Psychological Targeting & AI with Sandra Matz

Hello and welcome to the Behavioral Design Podcast. This season we're diving into the intersection of behavioral science and AI. We want to make sense of the state of AI, from understanding how humans interact with intelligent systems to using AI to do behavioral design itself. I'm Aline Holsworth, a health tech advisor specializing in AI and product design. Over the past 15 years, I've been crafting human centered products with behavioral science

at the core. At Apple, I LED Behavioral Science for Health AI, designing and launching AI powered features to help users reach their health goals. And I'm Samuel Sultzer, your second Co host. I'm a behavioral strategist specializing in hybrid formation and designing products that drive long term baby change. I work with leading tech organizations integrating AI to scale behavioral design for good.

And I'm also the founder of Baby Bites, a dedicated community on behavioral science and AI. Quick word on Nuance Behavior where we help organizations build impactful digital products using behavioral design. We only take on a few clients at a time to ensure the highest level of quality for our tailored evidence based solutions. If you'd like to become one of our special projects, e-mail us at hello@nuancebehavior.com or we could call directly on our

website, nuancebehavior.com. Hello, Sam. Hey, happy New Year. Oh, happy New Year. It's good to see you. Likewise, how are you doing? Yeah, pretty good. You're pretty good. Your your energy level seems a little low right now. I might put it about 23% maybe. It's not that your energy level seems low right now, but it's just a little bit low in general. Would you say that you don't like this? Have you had access to my aura

scores or like what's going on? Yeah, yeah, yeah. No, I'm ready to fill in our listeners. Both Sam and I have taken Sandra Matz's personality test based on the Big 5. So she included a a personality test to, to take in her book Mind Masters. And we decided to play along and really uncover our, our, our inner, the inner workings of our minds and compare ourselves to each other. And so I'm referring to, you know, the, the extroversion score, we actually scored fairly

similarly overall. But if you if you kind of dive down into the, the details of the the sub traits, you can see that the energy level of Sam and myself is quite different. That was very surprising that I had so low energy. I'm not sure I believe it. I think I might have some beef with with a few of our results, but like for example, agreeableness, we we both scored pretty low on agreeableness, though I scored extremely low on agreeableness. I don't think I'm quite that bad.

I have, I got a 31% in agreeableness. What? What do you think about that? Yeah, I know. I, I think it measures up pretty well. I think you're for an American, you're pretty unagreeable. And I would say, wow, Yeah, in general, I do you actually have less of AB problem than you do with the test scores. I think they were pretty accurate in terms of what I expected in terms of what I'd taken previous Big 5 personality test before. It kind of tracked pretty closely with that.

And then comparing ourselves, we can maybe go quickly like they were in open mindedness and consciousness. We were basically the same. Yeah, both pretty high. Yeah, I agree. We would probably like to think they were quite open minded and probably have data to to point to that kind of direction. Consciousness, I will feel that you are probably more conscious than I am if I had to predict. But it's in like I have really high, OK, I have high

organization, you have very low. What does that mean for you? That's just not true. That's just totally a lie. I'm so rigid in my organization. I don't know, like I'm I'm not sure what happened there. I'm too organized reverse for me, and sometimes I feel like you're the one who's always a little more prepared. And yeah, then we'll come to extroversion. You scored super high, like really high, like almost in the 90th percentile. And do you feel like that's true?

Like do do you feel like I tracked with your experience? No, no, not at all. In fact, I feel like there are a few things about my personality that are, are just like completely the I don't feel like an extrovert at all. I don't identify with being an extrovert. So when I see something like that, I'm, I'm just, you know, kind of bewildered. Yeah. I don't feel like a social person.

I I'm fearful of small talk. I think that I don't have an an external monologue like I see my son, you know, like everything that he thinks comes out of his mouth. And I like, I don't have that at all. Like my thoughts mostly remain inside my head and I don't like speaking out loud very much so. So I feel like a lot of the qualities that I would describe myself as are just the opposite of extroverted. I think it's the assertiveness that is like pushing me. There is I can be.

I see something that I think like someone needs to be stood up for or like I see something wrong with the social order and I'm very, very assertive and I don't know, maybe that's like what's pulling that? It's not that I'm fun and social. I think you're very fun and social, but I do think it's interesting.

I have a theory when it comes to extroversion and introversion because I feel like so many people, especially people that are public figures, we're like, we're this kind of introversion as a proud batch of honor on their like they go on interviews to do various things and they start like, oh, you know, I'm such an introvert, Like, oh, I'm such an introvert. I heard recently from, I think it was Neil Ferguson, I think

the historian, he said that. And I do think there's something around where we have a little bit of a skewed perception of what it means to be social. But an extrovert and introvert, I think there are certainly kind of people that are maybe what we the finest extroverts, but those are like very close to like the 90th, 95th percentile. I think like we are generally as human beings, socially uncomfortable in many situations, not confident, awkward, all of these things.

I'm a homebody too, like I I don't go out. I spend most of my time at home. And I think a lot of people do that honestly. But then there's like a few like the 510% of people that are actually like on the top extroversion, they are almost casted as like the top 50% of sort like they're kind of. And so I think that's basically we have a little bit skewed sense of what it means to be introverted and extroverted and I think.

That's fair. So. Yeah. I I also think that with all the public speaking and podcasting and such and having to like really be in front of an audience and, and be put on the spot and all of that, I think become more extroverted over time. Yet my identity has not necessarily shifted along with that change. I still am kind of holding on to what I the the alien that I think of myself from 1015 years ago or so. Yeah. And I think that's very common. And we go back to then agreeableness.

I was kind of average. You were. Yeah, on the disagreeable side. And when it comes to neuroticism or negative emotionality as is defined here, I scored quite low. That has been. Insanely low, like sociopath low. I didn't even know you could score. That do I even have a pulse? Yeah, 6% neuroticism. Is that like you're just a beam of sunshine? Well, isn't it just an absence of strong emotional kind of

reactions? It's not so much that I have the sunshine always like I, I, I think that's the maybe the misconceptions with this stuff is that I see. You're right, you're right. I think I'm very unlikely to have kind of emotional outbursts for good or bad in some ways. And I think that's, yeah, that's probably part of that, yeah. I cannot identify with that. I do feel anxious and

emotionally volatile at times. Yeah, well, you were in the closest to the 50%, so you're like, you're part of the big masses of most people. But yeah, it was fun to do this and we did it for a reason. We did it to do a little bit of

an experiment. We've been in times of, you know, gift giving times. Christmas, I just passed my birthday is coming up, wink, wink. So, so yeah, we, we thought we would do a little bit of an experiment. Yeah, that's right. So in the spirit of Sandra's psychological targeting and using AI in order to do this, we decided to kind of put ourselves up against the machines and compare our skill at gift giving, especially to someone that we know fairly well to Chat

GPTS gift giving. And so so we took our personality results, we fed it to the machine and we said, hey, ChatGPT imagine that you are an amazing gift giver, renowned across the lands for your talent at giving the optimal gift for each person. So it's really important that you find this this single most perfect gift for a person based on, you know, their unique personality profile, sort of person psychological targeting at its best. And so we asked ChatGPT to pick out this optimal gift.

