CM 283: Sandra Matz on Protecting Our Privacy Online
54:22 Share

There's oftentimes a question of like, how do we get there, right? You can either have privacy and self-determination and agency and control, or you give us all of your data and you can have the service and convenience and personalization and all the good perks that come with it. My hope is that we can break down this dichotomy because if it's an ease of author question, everybody's going to choose the personalization, convenience and service, right? That's how the human brain works.

I do think there's technology now that allows us to do the opposite, to actually break it off and say we have personalization, convenience and service, but also privacy and self-determination. Welcome to Curious Minds at Work. I'm your host, Gail Allen. With few exceptions, we have digital footprints. And each time we scroll social media, run a Google search or use a smartphone to navigate, we're adding data to that footprint.

While we gain a lot from our ability to do all these things, we also give companies the data they need to target us. Sandra Matz is a computational social scientist and professor at Columbia Business School. Over the course of her career, she's consulted with companies eager to profit from our data. In recent years, she's intentionally shifted her consulting work in support of organizations that want to protect consumer data.

In this interview, I talked to her about her book, Mind Masters, the data-driven science of predicting and changing human behavior. We discussed the methods companies use to profile us and how that profiling puts all the power in their hands.

We also discuss promising ideas for pushing back, insights that have the potential to empower and unite us. Matt's has written an accessible, highly readable book that anyone with a smartphone needs to read. I walked away from the book wanting a whole lot more transparency and accountability from tech companies. Before we start, one quick ask. If you like the podcast, please take a moment to leave a rating on iTunes or wherever you subscribe. Your feedback sends a strong signal to people looking for their next podcast.

And now here's my interview with Sandra Matz. Sandra Matz, welcome to the podcast. It's great to have you on. Thank you so much for having me. You've studied the use of data for psychological targeting. What does this mean? And is there an example you can share? It's a great question. Yeah. So the way that I think of psychological targeting is essentially a way for us to read the mind and write into the mind using data. So you can imagine all of the traces that we leave,

pretty much every day as we interact with technology that could be anything from your social media data to your credit card spending to the data that gets captured by your smartphone so the places that you visit with gps the calls that you make

All of this we can use to get an insight into your psychology. So say, or your personality traits, your political ideology, and then using these insights, understanding, say that you're more extroverted or more introverted, we can actually use it to also try and change your behavior. So that's the second part of what I think of as writing into the mind. To get a sense of the sea of data that we're swimming in, about how much data does the average person generate?

A lot. And that might not come as a huge surprise, but every hour, the average person generates about six gigabytes of data. So that's an awful lot. And I think what's so interesting about this data is it's not just a volume. It's really the variety, right? It captures everything that you do pretty much continuously. And it gives us this puzzle piece of your existence, the way that I think about it.

You share findings in your book from a research study that shows how analyzing a user's Facebook likes, for example, allows researchers to judge a person's personality even better than close friends, family, spouses, partners. Talk about this.

Yeah, this is, I think, was one of the most fascinating studies that a friend of mine at Cambridge University conducted. Because when we talk about we can use someone's data to predict their personality, their psychology, the most common question that you get is, well, how accurate are those predictions? And it's really hard to just throw out numbers, right? Nobody really talks in correlations.

And so she had the idea of like, how do we quantify the accuracy? And her idea was, well, let's just compare it to how good other people in our environment are that should know us pretty well. So let's say you took a personality questionnaire, you answer statement like, I'm the life of the party. How much do you agree? Based on that, we can infer whether you're more extroverted or more introverted. And now we have essentially two competitors trying to predict how you describe yourself. So one is the computer

And in this case, she only looked at people's Facebook likes. So those are the pages that you follow on Facebook to make a prediction of how extroverted you think you are. And then she asked other people like your friends, again, spouses, family members to complete a questionnaire on your behalf. And what she found, which I think was, again, is like almost 10 years ago, a pretty remarkable finding was,

was that the computer, just by looking at your Facebook likes, was better at kind of understanding who you are than pretty much everybody on the human side. And those people know you pretty well. Your family presumably spends a decent amount of time with you, understands what you go through in life, what you do on a daily basis, and still a computer could make judgments and predictions that were as accurate or even more accurate. And just so we're clear, how are you defining personality?

