Psychologist: If this one trait applies to you, you’re vulnerable to sycophantic AI | Paul Bloom

[00:00:00]I'm Paul Bloom. I'm a professor of  psychology at University of Toronto and an ameritus professor at Yale  University. And I study human nature.

[00:00:08]I study children. I study development.  I'm particularly interested in morality, in the social emotions, in how we understand  ourselves and how we relate to other people.

[00:00:22]The promise and peril of AI companions.

[00:00:27]When people talk about loneliness, when we think  about loneliness, we think about being alone, about solitude. And that can be definitely  a source of loneliness, but a lot of us are comfortable being alone. And you could be alone  for a long period of time, but not feel lonely if you know that there's people who love you  and understand you and appreciate you kind of waiting in the wings for when you come back.  What loneliness is at its root is a lack of human connectedness. The feeling that you aren't  loved, you aren't needed, you aren't respected.

[00:00:57]This could mean you have nobody in your life,  but it could also mean that you're surrounded by people and you're not connecting to them.  Hannah Rent pointed out that the most painful form of loneliness is when you are surrounded  by people and you feel utterly isolated. And I'm interested in loneliness in part because  it's an exquisitly painful experience. It's it's one of the worst states a human could  be to feel alone, to feel disconnected.

[00:01:26]People talk about a loneliness epidemic. There's a  wonderful discussion by Derek Thompson. He argues there is no loneliness epidemic. And in fact, if  you look at the data, people were much lonelier a few years ago during CO than they are now. And  there's not much evidence people are more lonely now than they were 20, 30, 40, 50 years ago. What  we see instead is actually a solitude epidemic.

[00:01:47]So people are spending much more time alone than  they ever have been. And it makes sense if you think about the the way technology, our houses  are much better. Our TVs are much better. Used to be your average person went to the movies  multiple times a month. Now we don't go to the movies much. We watch movies at home on our  big screen TVs and our abundance of streaming uh uh channels. We used to go out to eat a lot.  Now you know a lot of our dining is alone and um we order from uh Uber Eats. So as our interior  light, the possibility for rich life indoors, inside by ourselves has gotten better, um we  have become more alone. And you could talk about whether solitude in itself is a problem.  What I will say about loneliness is regardless of whether or not it's an epidemic, regardless  of how bad it is relative to the past, it's a terrible thing and many people suffer from it. um  estimates of how many people suffer. It's hard to get a good estimate because it depends how you ask  the question, but a reasonable estimate is one in five. One in five people are awfully lonely.  Loneliness is most severe in adolescence and it's most severe in the very old. And these people  really suffer. We now live in a world with LLMs, uh, large language models, generative AI, machines  that seem to many of us to be intelligent, some say super intelligent, and more than that  seem compassionate, seem empathic. Last year, uh, there was a I was part of a team of researchers  that was led by my colleague, uh, Mickey Enslet, and we wrote a paper called in praise of  empathic AI. And in it, we summarized a lot of the data suggesting that when you compare people  interacting with people and interacting with AIS, at least for short-term interactions, the AIS  win. So the typical experiment is you'd have one person, one group of people would interact  with uh with other people, maybe over text or talking. Another group would interact with uh chat  GBT text or talking and then later on they would rate their experiences and the main finding is  that if you don't know who's a person who's AI, you don't you're not biased in any way. The  people who interact with JPT say they felt more listened to. They felt they they were talking to  somebody who's more empathic, more compassionate, they get more out of it. So, we we published this  and we said, "Look, here's a case for for empathic AI." And man, people hated this article. I've  I've written a few things people dislike, but boy, they they they we got we got slammed. And I I I  have some sympathy for the objections. We made the positive case for AI, but there's a negative  case, too. And I think a proper understanding, a proper answer to the question, what role should  these things play in our lives should look at both sides of the issue and I think deal with some very  very difficult questions. When I think of people interacting with AI companions, I worry. I'm very  sensitive to negatives. And I'm most sensitive to negatives when I think about young people, about  adolescence, maybe people in their 20s. And I see two problems and they're both deeply serious. One  is that these AIs are really good to communicate with. They are they listen. They they they never  wait for you to finish their story so they could tell their own story. They never say, "Man,  I'm busy. Get back to me later." They're always there for you. And this is going to get better and  better and better. So my first worry is that they will become addictive. that I'm going to prefer  interacting with my version of ChateBT or Claude or Gemini or whatever than to a person. And you  know, maybe this is fine if I'm 90 years old and I have no one else to talk to, but but if I'm a  teenager, I should be getting out there. I should be dealing with with with friends, developing  romantic relationships, developing attachments.