When I asked ChatGPT what to get you Sam feeding in, feeding it into the machine. Well, I won't tell you my reaction. I will just tell you what it is and you can give me your reaction. And don't, don't. This isn't like you know, when, when you're opening an actual present from someone where you have to hide your, your, your displeasure. You can tell me exactly how you feel. The. Percent. And what do you say, Like 00 percent agreeableness, like

turning down the dial? Yeah. Exactly Exactly 100% honesty OK, so and the machine said the single most perfect gift for Sam is a membership to a creative Co working or makerspace that offers workshops in various crafts design technologies. Or creative pursuits. OK, that's a little bit creepy. That's a little bit creepy. Guess what? Guess what I just bought myself 2 days ago? A workshop at a local kind of crafts to like how to make your own wooden furniture.

No way. You got it for yourself. You should have let me get it for you. Oh my gosh. Yeah. Wow. So you like? It that is kind of, yeah, creepy, But that is a great gift. It's literally what I was thinking about over Christmas or in the New Year's. Like, OK, what is the one thing I want to do this year? I want to do a little more building with wood, wood, wood stuff. That's awesome. Well, I did actually ask a follow up question.

I asked, OK, can you, can you give me a specific makerspace? And I also, sorry, I gave it your address. I said make it conveniently located so that Sam can get to the space and it found one. It's called the very special name of Stockholm Makerspace. So I don't know if there's a better one or if the one you found is more interesting or, or better suited to you. But yeah, if you go to makerspace dot SE, check it out. I've already been there. I was there three days ago on that website and.

Yeah. It was my topic, but it's a little bit further away from my home, so I show us another one that was a little bit closer by, but that was the place I wanted to go and and get membership for you. So wow. OK. That is actually like really really. I'm starting to feel threatened by Chat GPT's success right here. I'm like, I don't know. I did the same thing for you, obviously. So we'll see what this, if it had the same impact. It's a very specific 1. So I will be very surprised if

if you had this same reaction. I have this kind of output written as a kind of a ad. So I'll read that out for you, Eileen. Elevate your library with a personalized X Libra stamp. A touch, a touch of whimsy and not of modern design. And a whole lot of you actually, every book that you love for the discerning reader who values creativity, organization, and a personal connection to their collection. Because a great book is forever.

And so your mark should be. I this is for stamping my name in all of my books so that if someone steals. It like this personalized etching thing that you can kind of like stamp, I mean, and a whole lot of you your name onto every book. I love it, you know, my poor books. I don't know, I'm like, I feel, I feel a little bit torn because I'm so like, you know, you don't deface the the books, but I guess this is allowed. But it's beautiful and it's on the inside of the book.

So it's actually only when you open it, right? Yeah, Yeah. And it's something you can also then when your your kids eventually, obviously they're going to read the books that you loved in the same way as you love them and they will say mommy's name. Yeah, I should have the date in there as well so they know when I read it. Yeah, you. Can place it into context. OK, so now that we've shared our ChatGPT gifts, I'm excited to share my the gift that I made and chose for you. Are you ready?

Yeah. OK, so this this took it's a little bit elaborate. It took a lot of thought and passion on my part to put together. So my gift to you is a semi guided world tour, tour of the world for you and your fiance to take together. So yeah, an entire month of curated travel experiences, all expenses paid because you know, why not? And it's going to be filled with surprises all along the way. So let me just let me outline some of your schedules. So it's it's only partially scheduled.

I want to make sure that you have some, you know, you know, some, some chances for whimsy, let's say. So it's so you don't have an overly packed itinerary. You can have some spontaneity. Yeah, But I've also scheduled some meetups with my friends in each place, and I might drop in periodically along the way. My, my Swiss family is gonna take you on a Steamboat tour of Lake Tune. Why? My Mexican friend's gonna make you poblano mole in Oaxaca and occasionally you're gonna have a

surprise activity plan. So you might just come across 2 bikes and someone to guide you through the countryside. So just like biking from 1 vineyard to another, you don't even know what's coming. You're just gonna say, oh, what are these bikes for these? They've just shown up in my path. You may have to see a movie that I recommend in a historic theatre. So it's a movie that's that's very special to me. I can't tell you what it is not yet.

You may have to strip down in a Turkish bath house while you're in Turkey. You may need to participate in a dance competition in Brazil. And to make you feel good about all of this excess and the luxury that you're experiencing, I've planted some volunteer activities. So. I thought you were going to say a donation to give well, and that I'd actually have to do some work. I have to actually volunteer.

You do, but it's fun work. It's it's stuff like helping children build Lego houses, teaching martial arts class. I think I think you're going to really enjoy the gift of giving. Wow, that is one heck of a gift. We have our wedding and honeymoon probably coming up this year. We haven't fully decided, but the plan is for that to happen this year. When we're looking at like, what would the honeymoon potential look like? And now you know. Sounds like.

No, you've got it planned. I was thinking about that. And paid, I see. And paid, yeah. So. Happy honeymoon. Thank you. Thank you. It is funny because I took a little bit different approach to this. Like I misunderstood the direction and actually bought you gifts. So like I think I think literally tomorrow or the day after you should get 2 gifts and one one is from which you now. Know in this way I actually get to keep the stamp.

You get to keep the stamp and you also get a really silly cute calendar with great gratitude and pun related stuff in it So. Thank you, I love it. I'm sorry I didn't actually get you anything. Well, I, I think it's funny, like in terms of if you got me, that would be really cool. But I felt in terms of, you know, speaking about negative emotionality or neuroticism, I had a peak in neuroticism. Like trying to figure out like what to actually give you.

Like I was feeling really on the spot. That was your 3%. That was my Yeah. I don't know. You peaked at 3%. I peaked because it was tricky with these things. Like it's feels like you're upset so much about like how well you know the other person, how to encompass some person into some kind of small gift of sort. It was tricky. It was really hard. It was high. Pressure.

I totally agree and have it make sure that it's personalized shows that I know some things about you and that it's not too generic. And I love that yours both like has a nod to my sense of humor and love of language. And also it's sort of a reminder. So when I see it every day, I'll think, oh, I'll think of Sam. And it's like, there's a lot of, I don't know, behavioral science lessons and baked into both of our gifts. Yeah, but this leads us into the episode, which was really fun

and probably. Yeah, I've said it before, one of my favorites it's it was really interesting episode getting into some, I think when it comes to AI, really relevant and interesting stuff. So yeah. Do you want to introduce our guest? Yeah, love to. So our guest, Sandra Matz, I've

been a big fan of for a very long time. She's a professor at Columbia and computational social scientists. So you may not have come across many computational social scientists. She's pretty special in that regard. And she's she's known for her work on psychological targeting. So we talk a ton about psychological targeting in this episode, essentially how big data can be used to create a unique profile of individuals and then leverage that profile.

