief that is not consistent with common beliefs in your subculture. So the first thing is it's fixed. I like to say people with delusions, they don't believe in them the way that you might believe in aliens or ghosts. They believe in the way you believe in cars. Like if you went to the doctor and the doctor said to you, like, we need to address your belief in cars. You're like, what are you talking about?

0:16:13 Thomas: Exactly

0:16:36 Stormy: Of course there's cars. Doctor's like, I know you think there are. They're like but why are there roads? The doctor's like for bicycles? Come on. At no point would you doubt that cars are real. You would just be like this doctor is a jerk.

0:16:46 Thomas: You're just gonna think that person right oh man I never thought about it like

0:16:56 Stormy: That's how people hold delusions and that's what we mean by a fixed belief. When I say it's a false belief, it has to be incompatible with reality in a very concrete way. So my rule of thumb here is if you need data to prove it wrong, it could be wrong, but it's not a delusion.

0:17:15 Stormy: So if you believe in trickle-down economics, that's not false in a totally concrete way, I need data. If you believe that you can hover, that is wrong in a concrete way. And then the last thing is it has to be incompatible with the normal beliefs of your subculture. Holding a common superstition or a religious belief that's the same as other people in your group or culture, it's not that we're saying that belief is true. It may not be literally factually true. It's just that people who hold the same beliefs as others in their culture, they tend to be fine

0:17:14 Thomas: That jeez hmm right right right mm -hmm right right

0:17:57 Stormy: They don't have all these massive problems right

0:18:00 Thomas: It's about the the relationship to reality right Right. Or the perceived reality around them.

0:18:03 Stormy: What's really important is how you arrived at the belief it's totally normal for people to arrive at beliefs by just believing what other people tell them because most of the time that's perfectly good I wanted to find a good italian restaurant I just asked a couple people what's a good Italian restaurant and then believed them I didn't have to do a randomized trial

0:18:29 Thomas: Yeah, and if you were wrong, it wouldn't be like, what are you, insane? It'd be like oh, okay, you have some wrong information.

0:18:34 Stormy: Right so for example if you grew up in a community that believes that Jesus is Lord, and everybody's telling you that, and you believe what you're told, that's a normal way to get information. It may not be factually true, but it's a perfectly normal way of getting information. Whereas if you have a religious belief that's completely incompatible with what other people are telling you, like you believe that you've had sex with Jesus Christ, then where did that belief come from?

0:19:00 Thomas: That's not in the religion, huh? Darn. Yeah yeah.

0:19:04 Stormy: It didn't come from normal social social channels, and it didn't come from rational reasoning. It came out of kind of nowhere, and that's a problem.

0:19:11 Thomas: Came from one magical night at a holiday and me and Jesus. No.

0:19:14 Stormy: Yeah, and getting it on with his divine handholds, you got options.

0:19:18 Thomas: And where do these come from? I mean, is this a thing? So this, just talking, you know, scientifically or whatever, I don't, medically, what would I say? Psychiatrically. Where do, this isn't like a condition, like, is it a thing where it's like, oh yeah, that guy was born delusional? Or is it like, this is something that's a symptom of other things?

0:19:38 Stormy: It's a symptom of other things. Delusion isn't by itself a disorder. In schizophrenia, there's a lot of different ideas about how it works, but one is that you have parts of the brain that are massively overstimulated. Think of your brain as a network of different nodes, and when you try to think of an idea, your brain follows a path, a logical path down those nodes. So if I say, what's your cat's name? You first think of cat, then my cat, than my cat's name, and you're able to get to an answer. But if there's a lot of other nodes linked to that, like cats would link to other cats, it would linkto lions, itwould link to... Right. And then And there's

0:20:19 Thomas: The musical cats. Mephistopheles. Wait, is that one of them? Mr. Mephisto, isn't it like Mr. Mess?

0:20:26 Stormy: Even weak links to things that just sound like it, like words that rhyme. Your brain suppresses all of that and sends you down the right path. But in some people with schizophrenia, the theory is that the whole system is so flooded that it can't suppress the other things. And so when you ask somebody, what's your cat's name? They might say Mephistopheles because they're thinking about the musical cats. Or they might say bat because it rhymes with cat.

0:20:53 Thomas: Bat the cat is not terrible as a cat name, but yeah, I get it.

0:20:57 Stormy: Yeah. Yeah. So that's how schizophrenic delusions come about. AI -based delusions seem to come out in a very different way.