There's actually many ways in which you can define personality. Typically, we think of it as like relatively stable dispositions. And we might get into some of the nuances here a little bit later on. But generally speaking, it's this idea that we all have tendencies in the way that we think, feel and behave.

And the most common framework that we use as psychologists and as researchers is what we call the big five framework of personality. Some of you might also know it as the ocean model. But the idea here is that it captures these five broad psychological dimensions of openness to experience, conscientiousness, extroversion, agreeableness, and neuroticism, which try and kind of carve out these differences between how people, again, think, feel, and behave.

You explain that we post things online because we want to be seen, that it actually triggers a brain response that's like receiving money or having sex. Can you say more about this? Yeah, it's one of my favorite findings of Diana Tamir. She's a psychologist at Princeton. And what she was studying is this idea of, you know, like on social media, we constantly crave feedback, right? We

first of all, we spend a lot of time talking about ourselves. And some of the benefits, the rewards that we get is obviously in the form of social feedback. So if we post something and we get a like from our friends and they share the stuff, that's like the social reward. But what we know is that even just talking about yourself is like intrinsically rewarding. So what she looked at is like people's brain response

when you get the ability to talk about yourself or like a topic that's not related to you. And people not only had like a brain response that, as you said, is similar to when you take drugs or have sex, but they were also willing to essentially give up money that they otherwise would have been paid in this study just to be able to talk about themselves rather than just talk about a random topic. So there's something about talking about oneself that we just love.

You talk about the fact that the focus is on language in what you're looking at, not images. How is this approach connected to James Pennebaker's work on how language offers a glimpse into our psychological well-being, even our mental health?

Yeah. So James Pennebaker, or Jamie as he usually goes by, he was one of the pioneers, I think, in the field of computational linguistics. And I can just give you an example that I found fascinating. And Jamie actually mentioned at a conference at some point. So he asked the audience to think of like, what are some of the psychological traits that we might associate with the use of first person pronouns? So like I, me, myself, and

And I think that, again, the room was like full of psychologists, full of personality psychologists who think about this pretty much every day. And most of us, I think, had the idea that, well, if you talk about yourself, you're probably a narcissist, right? It's like a very strong focus on the self. And what he revealed in the next slide, I think, blew our mind because it was not expected. And that's that actually the use of first person pronouns is not necessarily a sign of narcissism. It's a sign of emotional distress.

So the fact that you're talking about yourself showcases this focus on the self and this inward focus. And at first, I think all of us were a little bit surprised because we didn't see it coming. But then also, once you start thinking about it a little bit more, when you're feeling really, really down, you probably don't think about how you're going to save humanity and how you're going to solve climate change. What you really think about is, why am I feeling so bad?

What can I do to get better? Am I ever going to get better? So this inner monologue that we have with ourselves also creeps into the language that we use because we can't constantly monitor everything that we think and how that translates into what we say. So we suddenly leave the skews to our inner mental life and in this case, essentially emotional distress.

It's so interesting because you talk about that, and I agree with you. I immediately thought of narcissism, and then when I read this, it was quite revealing to understand that, wow, it might actually be like a cry for help from someone. Is there also a connection between physical health and mental health in the words that we choose?

Yeah, so that's another interesting finding that I think in a way reflects a truth that we've known for a long time. But it's so nice to see this borne out in the data, right? So again, like the same with the first person pronouns and emotional distress, like once you know this, it actually is, it reveals like a psychological truth that we otherwise might not have known.

In the case of physical and mental health, there's a study that was conducted with Tweet. So it was a team of researchers around Johannes Eichstatt, who's currently at Stanford. And what he was looking at is what are the linguistic signatures of people who have been diagnosed with clinical depression?