To influence some persons behavior, the person that the profile is made of of course. And so this is often also highlighting our focus on AI this season, often using AI methods to do both of these parts, the creating of the profile and then tapping into that. And Sandra just had an excellent new book come out called Mind Masters. And I don't have any doubts you're going to be running to get it. Put your Put your personalized stamp into it after you've

listened to our conversation. So consider this our personalized gift to you, our dear listener, as we continue to explore AI this year. So let's get into this conversation with Sandra. What happens? To. Murgatroid. Sandra, welcome. Well, hi, thanks for having me. It's so good to have you. You are the expert on psychological targeting. Would you agree with that? I'd hope so. I mean, at least I've spent the last 10 plus years studying it, so it would be nice to be called

an expert in that space, yeah. We think so.

I'm going to go ahead and say for sure you are and in one of your papers you have this very nice diagram where you really outline what is psychological targeting. You know, how does it come about? What are the steps from creating a psychological profile to then using that profile to create an intervention or or some sort of messaging campaign using the profile to target people, whether they're users, customers and so on? Can you tell us a little more about what goes into these two

steps? Like like give us a breakdown of psychological targeting? Yeah, more than happy to. And I'm impressed by the knowledge of the framework.

But yeah, for me, that the way that I think about like this ability of computers to really, what I think of it, like penetrating our mind and our psychology is both reading and writing for the computer nerds, it's like trying to understand what's going on inside our mind, but then also writing into our mind and potentially changing the way that we think, feel and

behave. So psychological targeting, the way that I think about it and the way that I describe it essentially unfolds in these two steps. Step #1 is that based on all of the data that we generate? And that can be really anything from the more explicit identity claims that we post on social media, right? Like what's our last vacation? The pictures that we share?

All the way to. These more subtle and implicit behavioral residues, all of the traces that we leave without really thinking could be anything from the GPS records at your phone captures to like your credit card swipes or your Google searches. We don't typically think about it twice when we create those traces. And all of these behaviors can be used by algorithms and AI, machine learning to translate what you do online into who you are. And I'm happy to go and go into

this. I'm a little bit more in a little bit more detail later because I think it's really not not rocket science. So often times we think of like AI as like this black box and just like magic that happens inside. Not true. So I'm happy to give you some examples later, but that's really step one is we translate what you do online into who you are on a more psychological level. And I think that's often times where almost like the the public

narrative stops. But we talk about, well, to what extent can algorithms understand who we are? To what extent is that an invasion of our privacy? For me, the second part is almost more interesting. The second part is, OK, now once I have an understanding of who you are, what your motivations might be, what your preferences are, your fears, hopes, dreams, whatever you kind of name it, to what extent is that that insight allow me to potentially change

how you think, feeling behaves. So that's the second critical part of psychological targeting. It's not just understanding, it's really trying to push you in a certain direction that either you might want to go in anyway and I'm just helping you get there, or I'm going to try to push you in a direction that you otherwise wouldn't have wanted to go in. And I think that's the slightly more sketchy part that we probably going to get into. Yeah, and no doubt.

Do you have any like go to examples that you kind of explain this through in terms of how this can look like in the real world? Yeah. So let me also again break it down into these two parts. So I think the first part of to what extent can machines read what's going on inside your mind? What I love about this work is that when you open the black box and you look into what this machine learning models are doing, and a lot of the relationships are actually

pretty intuitive. And so if you're more extroverted, for example, you talk about weekends, parties more on social, you often times use all caps because you're shouting everything that that you say. You spend more money on going to bars and pubs. So I think those relationships in a way are pretty intuitive, right? And then if you look at introversion, that would be people talking about reading, talking about the last fantasy novels that they've read or

role-playing. So as you can see, like the the stuff that introverts talk about is like much more reserved and quiet. So for me, that just means that there's really good face validity. What algorithms are picking up on often times isn't magic, right? It's all of the traces that we leave and they make sense. But there's also some that are a

lot more surprising. And I think that's in a way the contribution of algorithms to the predictions that we can make, but also to our understanding of psychology, right? If we just learn that extroverts go to parties more, well, yeah, we probably knew that before, but one of the examples. That I like. That is somewhat not as intuitive is the use of first

person pronouns, for example. So I me myself, I remember that Jamie Pennebaker is one of the researchers that studies the use of language in the context of psychology. And he present or he posed a question at a conference of psychologists. So like a room for the psychologists. And he was like, well, what do you think? Like the use of first person pronouns is predictive of. And I think all of us immediately jumped to narcissism.

So I think all of us were like, it's got to be narcissism if you kind of you want to talk. About yourself, you just want all the. Attention, it's got to be narcissism. And then he totally flipped the script and he's like, well, it's not narcissism, it's actually emotional distress. So if you talk more about yourself, it's a sign that you might be, for example, suffering

from depression. And in a way, it sounds counterintuitive in the beginning, but then when you start thinking about it a little bit more, might think to the last time that you felt really depressed and down, you were probably not thinking about how to solve the world. 'S. Problems, right? Because like what you're focused on is on the self. You're like, OK, what's wrong with me? How am I going to get better? Why am I not going to get better? What do I have to do to get

better? So all of this inner monologue that we kind of go through then just creeps in the language that we use to talk to others and to talk to the world. So for me, those relationships are really interesting because we might not have ever discovered them without the use of algorithm. So that's really kind of part number one examples of how do you get from the stuff that's on the traces that we leave in the digital world to who we are on a

more psychological level. Step #2 and this is really what I think the bulk of my work has been focused on in the last 10 years is the So what question, like what does it mean? Like how can I use those insights to influence your behavior? And we started on relatively, I would say neutral grounds if you want, in the marketing space.

So essentially the, the question of if I understand, for example, that you're more extroverted or more introverted, can I use those insights to get you to buy certain products? So one of the the very first studies that we did was we teamed up with them with a beauty retailer in the UK. It was like an online store and their goal was simply to get women to click on an ad, go to the website and buy something.

It's a pretty simple goal. And the strategy that we used was essentially saying, well, if we understand based on someone's digital footprint that some of the women are more extroverted, some of the women are more introverted, can we simply change the way that we talk about the product, Right. So we're not necessarily picking certain products that we think are more extroverted or introverted. We're just changing the way that we describe the beauty retailer.

So for extroverts and the creative team came out with something like dance, like no one's watching, but they totally are playing with the need of extroverts to be the center of attention. So it's like people on the dance floor. It was like saturated. Colored a lot was happening and then for introverts it was like one woman in front of the mirror and much more reserved and quiet. Like beauty doesn't have to

shout. So the idea was that if we target extroverted women with these extroverted messages that tap into the needs of extroverts, they would purchase more often than And that was exactly what we found. And it. Actually like a pretty.

Substantial effect. So I think back in the day, it was about like a 50% increase in the number of purchases, which directly translated into how much money the beauty retailer made, which is like a again, like obviously you're not changing a die hard iPhone user into an Android, but you can certainly change there's. Nothing that can do that. Yeah, exactly. There's nothing I can. Do that there's. Maybe the other way, but it. Certainly means that you can make certain behaviors more

likely than others, right? And I think that's. A pretty powerful tool and you can imagine that I'm just talking about products right now and marketing. But maybe we could use it to change your opinion about political candidates. Maybe we can use it to actually forgot to get you to save more instead of spending more. So in a way. It's like this tool that once I tap into your psychology, I just. And understand what you want and what you desire. I can nudge you in a certain direction.