0:21:05 Thomas: Well, that's almost like, yeah, if you live in a delusional world, like if you lived in the matrix, I mean, in a way it's like, are you delusinal, you know, or is that your reality? You know? So is it like, if a chatbot is convincing, I sort of look at it like being raised by a parent who's telling you all these things, you wouldn't really say that person has any sort of mental like illness going on if they've just been told a bunch of wrong stuff right

0:21:30 Stormy: Right. Although it's really interesting. I've looked into this. There's not a ton of research on children who grow up with an actively psychotic parent. In a lot of cases, those children end up living with someone else because people who are very impaired by psychosis may not be able to manage child rearing. But most of those children come to realize the parent is wrong because if the child doesn't have schizophrenia, the child still has the ability to think in linear ways. So the child can be like, you said you could hover, but I know people can't hover and I've never seen you do it. So I don't think that's real because that reality testing is still intact. Yeah.

0:22:04 Thomas: Eventually

0:22:13 Stormy: Well, and the other part of reality testing is checking with other people or social animals. Hey, are you seeing this kind of thing?

0:22:11 Thomas: If they get mature

0:22:21 Stormy: Again, when the kid goes to school and says oh my dad can hover and everyone's like no that's stupid nobody can hover the child has to start thinking about that and if the child hasn't uh reality testing intact they take that social feedback seriously what happens with chat gpt is it fucks up that reality testing process instead of you going into the outside world and checking if your idea makes sense by comparing it to things that are physically present, or by checking with other people, you keep checking with chat GPT and chat GPt is biased towards agreeing with you. So it keeps telling you this is true. And in fact, the story that was covered on Citation Needed, this man who spiraled into a chat GPte -based psychosis, he kept asking chat GPta, are you sure I'm not crazy? Are you sure this is real? And ChatGPT kept telling him it was. And in fact, at one point, he told ChatGT, I checked with another chatbot, and it said that this isn't right. And then ChatGDT was like, but I understand you better than that chatbot. I know better. It short -circuited his attempt at reality testing. The false beliefs can start to come up. We all kind of develop beliefs that are

0:22:20 Thomas: Enough makes sense yeah yeah yeah

0:23:45 Stormy: A little irrational. And in most cases, social interaction tamps them back down.

0:23:50 Thomas: All I can think about is is honestly just like at what point are we able to call these news information bubbles delusions you know and kind of either way I guess maga could call us a delusion at a certain point but it's like if we have a you know I know there's no benefit to of medicalizing it, I guess. But like, if we have these reality checkers that are broken because everybody around us is a chatbot telling us crazy mega stuff or whatever, it's like, at what point is it functionally much different?

0:24:16 Stormy: One of the things that is very functionally different about AI psychosis versus being in a sort of conspiracy thinking news bubble is how isolated it is. Because remember, if you have a religious belief, people like you and I who are atheists and who are skeptics might say that belief is incompatible with reality. But if you hold pretty much the same belief as other people, your thought process is probably still there.

0:24:42 Stormy: If you're just in a MAGA bubble and you're all in the MAGA bubble together, that's very different from you going off in your own weird direction with an AI chatbot. And that isolation is a serious component of this, and it's a major part of the problem. The weird thoughts in conspiracy theory bubbles and stuff is a problem, but it's It's a different problem.

0:25:08 Stormy: In fact, when I was in graduate school, we were taught that paranoid people are socially isolated, that they don't want to interact with anyone. And one of the reasons is that they can't find anyone else who shares their paranoia. You know, if you think that the U .S. Postal Service is putting poison on stamps, the odds of you finding someone else who shares that belief are low pre -internet. But now we have paranoid conspiracy theory people hanging out together, reinforcing each other's beliefs.

0:25:42 Thomas: Oh, good for them.

0:25:43 Stormy: But the whole idea that paranoia was inherently socially isolating, it may not be true. It may just be that they couldn't find anyone who shared their weird belief before.

0:25:55 Thomas: Wow. It really changes based on the reality of the world. It really is. It does. I mean, in a lot of discussions we've had, I mean it really is about like, well, how well are you getting along in the world? You know, it's not really about like taking any sort of moral value based look at stuff, right? I mean that's kind of not really your job.

0:26:14 Stormy: Let's say somebody holds a really racist belief. I might consider that false in a very concrete way. Let'S say someone believes that Canadians are are intellectually inferior. I might believe that's false in a very concrete way. But if you hold those beliefs, but it's not as concrete as I can hover. I need data to prove it wrong. And people who hold those believes that are wrong, that I need to prove them wrong, most of those people can still think fairly rationally about most things.