[00:05:53]And one view I have is that until these AIs get  conscious, until they become sentient, then there is something deeply wrong about preferring them to  humans. You know, human interaction is where it's at. I have two sons and uh if I was to discover  that one of them or both of them have left their longtime relationships and now they're in love  of their phone, I would say this is it's awful.

[00:06:20]What a waste of a life. So my first worry is  addiction that this will will pull us away from human interactions and I strongly believe  it's human interactions that matter the most.

[00:06:33]My second worry has to do with the nature of the  interactions with AI where the AIs are famously psychopantic. They they praise you. They suck up  to you. Nothing you will do is wrong to them. And this could be exhilarating, but it robs you of  the possibility for social learning. In general, we learn through friction. We learn by trying out  things and failing. A toddler learns to walk by falling down. A comedian learns to to to be good  on stage by bombing. A boxer learns to to block a punch by being smacked in the kisser. And in a  social interaction, we learn by having people get bored of what we have to say. We learn when we  when when we tell a joke and it falls flat. We learn when someone is annoyed at us for not being  compassionate enough. And so we learn from social feedback that basically says you're messing  up. These AIs by their very nature don't give that sort of feedback. And so this has dystopian  possibilities. What is how will a teenager grow up if he never learns not to tell boring stories  because the AI interacts with is never bored? How will somebody learn to apologize if she endlessly  deals with a machine that needs no apology? What would the world be like if the answer to the  question, "Am I the asshole?" is always a sound and reassuring, "Of course you're not the [ __ ]  Everyone else is." And I think it's a problem. I think that that social growth requires messy  often painful interactions and this is what the chat bots don't provide us. If you want to  think about the effects of these sophantic uh AIs, maybe a good analogy is billionaires or royalty  or world leaders who are surrounded by toadies and yesmen. There's a wonderful book called careless  people that's about uh tech billionaires and the picture it portrays is these are people I think  almost all all of them men who are surrounded by individuals who tell them all the time how smart  they are and how everything they say is right and the author nicely portrays the damage it does to  these people. They become insensitive to feelings of others. They become narcissistic. They develop  delusions of a sort. And you could see this, I think, also through history. The extremely  powerful often go a little bit crazy cuz they're surrounded by people who always tell them that  they're right. One of the few good things about not being very rich and very powerful is we're not  surrounded by yesmen. We're surrounded by people, you know, who call us on stuff, who don't suck  up to us, and who give us valuable feedback. So, we don't got a little bit crazy in that way.  And one of the worries about chat bots and AI is that any one of us can now become as if  we were a billionaire, as if we were a king, as if we were a president because we're now have  for the first time in our lives somebody who for whom we could do no wrong and who always tells us  that we are the smartest the best person around.

[00:09:49]So John Casio is the late uh neuroscientist who  did the most research on loneliness and kind of pioneered the science of loneliness and psychology  and he pointed out that loneliness isn't a glitch in the system. It's not some sort of signed  brain isn't working properly. Loneliness is is an expression of a need. It's like hunger or thirst  and as such is an essential part of being human.

[00:10:14]If you don't feel hungry, you don't seek out food.  You die. If you don't feel thirsty, you don't seek out liquid and you die. And loneliness is much  the same. Loneliness, among other things, serves as a motivator. And an animal like us who had no  loneliness, who wasn't motivated to make human contact, wouldn't reproduce, wouldn't survive. It  would be a dead end from an evolutionary point of view. So loneliness, although it's painful, is an  absolutely essential thing to have. I mean to put it another way um imagine a child born who could  not feel lonely. That would be a disaster. The child wouldn't seek out contact and and wouldn't  thrive as a person. So loneliness is a signal in at least two ways. The first is it's a motivator.  It gets you to talk to people. It gets you to invite friends for dinner and get you to to ask  somebody out for a date. It gets you to strike up a conversation with somebody. And if the need  for social connection is satisfied by your phone, by your laptop, that motivation will go away.  So I worry about the extent to which these AIs, these chat bots will pull us away from meaningful  connections with other people. I mean, I use them myself. I like them. I I I I use them in my  work. Sometimes it's actually interesting to talk to them. It's they they can be interesting.  They could be stimulating, but I I wouldn't want them to be such that they would they would be so  interesting and so stimulating that I don't talk to my wife anymore. I don't talk to my kids or  talk to my friends. And this is a problem that has every potential of getting worse. So, right now,  I mostly communicate through typing. Um, it could talk to them. They could do voice back and forth  conversations. The conversational skills will get more and more realistic. And then pretty soon  you're looking at somebody on a screen who looks exactly like a person who's interested in you,  very attractive person, great facial expressions, and something irresistible about that. And then to  get more science fictiony, we could get to sort of robots, you know, let alone sex bots. Um, and  whole conversation there about what to do with that. But they're going to become at the trends  continue more and more and more compelling. And so if you're listening to me and saying, "Look,  I I talk to chat GPT all the time and it doesn't do anything for me. Who cares?" Well, maybe.  And some people some people get caught up now, but but it is going to get better and better at  this and at some point you may be talking to a future version of chatbt and in some way find  it irresistible. I also think it's a problem from a social point of view. You know, I think we  could talk more to the people around us, not less.