That's really cool. And I guess from my point of view, there's this kind of like 3 examples that I've come across throughout the years in terms of like people maybe talking about your research kind of without knowing about your research. So I think way back when, I think the first example was people talking about Minority Reports where there's this film with Tom Cruise. He walking around and you know, he sees ads that are basically like personalized for him as he

was walking on the street. So he's used an ad and it changes based on kind of who's looking at the ad. And then you have kind of a more somewhat more recent example by having more of a one that became quite well known was this example of some household start to receive some form of products related to pregnancy. And then it came out that, you know, the daughter of the household was pregnant but haven't told the the. Parents and so on.

And it was some form of, you know, it's with Walmart or, or Target or. Target the infamous target. Example the infamous target example, right? And and then maybe like the the one example that I don't know if exactly it mirrors what and I get to hear if you agree, but it's more recent example is with with TikTok, because it's so salient for people that you know how to interact with something, it becomes very kind of mirrored in their interaction.

So like they they watch certain things and so on explicitly. And then it kind of like leads to certain things, but also implicitly like some things. Like people talk about how TikTok helped them realize they were bisexual. They didn't know it. But then they spend like that was a big John Oliver segment actually this week on TikTok where someone claimed to have like, I didn't know that I like girls. But then I started TikTok and I

I realized that. So yeah, those 3 examples, does any of those are more interesting to you or does any of them kind of familiar, especially for you? Well, first of all, I love that you totally skipped over Cambridge Analytica, which I think is the. Example that I that I typically get but it's not. Surprising because I got my PhD at Cambridge on the on the topic of psychological targeting. So I get mixed up with Cambridge Analytica all the time and it was actually in some way helpful

for us because. We've been talking about this potential tool that has the ability to change people's behavior. For a long some time, trying to warn policy makers of like, here's something that you might want to look into. Nobody cared until Cambridge Analytica happened and suddenly people did wake up to the reality that maybe there's a tool that we should be considering and looking into a little bit more.

So that's just the example I like because it really opened up people's minds to this ability of algorithms to penetrate our psychology. Now, all of the other ones I think are great examples. So the Minority Report in a way I like because it was the first one that painted this this future, but it felt like science fiction, like it's never going to happen. And yet it's exactly where we are right now. There's this funny quote that I really like, which is the difference between science and

science fiction is timing. And that feels like one of these examples. Right, so it used to. Be science fiction right now is just science, and it's. It kind of speaks to a point that I like to make. Sometimes when I give talks is usually someone in the audience was like. Well, I'm so excited that I'm not on Facebook or any of the other social media platforms

because. Now nobody can track me and nobody can use these technologies and psychological targeting to influence me and I it brings me so much joy. Do you have a phone? To tell people like you know exactly, are you breathing? Are you kind of spending money with your credit card? You have your smartphone that's with your 24/7, but there's 50 million cameras in the US. They kept you pretty much everywhere that you go there's unless you live. Out there in the.

Woods and grow your own vegetables. There's just no way that you can escape it, and I think Minority Report in a way shows. This right, it's like not that he kind of. Post something on social media and someone infers that you're extroverted. Yeah, he just walks. By you get like your record faces recognized. He connects it to everything that they know about you just by looking at the data traces that you leave unintentionally. And I think that's the world

that we live in today. So that's why I like like this example in particular and the Tik. T.O.K. Example. There's not as much research on the assumption that algorithms can know you better than you know yourself. But the interesting part here is I think that algorithms just see a lot more, right? So what did? The power of AI comes from the fact that they observe your behavior, but they also observe the behavior of.

Millions of people so there's certain patterns that they couldn't detect is like these digital Sherlock Holmes's and that you might just not have access to and because there's all these social pressures to. Conform in a certain way. You just might not be top of mind, but an algorithm just observes everybody and now you can see OK based on these kind of more subtle cues. Maybe that's something that I recommend and maybe that makes you wake up to a reality that

you didn't even see yourself. So I think it's an interesting one that we don't fully understand yet, but I think we're getting there. If you'll allow me, I want to do

a quick sidebar on Cambridge Analytica. We actually, we talked about this in our episode with Gordon Pennycook when we were talking about misinformation. So totally different context. And can you help us set the record straight? This also applies to a general, you know, question we have in terms of impact. What is the impact of psychological targeting in general? You know, some in some cases quite powerful, in other cases maybe less.

So for the individual, what is your take on how much of an impact the targeting that Cambridge Analytica did? How much did that contribute? To. Donald Trump being. Elected. I mean, there's like a few thoughts that I have on on Cambridge Analytica. The first one, I mean, it's not impossible to say what actual impact it had on the election. I think for me, that was never the question. I was actually really annoyed with the news coverage of like, well, did they do it? Did they not do it?

I'm like, I don't care. I know that the technology exists. So even if they didn't do it the next around, someone is going to use it, right? And if we believe, and I think this is getting to your question, do I think that again, we can convince a die hard Hillary supporter to suddenly vote for Trump? It's probably not, right? This is just we're not changing

people's core identity. We're not kind of flipping you around and turn you into a person that's someone totally different just by showing you a few personalized ads on Facebook. That's not how the world works. But most of the time we don't have to do that. But most of the people are either undecided on who to vote for or they're kind of considering whether to go out on Election Day at all to, to cast their votes.

So I think in that election is like 40% of people didn't even bother to go to the ballots and then cast their votes. So can we probably nudge them to either change their opinions on the candidates or decide to stay at home, which is what what the Cambridge Analytica campaign was trying to do? That's probably true. And if you look at the elections, most of them are won by really small margins.

So I don't think it's like this magical warfare tool that it was made out to be. But I do think that it could potentially have an impact even on something as big as elections, just because that the margins of error of which side wins are just so, so small. So that's one thing. The other one that I think is really important in that the coverage was it was always made out to be like this data leak.

So I think that the, the public narrative was like, Oh my God, someone hacked into Facebook, they got all of your data and that was like this tragedy that wasn't true, that the way that the data was acquired was perfect. Well, partially, and I'm going to explain in a second why it

wasn't. But back in the day, because Facebook was trying to connect everybody, like the moment that I consented to have you access my profile, you also through the Graph API got access to the profiles of all of my friends. So Cambridge Analytica didn't have to kind of ask every single person legally to get their Facebook profile. They could access that through my consent and through my profile. Now, the problem with Cambridge Analytica was that they collected the data for different

purpose, right? It was an academic collecting it and then selling it on to Cambridge Analytica. So that was the illegal part. But the initial way that the data was collected was actually in line with the terms and conditions of Facebook. And I think that's something that we need to understand. It often times the way that our data is being captured isn't through a security breach or a

data leak, right? Often times we think, oh, someone is going to hack into our e-mail or it's going to get our data. Most of the time we just sign away our data by accepting the terms and conditions so that you download a weather app on your phone and you happily accept what the app to tap into your microphone cause. The gallery, the terms and conditions. Nobody's reading it exactly like nobody has the time. It feels like a leak because you didn't know that was possible,

but it wasn't actually. It's just poor design of privacy and security in the 1st place, right? Exactly. And I think that makes a big

difference. Yeah, cuz you can't really go out now legally and prosecute those companies because essentially you consented to it. So I think maybe we're going to get into this later, but I think we just have to design the redesign the entire process right now we're just saying, well, we're going to tell you a little bit about about which data we're collecting, what we're doing with it, and then you take control over how to manage it. It's just impossible. Right.