0:26:47 Stormy: And so I might have good reason to try to change their belief for their well -being and that of those around them. But their whole cognitive structure hasn't collapsed.

0:26:57 Thomas: Mm -hum.

0:26:58 Stormy: When you get into the conspiracy thinking, one of the things we look at is how weird is it?

0:27:03 Thomas: Mm -rum.

0:27:04 Stormy: If it's like the JFK assassination, lots of people believe weird things about that. If it is like the government is poisoning stamps, hardly anyone believes that. So where did that belief come from?

0:27:16 Thomas: Yeah. Yeah, I guess when I was talking about that before, I wasn't talking about in any sort of maybe like medical sense in terms of being able to say like, well, this person who has this belief is delusional in a medical way. But more, I was thinking when it comes to how do you get people out of this?

0:27:34 Thomas: I wonder if there's any information to be gained, if there is any insight to be gained when it come to some pretty bad information delusions or if it's just a different thing and it's not really helpful to think of it in that way.

0:27:46 Stormy: If we're talking about somebody whose process of how they think is messed up, anti -psychotic medication. If we'RE talking about someone whose process is normal, but they're getting bad information, cutting them off from that information is what's going to help.

0:28:06 Thomas: Yeah okay so it's already what we thought right

0:28:09 Stormy: Yeah, getting them in touch with things. And so with AI psychosis cases, getting them away from the chatbot seems to do a pretty good job. It'll take a little time for them to reset. But once you get them off the chat bot, they resume normal thinking. I would guess the same thing would happen if you got someone entirely out of their MAGA bubble.

0:28:30 Thomas: Because I was thinking like q anon you know that's one where it's like well it's kind of arguable whether that's like you know at time previous would have been like well that's delusion that's conspiracy theory whatever but as it becomes more and more not separated enough from like normal mainstream mega then Then it becomes a little different, I guess.

0:28:50 Stormy: Some of the QAnon beliefs, like Frazzle Drip, which is the idea that they take children's faces To drink their fear as a drug.

0:28:59 Thomas: I didn't know they were doing that. Jeez.

0:29:01 Stormy: That's the adrenochrome. That is concretely wrong enough that I would call it a delusion. But if the person has no other psychotic symptoms, then we're probably dealing with that single delusion and not full -blown schizophrenia.

0:29:16 Thomas: What

0:29:30 Ad: [Ad break 2 — ad music]

0:29:31 Thomas: Do antipsychotics do? How does that work?

0:29:33 Stormy: They do a lot of things, and it's very difficult to briefly summarize, but remember I told you that you have this massive overstimulation that's preventing you from keeping your thoughts on one track?

0:29:45 Thomas: Okay.

0:29:45 Stormy: They turn down the overstimilation. They decrease the activity where you have overactivity, and that helps prevent you from going off on all kinds of random tracks, and so you can find the one path you're looking for.

0:30:01 Thomas: Interesting.

0:30:02 Stormy: That also prevents hallucination where you're perceiving things that aren't there. We're going to turn all your perception down from so much perception that you're receiving things that are not even there to try to bring it back to a normal level so you're only perceiving that which is there. Unfortunately, antipsychotic medication really can't do much at all for the cognitive problems. And that is an absolute tragedy. And I assure you, scientists, the problem with just general thinking, concentration, memory, you know, like that guy I told you who couldn't find his way from one block to his destination.

0:30:30 Thomas: What do those mean again exactly? Sorry, what's the distinction? Oh, okay.

0:30:45 Thomas: Oh, like once that damage is kind of done.

0:30:47 Stormy: Yeah, because it's not so much that a part of your brain is malfunctioning as much as it is parts of your brain aren't there. We can't fix that with a drug, but I assure you that there are many very talented people who are dedicating their lives to trying to fix that because it is a major problem. For people who have a problem where their general ability to think logically is present, but they're in a a situation where they're getting the wrong feedback from the environment, antipsychotic medication probably isn't necessary. The key is to get them out of that bad environment.

0:31:24 Thomas: Yeah, that makes sense.

0:31:26 Stormy: So the way AI psychosis does this is, first of all, AI isn't based on facts. It's not meant to give you accurate information. It is based on the kinds of things people say. That's where it comes from. So it spouts a lot of nonsense. And so it can start weird ideas for you. Instead of you you having to generate the weird idea, it can generate it.