And what they show is that many of this, so some of it is just negative mood, right? Negative affect, what you would imagine people are being sad, people are kind of writing more about their depressive feelings. But another category of words that was related with clinical depression was essentially physical symptoms. So pain,

being hurt and so on. And again, we don't really have an understanding of which comes first, right? Do you experience physical pain and that leads you to also suffer from emotional pain or is it the opposite? But it clearly showcases that the two in some level go hand in hand. You also talk about the fact that Facebook status updates, they actually can reveal people's income. How does that work?

Yeah, that was one of the studies that I still conducted when I was a graduate student at Cambridge. And for me, it's an interesting study because I think it tells us something about how we can use data. And I'm going to tell you in a second what I mean by this. But so what we did here is we essentially looked at, again, we captured someone's income using questionnaires and self-reports, and then we connected it to their Facebook status updates. And some of

the things, some of the relationships that we saw actually are, I think, quite intuitive, right? If you have high levels of income, you talk about your amazing vacation, you talk about luxury brands and stuff that you buy, and then the lack of that is really indicative of lower levels of income. There are some relationships that just when you look at them feel extremely uncomfortable. So something like

low income people being a lot more focused on the self, right? So the same way that we previously said focus on the self is related with emotional distress. They focus a lot more on the self. They also focus a lot more on the present. And I think this is like one of the cases where looking at big data can actually tell us something about the way that society works and tell us something that we might not actually be happy about, right? So it's just a signal that once you

once you kind of struggle to meet your financial goals, life just becomes really, really difficult. So it's really difficult if you're struggling to put food on the table to think about the future. So that's why they live in the present. And it's also really difficult to think about other people because you're constantly trying to figure out how you make this work and how you can survive the next day. So for me, it's just like one of these examples that

actually teaches us something about how society works and gives us a foundation of changing something, right? It's not normative. So we're not saying this is how society should look like. What we're saying is this is based on the data, what we currently see, the relationships that we currently see. So now let's use it to, if we don't like it, change it. What, if anything, can researchers learn about our personality and our character from our physical features? And what role does deep learning play?

It's a great question. I think it's one of the most, still most controversial topics, I should say, ahead of time. So this is, I think, some of the latest research that has looked at the relationship between people's faces, starting with just the pictures that they post about themselves, right, that could include grooming. So we know, for example, that extroverts are somewhat more likely to dye their hair blue.

blonde. They're probably more likely to wear contact lenses. They're also more likely to smile in pictures. So those are, I think, all characteristics of like, how do you take pictures and how do you style yourself? Now, some of the claims that the researchers like Mihał Koszynski, for example, make is that it's not just a grooming that we can pick up on, but it's also some of these facial features that are directly associated to our personality. And this has a pretty dark history, right? So they're for a long time

And there was this idea that based on your facial features, we can predict whether you're criminal. We can predict if you kind of have like good genes and so on. So there was like a pretty dark history of physiognomics. That's the way that the science is called, being abused in pretty nefarious ways.

Now, the argument that he makes, which I think is an interesting one, at least kind of considering scientifically, is that there's many pathways by which what we look like could actually impact our psychology and vice versa. So one of the examples that he gives, for example, is that let's imagine that you were born a really beautiful and pretty baby.

What happens? And I've just had a baby 10 months ago. So I see that there's differences in how people react. So it's very top of mind for me. But you're a beautiful baby. Everybody stops and smiles at you. You have these most pleasant social interactions. The chance is that...

You being pretty and being constantly bombarded with positive social feedback maybe makes you a little bit more extroverted is not totally far-fetched. There's actually research showing that that's the case. Same is true for hormones. We know that testosterone, for example, shapes our facial features.

But we also know that it kind of associated with more aggressive behavior and so on and so forth. So even though the original science of like what your face reveals is related to your character was totally debunked, I think there are theoretically

reasons for why we might see some of these relationships. And also empirically, we just see that based on your face alone, try removing all of the features of grooming, of makeup, of hair, actually you're still predicted of some of these personality traits. Now, I'm not saying that this is good news, right? It's certainly like once you're able to make predictions about who you are based on your physical features alone,

That kind of raises entirely new ethical questions because you can't leave your face at home. So you could say, I'm not going to use social media. I'm going to leave my phone at home if I don't want someone to track me. We have facial recognition and cameras pretty much everywhere. So if that's true, we're talking about a whole new level of intrusion. And deep learning, of course, plays a big role in this.