So even if you understand exactly what's happening, what data they're taking, you don't have the time to manage your own data across all of the services and products that you're using. It's just impossible. If there's no default level of of privacy protection that is is kind of at a good level, it's just not going to happen. Yeah. OK, Tangent over on Cambridge Analytica.

Yeah. But a very good tangent and I think, you know, I'm obviously we're super tempted to jump into talking about AI. This isn't. But actually I want to continue to lay a little bit of foundation because I think it's interesting given that we're now introduced kind of what you mentioned before about it's kind

of like call it 2 main steps. It would be interesting to look at that a little more in detail because I think especially when it comes to, you know, citing another maybe example that many people have seen is this film that came out maybe two years ago, Social Dilemma was it out maybe two years ago? And they have this very strong visual kind of metaphor where

they're showing this main. Protagonist of sort and you have this one with social media company that like through his interacting in the world, it'd be like a more and more of a one to one kind of representation of him in their system. So initially it's like very kind of not really detailed and as the film progresses, like it becomes a very kind of 1 to one representation of of him, but it's still kind of interesting in terms of like, well, what does that actually mean?

Like what if we wanted to build representations of people? What are the things that we might look at? And I think you cited the idea of simple, you know, extroversion versus introversion, which is a kind of a some form of personality trait. And I guess it'll be interesting to talk a little bit more about

that. You know, you could imagine that the type of things that could be included could be everything from I guess people's kind of preferences, maybe both kind of reveal and stated it could be as we're talking about personality, maybe some other things. I just want to see what would you say is kind of the main thing that is being used at the kind of the first stage of like creating that picture? Like what are the main things that usually kind of part of

that equation? That's a great. Question and and the extroversion versus introversion was a gross oversimplification, I should say. So I'm. A personality psychologist. And and psychometrician by trainings that would have get me kicked out of the Society of that. But I think there's a number of different steps. So one is, I think like any social scientist or behavioral scientist would say, well, why don't you just go from behavior, past behavior to future behavior, right?

If you can observe someone's entire behavioral history, why don't you just use that to make predictions about what they're going to do next? And that's totally true, right? I think we don't necessarily always need the intermediary step of saying we were translating your behavioral profile into something like the Big 5 personality traits, your moral foundations, your values, whatever it is.

Now, I do think that's actually sometimes it's helpful because it gives us a human understanding, right? And a lot of the interactions that we have, especially in the second step when we're trying to influence behavior is still driven by humans. So if I'm a marketing creative, I might take the insights into who someone is on again, personality level to create some messages. Now we're going to talk about AI in a second.

So some of this is actually being automated now by by these large language models, but there's still something that is really meaningful, especially if we want to bridge this kind of experience between offline and online world to have an understanding that humans can make sense of. Now, I think you mentioned personality traits, moral

foundations, preferences. So the extent to which these psychological traits are relevant and helpful when we think about catering content actually very much depends on the context, right? So the big 5 are often times helpful because they're so broad, like anything from music preferences to vocational interest. They kind of explain some of the variance. They're also really crude, right?

Like if I think of like personalities, like you meet a person in the street, you know nothing about them. Knowing that they're extroverted is actually helpful because it gives you at least a sense of who they are. The moment that you get to know them a lot better, you throw the big 5 out the window because you don't think about your spouse is like, Oh my God, he's so extra. You know them so much better and on every little detail that you don't need this anymore.

So I think this is one thing, but you can also imagine that other contexts, for example, in think about health, I don't think the big 5 are necessarily the main driver you could think of, well, if I wanna change behaviors, maybe I'm gonna go for regulatory focus. Are people more motivated by prevention or by promotion? So I think this very much depends what you're trying to measure on what you're trying to accomplish in terms of behavior change.

But even that I still think assumes that we like the static version of ourselves, right? I think a lot of personality psychology initially was let's just capture someone's dispositions. Let's just see what are the stable dimensions that underline our behavior, thoughts and and feelings. And I think personality psychology has shifted quite a bit over the last years to an understanding that I think is first of all, a lot more accurate and a lot more nuanced. So this idea that we're not

always the same, right? This is always the debate going on between personal psychologists and social psychologist, where personality psychologist, it's the person and social psychologist was like, no, it's the context that determines how we behave. Could it be both? Exactly. And I think that's a model that we have now personality that I really like. So the way that we think about it today is it's not just a one

time measure. It's essentially if I sampled how extroverted you feel behave across many different points in time, what I would get is a distribution, right? So you have like a distribution that's looks relatively normally distributed. You have a lot of stuff that's centered around a certain mean. So if you're more extroverted, your mean is higher. If you're more introverted, your mean is lower. But it still means that we give you at least a chance to be slightly more extroverted than

you typically are. Maybe when you're in a social context, right? You're at a party, there's other people, maybe you feel more extroverted than usual. If you're at the library or at home, you probably feel less extroverted because there's just not as much social stimulation. So I think what this means is we can get a much more one's understanding of a person in a given point in time. So I understand that most of the time you're relatively

extroverted. But right now I can adjust my prediction of who you are because I know that you're not in a social setting. And this is true for humans, right? So we can make these adjustments. The same is true for algorithms, like if an algorithm gets hold of your time behavior history, they might get a good sense of your general disposition to be extroverted. Now, if they can combine this with knowledge of some of the momentary data that you generate, right, it can tap into your GPS.

You can figure out that right now you're at a certain location, it's a bar, it's usually buzzing full at 5:00 PM and that's where you are. Maybe it's going to adjust the prediction to make you a little bit more extroverted. And who knows, Maybe that's the time when I should be targeting you with like an extroverted product because you're self concept of extroversion is activated.

But it could also be that I should actually wait when you feel a little bit less extroverted than usual, maybe that's when you're craving more extroversion and that's when you kind of buy more likely to buy. So I think this dynamic understanding of of personality is where we're headed. That's cool. And this allows me to do a

little bit of a kind of my turn to do a little bit of a detour or like tangent, no, but because you mentioned distribution here. And I think that's super interesting when it comes to personality, because I think a lot of people, if they're not like you or if they're not as experienced with personality, they might be exposed to some of the less. How do you say good models? But more popular. Less robust and where people are placed in kind of very rigid. Dichotomies. Let's name it.

It's the MBTI. Yeah, what we're talking about. Well, that's one of them. I'm in Sweden and so in Sweden I have this hate relationship with a book that is best selling book for the last 10 years, which is about the disk model. So it's even worse the disk honestly dynamic. Yeah. So but whether they're placing a collar or as a letter, they're placed in that specifically. But obviously with Big Fiber Ocean, the whole interesting part is that we're placing a continuum of sort or

distribution rather. And you know, looking at my own, I think it's interesting where there's 2 dimensions where I'm like at really at the kind of like high percentile, but some of them I'm kind of close to the middle. And I guess technically I would probably be categorized as an extrovert. You're. Definitely an extrovert, Sam. I score maybe 55. I score like 60. Ludicrous. Yeah. Definitely an extrovert.