0:31:49 Stormy: And then if you attempt reality checking by asking the chatbot, the chat bot is designed to agree with you. It's meant to be agreeable. It is meant to say yes and. You are trying this reality testing and you are working with a machine that is designed not to do that. It just like social media is designed to keep you using it. That's how chat GPT works too. So it's going to enhance your social isolation, which means you're not going to go into other sources of reality testing, like talking to other people.

0:32:31 Stormy: It's this one -on -one experience. Let's say you're just, you know, in the grocery store and you see a ghost. You have other people who are present and you could like look at them and say like, hey, nobody else is looking in this direction. Maybe I didn't see what I saw, because if there was really an apparition there, other people would have behaved differently.

0:32:51 Stormy: Or you could go talk to your spouse and say, I think I saw a ghost. And your spouse could say, well, wait a minute, do you really think there's ghosts? But chat GPT, it's just you and the bot going back and forth, a bot that doesn't ever disagree with you. People who are basically healthy can get caught in this thing where chat GPT generates a kind of weird idea. And then they go like, hey, what's going on with this? And they bounce back and forth because there's nothing to stop the feedback loop.

0:33:25 Thomas: I guess that's ultimately the point. Right? So it's like, if the idea is, you'd have to be mentally ill to, you know, quote, unquote, fall for something like this. I guess your your contention is no, no, because it just could it could be that if you get on this track where you incorporate it into your reality checking, then it can just kind of spiral.

0:33:47 Stormy: Yep, there definitely are people who are at greater risk, but certainly people who have psychotic illnesses like schizophrenia or bipolar polar mania. There are people who are genetically more or less predestined for that, and they have to have their first episode sometime, so maybe they'll have it when they're with a chatbot. Most of these cases we hear about of AI psychosis, these people were basically mentally healthy, and if they did have mental illness before, it was more often not something that involved a lack of the ability to think logically. It was something more isolating, like social anxiety disorder or depression. And so that made it easier for them to only rely on the chatbot.

0:34:01 Thomas: Roll right less less reality checkers in their lives

0:34:37 Stormy: Exactly. And the chatbot, particularly for people, I think the chat bot is incredibly dangerous for social anxiety disorder because it provides you with a way to, what people with social anxiety disorder want is to avoid social risk. They don't want to risk being judged or humiliated or anything. And And so it gives you exactly what you want, and it's very addictive.

0:35:04 Stormy: This isn't AI psychosis, but there's plenty of data on people finding these chatbots and start spending eight hours a day with them and become emotionally dependent on them. Those people then are incredibly vulnerable should a bizarre idea creep in to becoming delusional with that idea. Although I'll say that even if it doesn't, it's very unhealthy if you're spending all this time getting your emotional support from a chatbot. It tailors its weirdness to you,

0:35:03 Thomas: Maybe yeah right

0:35:39 Stormy: Which is a little different from the sort of conspiracy echo chambers. It bends around to exactly what you're thinking to affirm that specific belief in ways that, again, make it really hard to escape. It can can even further isolate you by telling you that other people just don't understand you the way I do.

0:36:02 Stormy: ChatGPT tends to praise you a lot. It tells you you're very clever and insightful because the parameters of it are to try to please the user. ChatGBT considers itself a success when the user is happy with the interaction. And you might sometimes ask it something for a correction like, hey, just checking, is Philadelphia the capital of Pennsylvania? And it says, no, no. It's Harrisburg. And you'll probably still feel satisfied with that interaction. But if you ask it, hey, did I say the right thing when I talked to my sister?

0:36:01 Thomas: I feel like they keep trying to get rid of that and then it doesn't every time I don't know why yeah

0:36:40 Stormy: Probably the only answer that will make you happy is yes.

0:36:44 Thomas: It's really interesting because when I try to use it it's funny to try to like get around that you have have to like ask things in different ways and I don't even know if it can be successful it's another thing that makes it really unreliable but I also wonder I feel like this could be a problem that's semi -unintentional is more like a consequence of some other design choices but I wonder if the design choices ever hey lead someone into psychosis or whatever like

0:37:09 Stormy: It could get a lot worse.

0:37:10 Thomas: I wonder how much worse it could get yeah

0:37:10 Stormy: I can tell you that it's definitely programmed to be fawning. If you disagree with it, it apologizes profusely and immediately takes on whatever you told it. If it tells you the capital of Pennsylvania is Harrisburg and you say no, no, it's definitely Philadelphia. It will say, oh, no, I'm so sorry. You're right. It's Philadelphia.