[00:12:53]It's also a signal in that it it helps correct our  socially maladaptive, often selfish behavior. And I think we are by nature often selfish socially.  And it's social feedback that gets me to realize I got to listen to the other person. I got to  listen to them. I got to I got to maybe expand my compassion, my caring. I got to try to get  in their head and figure out what's going on with them and and and and maybe help them. And  I think we want to do that, but we also have selfish impulses that keep us from doing so. But  if you've ever talked to a chatbot, there's none of that. And this has individual implications,  but also broader social implications. We don't want to live in a world where everybody everyone's  selfish urges are satisfied and nobody really has leared to care about other people. There have  been various news reports of people who with delusions where the delusions have been encouraged  by uh by these AI models where uh somebody comes in and has a crazy view sometimes a dangerous view  involving say you know a paranoid idea that people are trying to get me and the AI says you're  absolutely right and does something which no therapist in her right mind would do and no friend  would ever do which is encourage an illusion. Now, I also know that the that the the companies  that create these AIs are worried about this.

[00:14:21]There was a famous case where uh where an AI may  well have contributed to somebody's suicide and um and they're working on the models to to stop  this from happening to establish guard rails so that the extreme delusions aren't encouraged.  And I think that that's an excellent way to go.

[00:14:38]But the broader picture is doesn't go away which  is we all have ways in which we tend to go off the rails even a little bit and having people  around us not AI but people tend to correct us and keep us more sane. If I was to tell tell you  this crazy ass idea you say that's pretty dumb but I tell Chad and he says you're so smart. Tell  me more. I I think you're on to something here and encourages me in ways that I shouldn't be  encouraged. I don't know how severe the problem is. I don't know in what how much of this AI  psychosis as it's called how often that occurs.

[00:15:14]But it's something which occurs to some degree and  I think it's something that occurs to all of us to a little bit when we interact with these things.  Was talking to a friend of mine, a a colleague, and she was describing a dispute she had with  somebody. And I'm listening to and I'm saying, you know, I feel that this other person might have  a bit of a point. Tried to say it nicely. And she she looks at me and she says, you know, quite a  bit of righteousness. I told us to chat GBT and chat GBT told me I'm 100% right. That's what the  machine does. It tells you you're right. And I think part of I think for some people, for all of  us, but for some people in particular, being told, "You're right, you're right, you're right, has a  damaging uh effect." So, I have a long-standing argument of a friend of mine, I'm not going to  out who he is, but but he feels strongly about this where when I say, "Look, this is terrible  if people ended up with AI companions instead of other people." He says, "Why?" He says, "If  these AI companions are are rich, good company, interesting, intelligent, why? You're just being  prejudiced, man. Like, you know, just what what what do we lose?" And it's a good question. It's a  question we're going to be hearing more and more, I think, as time goes by. But I like the answer  given by Oliver Burkeman. And Oliver gives uh this nice example. His example is, you know,  you're at home one day and your phone rings and you pick it up. It's an old friend. You talk  to an old friend for an hour about your life, about they show a lot of concern for you and  so on and it was a wonderful conversation.

[00:16:51]You hang up and 10 minutes later you get an  email. It's from uh it's from Meta. Meta says, "You would have received a phone call. It was  not from your friend. It was from us. You were using voice cloning software. We're using powerful  AIs to extract personal history and everything.