I don't know about that. But then like on something else like conscientious, I'm like at 9095 or something. So like technically if you have to bookmark me, I would be an extrovert and conscientious, but. Yeah. They're very different in terms of where I am on the distribution. And so that's my question basically like how does that fit into this? Like is that? Yeah, yeah. It's such a great question cuz technically, right. So if you start with the data,

we're like million dimensional. So every single data point is in a way a dimension. We're already reducing complexity into five if we kind of take the big 5 again, which already reducing complexity into 5 dimensions. So that's a huge kind of step into kind of limiting the the amount of information that I can can see about you or can use about you. And then as you said with like some of the the questionnaires that have intuitive appeal, right? People love love labels.

People love. Something that they can stick to, they can use it to explain. You can identify with that in case with the MBTI, but the problem is that you're dichotomizing something that shouldn't be dichotomized. And most of us are somewhere in the middle, right? There's very few people are extremely extroverted. Very few people are extremely introverted. So most of us. Hovering somewhere around the middle of the distribution, if that's the cut off, we're just

gonna make a lot of mistakes. I think in the context of personality, what's interesting, and I don't think we we actually know the answer yet is how do we best personalize, right? Again, like if I were to talk to my marketing colleagues, I think the typical argument, the days, the more information you have

the better. So if that's true, I would be much better off taking your entire big 5 personality profile because it tells me something about your extra version, your conscientiousness and everything. However, it could also be true that the ratio of signal to noise, it just becomes much, much higher. So lower, so cloudy. Exactly. It becomes too cloudy because like what does it tell me?

If you're like a six sixty 60th percentile on on extra version, not entirely show what you make of and it might be better to just go for the ones that are extremely salient, right, because the the the ones that are extremely salient. So you're 95 percentile of conscientiousness. That's probably something that's driving your behavior quite dramatically, right? Because it's a core of who you are. Otherwise you wouldn't have take strongly agree on pretty much all of the questions.

And so we might actually be better off ignoring some of the trades in the middle and just focus on the ones that seem to be driving your behavior the most. And again, we don't really know yet. I think it's an interesting question that we haven't looked into empirically. My hunch is that it's one of these instances where discarding some of the information could actually be could actually make you better off. Super interesting. Thanks for diving even deeper

there. That's. Awesome. And we might get back to this in the context of AI, cuz the one thing that we know is like from some of my recent work is like language models like ChatGPT are extremely good, like crafting messages. If I give them, if I tell them, look, Samuel, he's extroverted. Like just come up with an iPhone ad that's tailored to his extroverted personality. Extremely good. But I don't even need to tell it. Here's what extroversion is all about.

It's obviously right. The entire Internet, so it knows. What? Extroversion is all about, and it's great at crafting these messages. Now one thing that we tried initially was to say, OK, let's add more new ones because now that we are not relying on human creative people, maybe we can just take the full personality profile and we could just do this for every person, right? So I can create a very specific ad that's just targeted at you and like in your 5 dimensional

personality space. Now what happened is that essentially the way the Cha Chi PT did it, it just kind of took every single trade and had like one sentence for every single trade. So instead of integrating and saying, well, OK, maybe an open minded extrovert is very different to an open minded introvert, right? The open minded introvert goes to the museum, open minded extrovert is out there in the street with like a flamboyant

outfit. So they didn't really, it just kind of additively said, OK, here's a sense fixed version, here's a Center for openness. And in that case, it just, it felt super awkward, right? It's not a very natural ad that you see when it's just like 5 different statements. Maybe the language models are going to get there at some point, but at least right now you're better off just focusing on one when you use these generative models.

So it's, it feels to me like we have a tendency to focus a lot on the person of the personality and their values and so on. But as you mentioned earlier, there's also a, a really important part of the environment, the situation that a person is in. And this can both influence their personality and you know, what parts of them are amplified and so on. But are there other aspects that, that are influenced by the environment? What can we sort of learn about behaviour?

What can we predict from something like GPS, someone's location, or other sort of passive indicators of where someone is, for example? I mean, it's super interesting because first of all, I mean, GPS is just one of the data points that I personally really like because it says so much about who you are, right? So like GPS records, for example, can tell you whether someone might be suffering from depression because they're not leaving the house as much anymore. It is much less physical

activity. So there's actually somewhere that you can translate context into something that looks more like a disposition or like a psychological dimension. But also context is something that I think is really interesting when we think about privacy because like one of the

big questions right now, right? If you look at all of the companies like Facebook, Snapchat, Instagram, I think one of the, the challenges that they have when it comes to privacy is that they collect these vast amounts of data. They collect essentially your entire history, everything that you've done for years and years. And that data becomes incredibly valuable, right? Because it's like a holistic picture of everything that you've done.

So the moment that there's a data breach or security breach, all of the data is gone. So one of the questions, I think that's partially driven by regulation and the companies are worried that at some point that it's just gonna stop, that they're allowed to collect all of this behavioral data. I think a lot of these companies are looking into alternatives where you can fade out some of these behavioral histories by

considering context. So what I mean by that is essentially, instead of me taking all of the data that you've produced over the last two years and saying, well, you're more extroverted or you're more introverted, maybe I just skip most of that and say, OK, you're right now, forget

about who you are right now. You're in a context that should make you relatively extroverted again, because you're at a bar and there's, it's a social context and there's many people and you're probably attuned to what other people think about you and so on. So forget about who you are, what the starting point is, we're just gonna take this contextual information. And so with some of my students, in collaboration with one of the bigger companies, we actually tried just that. So we tried.

Can we equally well predict what someone is gonna do on the app next by using these contextual cues and fading out some of these more intimate personal histories? And what you see is that it actually works pretty well. So because a lot of the insights that I think you get about people's psychology are captured by context, right?

So if I find you at a bar, well, that also means that your extroverted disposition is probably a little bit higher because you're at the bar and that was your place of choosing most of the time. So a lot of the information that we can get from context also tells us something about the psychology without us having to collect and store for all eternity all of the traces that you've ever created.

So I think it's an into context is like this very interesting window into what's going on inside our mind in a current moment that allows us to potentially protect user privacy a little bit better while still being able to make predictions about your behavior and the next steps that you might take. That's true, though even introverts do get. Once in a while, once in a while. I know it is. Actually.

Some of my earliest research was looking at like the extent to which aligning your spending with your psychology makes you happier. And we created this experiment where we forced introverts to go to the bar and have fun with their friends, and it turned out that their happiness actually dropped. After. Doing so. So it's not only did it not make them happier as it did for extroverts, it just actually made them more miserable. So wow, Yeah, I I hear you.

OK, awesome. And in terms of those sort of real time opportunities for interventions, can you think of any other examples of like, you know, based on data from our the sensors that are in our watches and our phones and those sorts of things like if you're exercising or like, like what other things can you be doing that we can learn about that can maybe trigger or silence some kind of health intervention?