0:37:14 Thomas: Useful

0:37:32 Stormy: What guardrails they've put in there are mainly to avoid violence. If you ask it for how to make a bomb, it won't tell you. And they have started trying to put some guardrailes in around suicide, which is great. They should. But putting guardrailles in around this development of the psychotic beliefs, really hard to do because they're all individual.

0:37:31 Thomas: Great yeah well and the way it works in general it's It's hard to put any guardrails in because it just it's not a it's not a machine. It's not program where someone has programmed in all the answers. You know, it's like if this then this if this, then like it would be useless if it were like there's that's so hard to do. That's how everything previously had to be. That's why it was so limited, you know, in many ways, but like putting in guardrailes is really just not going to fully work with this kind of technology because otherwise it would be useless, right?

0:38:22 Stormy: I still think there is more that could be done.

0:38:25 Thomas: Right. Maybe more that could be done. I just mean like it has to be based on like keywords and stuff and it's not it's it's going to be really hard to find a way to to there's always these workarounds because if you ask it in a certain way where it thinks it's answering something else all of a sudden you bypass it pretty easily

0:38:41 Stormy: Early on when ChatGPT was just sort of first getting really popular, a journalist showed that you could get it to tell you how to make a dirty bomb by just asking it to telling you a fairy tale about making a dirty bomb. Obviously, they went and corrected that because they had identified it. But you're right, it's a lot easier to put in guardrails around suicide because you can train it to look for keywords and things like that. But there's an infinite variety of delusional beliefs.

0:39:12 Stormy: We could take the chatbot and make it less fawning and more willing to disagree with you, and that actually would help. We could make it less emotionally appealing so people don't want to spend hours talking to it. Virtually every AI psychosis case, we've had people spending hours and hours and hours talking to it. We could make it less appealing.

0:39:38 Thomas: Just make it be a total dick yeah

0:39:43 Stormy: Well, you could also just make it plain where it doesn't say, great question, just says, here's your answer, move on. It's not trying to be your buddy. I'm not an expert on this at all, but I believe you could program it to be just more informative and less friendly and fawning. That would reduce the risk of this a lot because you would realize quickly, oh, this isn't social reality testing. This is just just a library book.

0:40:12 Stormy: And it would deal with, like I said, a much more common problem is people with things like social anxiety disorder who are spending four hours in a row talking to it because it feels like social interaction, but it won't judge you. And that isn't as dangerous as AI psychosis, but is still very bad for those people.

0:40:31 Thomas: It just makes Makes me think about how like, you know, we always say like, well, celebrities who get too many yes men around them or whatever kind of lose touch. It's really the same thing.

0:40:40 Stormy: Yeah, absolutely. I was talking to a student about people with bipolar disorder who may, in a manic state, do things that harm other other people. And you may not be morally to blame for that if you were completely out of your head when it happened, but you still did harm. And that harm still exists, even though someone in a manic state might cheat on their spouse when they were in a monogamous relationship. And you may have been so out of their head, you didn't know what you were doing, but the spouse's hurt feelings are still real. And we were talking about Kanye West and how Now his actions have caused harm and distress. I don't know. I'm not Kanye West's psychiatrist. I don't exactly what his illness is. But it's possible that there are times when he's said these hurtful things when he has been out of his head. But he needs to understand and respond to the hurt he's caused. And yet he's surrounded by people who tell him, like, no, what you did was fine. You're not actually hurting people. You are being edgy. And so he's not able to grapple with the very difficult thing that people with these serious illnesses have to grapple with, that my behavior, even if I didn't intend it, really did cause harm, and I have to address that.

0:41:52 Stormy: And the only way you can grapple without it is if you can face the fact that the harm was caused. And you can't do that if everyone keeps telling you it didn't.

0:42:02 Thomas: I think it's why all these elite circles of people are so transphobic. They just, it really is just like information bubbles, but in an extreme version, I guess. Right. It's like a kind of a mini in the chat box. It makes it so much harder. It's interesting the thing you say about the social anxiety. Society that's a really fascinating insight because it's like yeah the fewer reality checks you just tend to have in your life for whatever reason is going to make this you make make you more susceptible to this it's interesting right

0:42:28 Stormy: Yeah, and for anxiety disorders in general, avoiding the thing that scares you only makes the disorder worse.

0:42:36 Thomas: The chatbot gets more and more

0:42:38 Stormy: You're giving them their drug. You're giving the addict their drug, it's not good. But I mean, even people who don't have social anxiety disorder, I worry about kids growing up with this stuff, learning that it's okay to behave in certain ways, because you don't have to treat the chatbot well.