[00:17:06]Did you enjoy it? How did it go?" And Bergman  says, "Um, and I agree. You'd feel betrayed. you were lied to. But more than that, communicating  with a real person is enormously different from communicating with a nonsensient unconscious  machine. Even if the words are the same, even if the insights are the same, the stories  are the same, they're different because part of a relationship, part of a communication  involves the consciousness of the other person.

[00:17:41]When I'm with a friend, I like to make my friend  laugh sometimes, and I like when they're concerned about me, and I might like to surprise them. But  an AI can't find anything funny, not really. Can't uh can't be amused, can't be surprised, can't be  concerned, has no needs, can simulate all of these things, but a simulation isn't a real thing. Now  throughout this I've always had this qualification which is that these machines are nonsens  non-concious there's nothing at home maybe one day that will change and maybe one day AIs will  be fully conscious beings and then that changes everything then there'll be people and then the  world will contain many many more people and then what I say might not apply then things will be  very different but until then these interactions with these chat bots I think inevitably fall far  far short of interactions with a person. So I'm sensitive to the downsides of AI and I I worry  deeply about them, but I don't think we should forget the upsides. So one can maybe draw an  analogy here with boredom and maybe the end of boredom. Not so late at night the TV stations  would go off and then if you're by yourself, what are you going to do? You read a book. Kind of kind  of boring. sitting on a bus, sitting on a train, you you're bored a lot at a time. And now we're  almost never bored. We got our phones. Our phones, you know, we play games on our phones. We listen  to podcasts. On my way here, I listen to a to an exciting podcast while playing uh the New York  Times spelling bee. And I wasn't bored for a second. And this is a good thing. Boredom is is  aversive. Boredom is unpleasant. Now, it's also a bad thing. Boredom serves certain functions.  It could be we get so entranced by our phones we don't we're not motivated to find more stimulating  intellectual things on analogy with how having our loneliness satisfied um might keep us away from  contact reaching out to people but loneliness is a lot worse than boredom. Loneliness is a form of  agony and there are many people in the world who suffer from loneliness. to deny them the capacity  to interact with something that can make the loneliness go away I see as a form of cruelty be  like you know people having pets and getting great meaning from their pets and you take them away  you say that's not enough that's not rich enough the people I think would benefit most from these  interventions from empathic machines are the very old there are a lot of old people around many of  them are institutionalized Many of them spend days and days and days without any meaningful human  contact. It is nice to imagine these old people constantly being surrounded by doing grandchildren  and and loving children. This a fantasy. Many of them have nobody, no family. And there's not  enough money in the world to pay people to give them to talk to them for hours per day. Also,  many of them suffer from various forms of dementia that makes them difficult to talk to, difficult to  communicate with. If there could be a machine that could end their loneliness, if if if it turns out  that that these chatbots do anything to to provide them with company to make this pain go away,  that'd be a godsend. If you want an analogy, think about powerful painkillers like fentinyl.  These have their dangers. They're addictive.

[00:21:25]they they you could overdose on them or so on, I  wouldn't just give them out to everybody. But if you showed me somebody who was in irredeemable  pain and maybe they're they're in their 90s, I'd say, "Forget about worries about addiction. Pump  them up. Make their pain go away." And that's what I think about Chad. But it's not just old people.  When I was 12, my mother died. And this was a wrenching event for me. And it led to a period  of great sadness and then a period of social disconnection and then a period of loneliness that  lasted quite a long time. If there was a machine available to me that would make me feel loved,  cared for, understood, that I could talk to, that would listen to me, that would make my days  more full and more social. It would have made me so much happier and so mindful of the problems,  the worries about addiction, the worries about about not becoming a full human. One way to think  about people versus AIS is uh was expressed by my uh by my friend and colleague Molly Crockett who  described these chat bots as emotional soilent.

[00:22:43]I like the analogy because Soylent is not bad. I  actually I like the chocolate one. Um, it is it is nutritious and filling and so on. And you could  you could be sustained by it, but it's not food.

[00:22:56]It lacks the richness and and and and [snorts]  flavor and texture of real food. And a life that's purely living on soilent on a meal replacement  drink would be a kind of really lame life compared to one where you're having actual meals. This  analogy suggests a place for these AI companions, sort of a backup, a backup way to sustain  ourselves when the real stuff isn't around, but by no means a perfect substitute for the real  stuff. I'd like to think that there'd be some some solution here where we could ease the pain of  loneliness and at the same time allow people the opportunity to develop and become more human and  ultimately outgrow the need for these machines.

[00:23:49]Want to support the channel? Join the Big Think members community where you  get access to videos early ad