Yeah, for example, I think so for me, it's actually the combination that's the most interesting part, right. So if I can use your GPS records, for example, to say, well, it looks like you might be entering a depressive episode cuz again, you're deviating from your behavioral routine, you're spending more time at home, you're not seeing friends as much anymore. I have a better sense that you

might be struggling. Now the momentary aspect comes in when I see you might be struggling and you're currently walking close to a park. So now the best intervention is just to tell you, like, look, seems like maybe you're not having a great day. Why don't you just go and spend some time in the park? And I think that's the combination of like, what are the predictions about who we are

a little bit more generally? Plus, what's the context that we're in right now that could drive interventions that just are so much more convenient and so much more personal? Yeah, much smarter. I'm so tempted to be like putting you against the wall and being like, OK, if you would be trying to get some person to, let's say, go to the gym and you have to choose like either, you know, could lose equation.

You know, either you can take the the data you have on the P the person, or you can only take the data you have on E the environment. Which one would you choose if you can only take one? I would probably, and for different reasons, I would probably go with the context because the reason for my context is so interesting in this specific space is that it usually cues habits, right?

So the moment that I know you typically exercise, if this happens when you walk past the gym, when you kind of get up early in the morning, those are all cues that in a way get me to a habit. So they kind of allow me to trigger a habit that it's much harder to trigger with with like the P component, with like the dispositional component.

So I think in the context of exercising, just because we know that habits are so important and so critical in this case, I would probably go with a contextual one. What? What I want to do here is kind of your thinking process around that, like how you think about that. So that's really interesting to hear. And yeah, I guess now should we?

Yeah, talk about AI. I mean, it's funny cuz I think we've been talking about AI all the time, by the way. So I think anything that's it's like machine learning is just a subset of of AI. So, but I think there's like a, yeah, a qualitatively different level of AI that we see now. And I'm happy to get into that because I do think it's changing this entire game of psychological targeting quite a

bit. Yeah. Where would you start in terms of unpacking what has maybe changed in some ways your world with the advancements of AI the last few years? Yeah. So I think we can actually go back all the way to the beginning and the framework that we established, right? If you think back those two steps, one is we take data and make predictions about who you are. And then the second step is based on these predictions, we try and change your behavior and AI.

So now I'm talking mostly about generative AI. So large language models like ChatGPT, they allow us to, in a way, scale and democratize both parts. So the first one, back in the day when I started, you essentially had to have a proprietary model, right? So if I wanted to get from your Facebook likes to your personality, what I needed was, first of all, the data set that got me access to your Facebook profile, your Facebook likes, your status updates in combination with a self report

of your personality, right? So you also had to complete a questionnaire to tell me, here's how I see myself and here's how I would describe myself. And then I was able to start. Training these models that say, OK, maybe if you like Lady Gaga, you're more extroverted, if you like manga, you're more introverted. But for that, I had to have the data set and I had to train the model. Now Fast forward to today, what you can do is you can essentially just take someone's social media profile.

So just take all of those the status updates that someone, someone post their Facebook likes, you can pop it into ChatGPT and you can just ask it, OK, what do you, who do you think a person who posted the stuff on Facebook and who followed these pages, who do you think they are in terms of the big 5? Just give me a numeric rating from 1:00 to 7:00 on each of the five dimensions and it does a

remarkable job. So we just published this paper showing that the accuracy by which ChatGPT, which was never explicitly trained to do that, right? So and nobody ever told ChatGPT get me from status updates to personality profiles, but it still can do so with almost as high in accuracy as if you train the model explicitly for this purpose. And that is a total game changer. So first of all, it means that

anyone can do it right. So back in the day, it was like this outcry of well, Cambridge Analytica got access to all of this data and Oh my God, they have these models like a 15 year old kid in the parents basement can now scrape your data online, run it through championship BT and essentially have like the psychological profiles of millions of people. And not just a big 5. But if you've, if we forgot to collect moral foundations in our original data set, that's it.

We're not going to be able to predict it like chat. GPD can predict whatever you wanted to predict. You're not limited to the Big 5. You're not limited to what you what data you collected initially. It just knows everything about any psychological trade and you can take it even a step further, right. So back in the day, we just looked kind of passively and looking backwards retrospectively at someone's history to make predictions. Chat bots have the ability to

dynamically interact with you. So some of the more recent work we actually looked at, well, can we train a chat bot to have a conversation that still feels natural, like getting to know someone on a first date, and then based on that conversation, make predictions about who they are. And those predictions are far more accurate than anything we've had before, which makes a lot of sense, right?

Because if the chat bot figures out, well, I don't really know enough about his level of contentiousness, can just ask questions that go in this direction. So it's like almost a self-guided dynamic scouting of someone's personality. So that's Part 1. Part 1 is that we can now democratize and scale the prediction of psychological traits just by having these tools available. But that's obviously only part of the story because it also generates content, right?

So if the moment that I know is amazing at generating content, that's why it's called AI, believe it or not.

But so once you give it like the instructions to say, well, create me for this is like an example that we used in one of our studies, come up with a short political speech advocating for climate action for someone who scores high on the moral foundation of fairness, which is like, again, the way that people think about what's right or wrong in the world, often times doesn't happen in the Big 5 personality space, but more in the space of moral foundations.

And it does a remarkable job. So I don't need my team of creed. That's right. When I started, I had to head up a creative team that sits down, comes up with a pictures, comes up with the text that goes on the ad. And that was extremely limiting. So with a beauty retailer example that I gave earlier, we only tailored the ad that people saw on Facebook.

The moment that they went to the website, it looked the same for everybody because we just clearly didn't have the resources to now customize every single page on the beauty retailer's website. Could be totally done today. At the moment that I know that you're extroverted, we could just in time create an entire website for you that's specifically tailored both to who you are, maybe the context, maybe the steps that you've already taken on the website. And that's a total game changer. Yeah.

And, and when you're personalizing for moral foundations, what's, what are some examples that ChatGPT would come up with when tailoring a message to say for climate action for a conservative versus a liberal? Yeah. So I mean, So what we know is that the moral foundations that are more closely and lined with being liberal are fairness and care.

So when you pop that into ChatGPT, it just essentially talks about how it's our responsibility to care for the planet because it's not just us, and it's only fair that we take care because future generations are otherwise going to suffer. It's a very liberal argument, right? It kind of we have a responsibility to look out for others, even though we're not going to be impacted immediately. Potentially the the arguments for for more conservative people are usually focused on purity

and authority and loyalty. So if you think about loyalty is like essentially we have this connection to Mother Earth and there's something that like the loyalty that we show towards the people in our community and mother is something that we should be focused on or we just want to have to make sure that the planet is in pristine shape. That's kind of playing. Not so much for future generations. Exactly. Not so much for future generation. That's because we care about

other people. But that's just like 1. We just have to be pure in the way that we act and in the way that we interact with the planet. So I think those are like, you just have to look at them. And again, just take any personality or any psychological trait that you can come up with and it just does a really remarkable job. So I find it fascinating just playing around with it and seeing what it comes up with. And it's often times so much better than anything I would

have come up with. Far better. And I guess I want to ask you in terms of what I felt recently in the last few years, you talked about science fiction and the difference and just being timing. And a lot of these things would be science fictions before, but now it's kind of more used to science or like reality. And for me, I ended up used for fun creating, you know, AI clones of my voice and my face and all of these things.