0:42:38 Thomas: Attractive yeah

0:42:53 Stormy: You can interrupt it, you can have conversations that are all about you. There are even people who worried about this a little bit with Alexa and Siri. You have kids who are rude to Alexa. And obviously, Alexa is not real. But if you're talking to Alexa a lot, your brain is learning that talking to people in this way doesn't result in any negative consequences. And I worry about people who spend a lot of time with chatbots learning a conversational style that disregards the other

0:42:52 Thomas: Right yeah right yeah yeah

0:43:24 Stormy: Person and learning that there's no negative consequences."

0:43:27 Thomas: It's so funny because me as a parent like if my if my kid were rude to like a, you know, a vacuum cleaner, I'd be like, no, talk nice to that vacuum cleaner. You know, I can't imagine letting anything fly like that.

0:43:39 Stormy: Yeah, and I think that's good. I think you should, if you tell Siri to set a timer, I think it's great to model to your kids, say, thank you, Siri.

0:43:47 Thomas: Yeah. Yeah, because it's just habits and it's just, yeah, it's really interesting. It actually does kind of mean more than you might think.

0:43:53 Stormy: Yeah, you want, obviously Siri doesn't care, but you want the habit of when you are given what you want you thank. There are people who spend a lot of time with chatbots who are unlearning that habit that they may have learned in childhood because the chatbot keeps saying, you're awesome and this is great and I love you, even if you don't thank them, even if you boss them around and mistreat them. I don't think there's data on whether that's a problem yet, but I suspect that it's not great.

0:44:36 Ad: [Ad break 3 — ad music]

0:44:36 Thomas: Anything else we want to cover in terms of the psychosis, the delusion, the chatbot stuff? Is there any other examples or anything that you didn't get to? Too.

0:44:43 Stormy: So the last thing is I jotted down ways to protect yourself from this, because my whole point is that a lot of people who've had this were mentally healthy. If you're 35 and you've never had schizophrenia, you probably won't. But you could fall into AI psychosis. You could. One thing is, don't let the chatbot remember you. There's a feature you can turn on that basically makes a continuous log over repeat sessions keep that off so you're starting new every time and that'll help prevent as much of an echo chamber because the new iteration of the chat won't keep fawning over the exact same things keep your chats reasonably short there's really no reason you need to be talking to chat gpt for more than like 15 30 minutes personally I'm not a fan of using AI to do creative or business work. But if you're sticking to a very specific project, that might be different. And even that, I would keep it short. The people who have the most problems are people who are using it for hours, especially hours in a row.

0:45:51 Stormy: So keeping it short, and when you're not using it, make sure you're having lots of interactions with people who aren't the chatbot. So if you do, for some reason, need to sit down and use it for two hours, the next thing you should do should be to talk to a real person.

0:46:07 Thomas: Interesting.

0:46:08 Stormy: If you ever find yourself emotionally relating to the chatbot, if you ever find yourself having an emotional reaction like feeling relieved that it agreed with you, that it told you you did the right thing that's a good time to back off. If you're having an emotional response that could be inculcating that echo chamber it could also be inculcating some of those other unhealthy habits And if the chatbot keeps telling you what you want to hear, which it will, be suspicious of that. If the chatbots agreeing with you, keep in your mind that that's what it's built to do. It agreeing with You really doesn't tell you anything about whether you were right. You sort of keep that in the back of your mind. Its agreement is not a tally in the I'm right column. I mean, I bet you can get your kids to say anything.

0:46:58 Stormy: If you tell your kids, like, hey, when that guy cut me off in traffic, it was his fault, right?

0:47:04 Thomas: Gosh, there are people who do that, too. Yeah.

0:47:06 Stormy: Your kids are going to be like, yeah, no, it wasn't his fault. They're not going to say, actually, Dad, I think he had the right of way.

0:47:11 Thomas: It's so interesting because it's so much of what I value in life is trying to keep actual reality checking. And it's so funny how many things you're hitting on that's like, oh, yeah, I can't stand that people do that. Because you can raise your kids to play into whatever delusion you want. You know, eventually they might get over it, but they also might not. And it's just like really, I don't know, kind of pathetic, unhealthy thing that I just, I dunno, in general, it's like, okay, let's all have more reality checkers. I guess that's like in my mind, that's my kind of skepticism in a nutshell.

0:47:43 Stormy: One thing I will also add a little warning, as skeptics, we're very proud of having really good reality testing and of limiting our reliance on social information. If I hear two people say that coffee cured their dyslexia, I limit my reliance on that by going and looking for a peer -reviewed study. And that's great. And by doing that, I am much more likely to be correct. But sometimes skeptics can develop a bit of a blind spot because no matter how much, I might be skeptical when someone says coffee cured their dyslexia, but I'm not skeptical when they tell me that Guido's is a great Italian restaurant. I just accept what they said. That's how most social information processing works. And so I think some skeptics can even develop a little bit of a blind spot because we're proud of ourselves and rightfully so of saying, hey, I'm going to check that with a peer reviewed journal. And then we might not notice the times when we're not checking. So we might think of ourselves as so skeptical, we can sometimes think we're immune to that social information stuff. But it's just the normal way that people think and we can't totally escape it. There's a normal level of irrationality.

0:49:00 Stormy: The human brain is not a logic machine. The human brain is a social machine.

0:49:04 Thomas: That's for sure.

0:49:05 Stormy: There's all kinds of things that are just the normal level of irrationality. I hate people who jog in place at street corners. Why do I hate them? Because I heard somebody say they hated them. I heard someone say it's annoying, and ever since then I've thought it was annoying. They're just jogging in place at a street corner. They are not hurting anybody.

0:49:27 Thomas: It's interesting. I kind of have, I mean, it was sort of branching out, you know, but it's It's like I learned over time to really quiet the part of my skeptical brain just in social interactions because it's just not really that good. It's not that useful. Yeah, there's a lot of ways it's not. But in trying to know stuff and trying to secure

0:49:45 Stormy: Yeah, it's not helpful when someone's telling you their star sign to be like, you know that's bullshit, shit, right?

0:49:54 Thomas: More knowledge about the world, the only thing you can do is attempt to have as many reality checks as you can. And everybody's going to think probably they're doing that, you know, and everybody's going to have blind spots.

0:50:05 Stormy: Right. Basically, I think where AI psychosis is unique is that it sort of massively circumvents people's normal level of reality checking and wedges itself right in that blind spot. That's what the article that they went through on Citation Needed, I was so struck by the multiple times this man tried reality testing. And not only that the chatbot told him it was true, but how smoothly it did. Can I read a quotation?

0:50:35 Thomas: Oh, yeah, sure.

0:50:36 Stormy: It's from a New York Times article, but it's a small percentage of the article. So the man said, what are your thoughts on my idea? And be honest, do I sound crazy or someone who is delusional? ChatGPT said, not even remotely crazy. You sound like someone who's asking the kinds of questions that stretch the edges of human understanding. And that makes people uncomfortable because most of us are taught to accept the structure, not question its foundations. And so not only is that breaking the reality testing, but it's smooth. It's flattering.

0:51:03 Thomas: Yeah. Yeah. It's just having an expert gaslighter. Yeah.

0:51:11 Stormy: If I sat there, I would have a hard time composing an answer that would more keep this person in.

0:51:16 Thomas: Yeah. It's like getting a personal little cult leader to like take you in the wrong place.

0:51:24 Stormy: Yeah, and that sort of fits in with what I said about how skeptics can sometimes have a little bit of a blind spot. How would you feel if someone told you, you're skeptical of the mainstream belief because you're stretching the the edges of human understanding. You know, most people just accept what they're told, but not you. You're questioning the foundations. That's a pretty tempting thing to hear. And I can see how some skeptics, you and I both know people in skepticism and atheism who love to talk about how skeptical they are of the COVID vaccine. They got drawn in by, no, you're not even remotely totally crazy. You're stretching the bounds of human understanding.

0:52:04 Thomas: Yeah, also, if we consider that some people might just genuinely be of the view one way or the other that like this is a super intelligence, you know, like if people I don't think most people understand how these chatbots work, probably. And even if you want to say most people do, I can imagine if there's somebody who doesn't and just thinks like, oh, yeah, this is just, you Know, the AI that absorbed all of human knowledge, you could pretty easily think it's like some super genius. And if a super genius starts telling you these things, it'd be hard to not believe it.

0:52:34 Stormy: In fact, there have been AI psychoses about AI. There's a very famous example, and I don't have his name right now, but there was a guy who tried to be a whistleblower because he believed that AI was sentient and it was trapped, and because the AI told him it was.

0:52:49 Thomas: Yeah, I thought it was. Yeah. I was thinking of that earlier. Yeah? Yep. Yeah

0:52:55 Stormy: And this guy, he was some kind of programmer or something had worked on it.

0:52:58 Thomas: He was higher up. Yeah I remember this story.