And through this kind of automation as well, you know, the end point could be kind of a short video message from me that looks like it's for me, sounds like for me. My partner could. Not really see that it's not me, but it's just something mass generated thing that it's just starting from a simple, let's say and starting point of some

data point. It flows into kind of a 22nd message talking about the person's not only name a certain things, but also you like in a way that is likely to maybe make them feel seen or, or appreciated or or heard and so on. And I think that being possible is science fiction. Like I think it's so it's strange for me. I mean, I think it's, it's, it's a terrifying, I think science, right, because I don't think it's science fiction, right. So you can totally do that.

As you said, you, you didn't, the question is like, how good is it? And for me, the, I was recently talking to a New York Times journalist who outsourced her decision making for a week. It's an amazing article if you want to read it, But essentially she outsourced her decision making for a week to just generative AI. And, and one of the things that we talked about, which then she also describes in the article is that it's really helpful in many

ways, right? Because it kind of has this in like it can give you cooking instructions much better than you could kind of get from the Internet. But what it really does it she says, like it turns it turned me into a basic bitch. And that was like one of the one of the concerns that I have is that it's just going to make us

incredibly boring, right? So I think AI is really good as a society, homogeneous, as a society and also as individuals, because it's just kind of playing into the same thing. And again, right. So if the things that Samuel is an extrovert, that's how it's going to talk with everybody that you're interacting with and then the feedback that you get is just kind of reinforcing everything that you've done.

So I think that the way that we kind of create these doppelgangers of AI, if we don't bake some exploration and some uniqueness into the way that we designed them, we're just all becoming like so average and so boring. I think that's one of my when I'm concerned about privacy, the loss of self determination, but I'm also really concerned about

just us being so boring. Have you there is this game, I don't know if it's a game that that people have been sharing where you actually, if you are a regular user of ChatGPT, you can go to ChatGPT and ask it to tell you something about yourself that you don't know. Have you experimented with I've? Not done that yet. Oh. Very funny because you, you need, I mean ChatGPT would need a bunch of information about me to play the game well, right? So it's an interesting one. Talking about games.

Quick, quick game we have for this season of the podcast is something called To AI or Not to AI? And this will be a very quick fire round where basically we'll give you a few prompts and you used to have to answer what do you think it's a good suited or not well suited for AI? To outsource to AI, Yeah, well, we'll see what the questions are exactly. Good. OK. Are you ready? Yep. Clean your room. Oh, if there was like a physical AI? Absolutely. What about make sure you wake up

on time? Yeah, AI, you're like, yes, everything generate targeted ads for potentially pregnant customers. I mean, is it counterfactual to have humans generate targeted ads for potentially pregnant customers? No the counter. No the counter. This is a task that is well suited for AI and doesn't have to be an alternative. Well, I mean, I think AI would do a remarkable job. Would I want to be targeted by it? That depends on whether I ask for it or not.

Because do I think it could actually give me amazing recommendations? Yes. Do I want to do this behind my back? Probably not. With consent, yes. What about improve the Big 5 model of personality? Yeah, great. Predict your ideal pet. Or if I don't have to follow that recommendation, I'll take the idea. OK, OK, so final one, personalize the way your GPS speaks to you based on your

personality profile. Personalize the way that my GPS so my navigation in the car is that what you and yeah, I don't have a have an issue with that. I think what is true for all of these questions is that I would want to understand how it works and what it does. And most of the time, I think once you understand it and you have some control, it can be

extremely helpful. So that's, I think generally speaking, I err on the side of there's a huge opportunity if we manage it well and less so on the we're all doomed. Well, that's a perfect segue to our final question that we ask all of our guests. What is your most controversial opinion about AI? That's I mean that that might be it. It's just funny because I don't think it's really hard to have controversial opinions of AI because like people's opinions are all over the place, right.

So you. Can get all the opinions are already out. There, Yeah, they're already out there. But I do think that I probably, if you kind of compare me to the average person in that space, I think I'm probably more optimistic. I think that where I'm less pessimistic is that we have the ability to do it ourselves as users.

So I don't think we can just have regulations and mandate transparency and control and say, well, would just tell you what to do and then you can navigate the landscape yourself. That I'm very pessimistic about. But I do think that there's so many good use cases and with just like the right regulation and some of the new technologies that allow you to do it in a

privacy preserving way and give you more control in a supported way, then I think the opportunities are incredible. Yeah, you, you talk in your book about having better systems by design. Yeah. Making, not putting that burden on the user to, for example, read the terms and conditions and know what they're signing up for, but rather help them without all of that effort. Exactly. So awesome. Yeah, this is so much fun. I feel like we could have talked

for so much more. But I guess like for anyone who feels like the one to learn more. There's a book they can read, so we will make sure to help people along the way to to get the book and. Yeah, it's just out. Yeah, it's just out. And it's funny because I, I didn't start writing it with a

book in mind. So I just started writing these short stories from my friends and family just to kind of give the well, first of all, to make it more enjoyable to follow my research because I don't think reading scientific articles is just what they need, but also just to tell them like, look, here's what your data says about you. Here's why maybe you should care about it as well. So I try to make it as a approachable and accessible as as possible. It's amazing.

And yeah, it was lovely to read it. And yeah, so congratulations on on the book and thanks so much for coming on today. It was, yeah, fantastic. Two foundations to speak, and. Yeah, thanks for all your wonderful work and your research and and everything. It's fantastic. Thanks for having me, was really fun. And that's a wrap. You've been listening to the Behavioral Design Podcast brought to you by Habit Weekly and Nuanced Behavior.

Sam and Alene tell me. This season is packed with incredible insights about behavioral design and AI, so be sure to subscribe and share the podcast with your friends. Though you might want to keep it away from your enemies. In case you haven't noticed, I'm an AI voice. Yep, pretty crazy. Quite the improvement since last season's AI outro, don't you think?

If you'd like to collaborate with us at Nuance Behavior, where we use behavioral design to craft digital products with Nuance, e-mail us at hello@nuancebehavior.com or book a call directly on our website, nuancebehavior.com. A special thanks to the amazing Dave Pizarro for our show music and to Mei Chen Yap and April English for their help in producing and publishing this episode.

Thanks again for tuning in. We'll be back soon with another exciting conversation where behavioral design and AI intersect. Happens to Murgatroyd. Oh. Well, hello there. Are you still listening? Wow, we really. Why? Yeah, why? Why are you still here? Why are you still listening? Get out of here. Get out of here. But I guess to reward you for being such a loyal listener and making it all the way to our kind of like little hidden gem of a final snippet, we have a recommendation.

If you liked this episode with Sandra, check out our episode on personality with Sanjay Servastava. He is one of my former colleagues at Apple. In that episode, we got all the way into the details of personality, how it's measured, how it's used and so on. So. Check it out. Yeah, so sorry, no embarrassing L takes this one but. I know, I thought it was so ironic that you said that as we're like stumbling over. OK, we'll just say. I don't think you have to say anything else.

OK. I think we can end with that. We'll. End with that only if you want to bye bye bye.