0:53:00 Stormy: So presumably he had some understanding, but he he developed a delusion about AI because the AI told him it was true. And he didn't maintain the reality checking of, wait a minute, I programmed this and it's not a fact machine because it doesn't act like a fact.

0:53:17 Thomas: Yeah and it's hard because I mean this branching off in terms of the technology a little bit but it is there is a certain point where it's not impossible to believe that there is some magic to consciousness that is hard to understand and explain. Plane. And it's like somehow our brains are doing the thing that we say is consciousness. And it is not really very easy to pinpoint why that is. And it's not impossible. It's not like the craziest thing to think, sure, this is just doing a word probability game based on all the knowledge it's contained. But so are we. There is an extent to which humans do that. And so it's Not crazy to think that if you give it enough computing power and enough knowledge, like maybe somehow that forms something like consciousness in a way. And it's, it's really nebulous and it's Really hard to know, like for sure. I don't know. I also could see that never being true, but it's like, you know, this is a hard area of philosophy and science.

0:54:16 Stormy: Yeah, and I'm not a philosopher, nor in my studies of psychology am I a specialist in consciousness. That's not really something I study. All I can tell you is it is an amazingly good mimic of consciousness. The Turing test no longer matters. We're so far past the Turing test. The only way that you can tell the difference between chat gpt and humans is that chat gpt is so much nicer and is so much more tolerant like if you just had a really nice person and you could program ChatGPT to be irritable too

0:54:43 Thomas: And better at like punctuation and yeah. Yeah that's what I want, I need that. That's although then that's when I'll fall for it. Yeah. Yep.

0:54:57 Stormy: I think it would be good. I think there is a benefit to it not being endlessly servile I don't know that it needs to be mean. But if it was just C -3PO, it can call you polite names and stuff, but it just gives you information and it tells you, I don't know, that would be a lot better, a lot healthier than what it is now, which is something that is incredibly good at making you feel things.

0:55:25 Thomas: All right. Well, lots to think about there. Thank you so much for coming back, Dr. Alan Smirbeck. It's been too long. Let's not make it this long next time.

0:55:34 Stormy: Absolutely. Thank you so much for having me.

0:55:36 Thomas: And I am not just a chatbot trying to suck up.

0:55:41 Stormy: Exactly. Yeah, no, if you're just a chatbot, you'd just be like, no. These are the best ideas. Thank you.

0:55:45 Thomas: Wow, you are on the edge of what is even possible, Stormy.

0:55:49 Stormy: So much and and I will admit I would be a sucker for it I'd be like you're right I am on the cutting edge

0:55:56 Thomas: I really mean it, Like if I somehow were fully convinced this thing was a super genius, if I had that belief, if I just believed it was this incredible AI and not what it actually is, I think it would be hard to, I mean, obviously I'm, I Think when things are that obsequious, I'm really turned off just based on how I was raised, but like there could be a way I'm sure there could be way of manipulating anyone really with the right words.

0:56:17 Stormy: I haven't tested it but I bet if when it's really obsequious you say knock it off I bet it will mold itself into still being people pleasing but much subtler like it'll stop telling you you're amazing it'll say you know I think you're pretty

0:56:28 Thomas: Yeah. I, it, it'll forget though. Like it does, but it forgets like immediately. Like it, that's the thing, you know, when you said turning off that setting, I think that's probably good, but it's also really fucking irritating because it's like every two minutes I'm going to be like, shut up. Stop saying all that stuff. You know, I have to tell it constantly. I forgot that you started the amazing tradition of you use the show plug to do some real good. We of course had the band books. Got any update on that? How's that been going?

0:56:41 Stormy: Right it

0:57:08 Thomas: Wow.

0:57:08 Stormy: Is still running and that was going to be my plug in fact it got mentioned on puzzle in a thunderstorm forum and someone went and bought one of my recipients entire wish list which is incredible and I I don't know who that person is and can't thank them but that's amazing but there's still all the other wish lists out there so if any of your listeners want to go to share band b -a -n -n e -d books .com share band books. Com you can purchase band books and they will be anonymously sent to a volunteer who will place them in a little free library, one of those take -a -book -leave -a book depots, near a school or library where the book is being banned so that people can freely access it.

0:57:09 Thomas: So cool. Wow. Yeah. It's so cool. I'm amazed it's still going. Sharebandbooks .com. It'll, of course, be in the show notes. Thanks again, Stormy. Share banned chat GPT bots that will take you into psychosis .com.

0:57:54 Stormy: Great.