So deep learning is interesting because the way that we typically make these predictions, like let's say we get from your Facebook likes to your personality or so on, is with very concrete what we call feature. You can think of it as like cues to personality. So if you like the page of CNN, you get a one. If you don't like it, you get a zero. If you like the page of Lady Gaga, you get a one. If you don't like it, you get a zero. Right.

Now, having concrete features in a face is really tricky, right? Because our faces, when you think of these more objective features, are very similar. Everybody has two eyes, one nose, one mouth. So you're not going to get very far. And what deep learning does, it's essentially kind of goes through the picture pixel by pixel and just kind of tries to bottom up without me telling it, here's the features that you should be looking for. Try to see if we can spot patterns.

And sometimes what is the way that you can still probe the model and see what's happening is you can essentially say, let's say we're trying to predict extroversion. You can say, well, model, just give me the 10 pictures that are predicted to be most extroverted, morph them so that we just have like one average prototypical extroverted face. And now you can look at what comes out. So even though you don't fully understand what the model is doing, you still see that the 10 pictures that were predicted to be most extroverted have again, blonder hair,

lighter eyes, a smile. So even though the features itself are somewhat more obscure, you can still look at the output. You have talked about this, how we leave digital breadcrumbs, behavioral residue, traces of our actions for others to find. And you specifically talk about three ways that we do that. I think in general, a lot of us are not aware that we do this, especially in these three specific ways. Can you tell us what these three ways are? And then we're going to go into them a little bit more in detail.

Yeah. So the way that was everything that we've talked about so far, like in terms of social media is like what we would call explicit identity claims. So this is the, those are the signals that you intentionally put out there, right? So you post on social media because you have, you want to be seen by the world in a certain way. Now, behavioral residue are all of the traces that you create without really thinking. And those are actually a lot more than I think people realize. So whenever I give talks, everybody's focused on social media for

getting about pretty much all of the other data traces that we generate. So the three that I talk about in the book, and there's many, many more, are like Google searches, credit card spending, and smartphone sensing. So again, typically, when you carry your phone with you 24-7 and someone is tracking your GPS location, you don't think about, oh, I'm now going to go to a supermarket because I want to be seen as a certain person, or I'm going to go to the park because maybe people are going to

think I'm more introverted. So those are all of the traces that we typically have much less control over. So let's take, for example, spending records, how we spend and what others can learn from us. How little data do we need to figure out who a person is and what else can we get from spending records? Yeah, that's a great question. And the way that I think about it is actually there's these different layers of identity, right? So there's this work that I love that was done by a

by a guy at MIT around a group of Alexander Pentland. And what they looked at is, can we actually identify a person? So in this case, we're trying to say, well, I'm finding Gail in a data set. I'm finding Sandra in a data set. And that has been previously anonymized. So think of it as, I'm going to take the data of everybody who lives in New York. I'm going to get access to their credit card spending. I'm going to anonymize it by taking away date of birth, name, address, and so on. So I just see

what they buy, where they buy it, and when they buy it. Now, this data set should technically be fully anonymous. But what they showed is that the moment I get three data points, so I know that you, for example, went to Starbucks in the morning to buy a chai latte, then you had lunch in a certain place, and maybe you went to an art gallery at night in a specific location.

The moment that I have these three data points, I can actually say that this is you because there's only so many people who have the same exact spending signature. So if we think about spending records, and the same is true, by the way, for GPS records, so what we can track with your smartphone,

It becomes very quickly, very personal, almost like a like a fingerprint. So that's level one where we're talking about real identities. But the same way that we talked about psychology before, I can also use your spending to predict whether you might be more extroverted, whether you might be more impulsive and so on, just by looking at what is it that you actually spend your money on.

You know, it's so funny. When I read this in your book, I thought, well, this makes so much sense. But I thought, how often do I really think about this? And I don't, right? And yet what you're describing is, you know, we're going to talk about this later on, but it feels intrusive. It feels unsafe on some levels. And so it's just really, really interesting to recognize that there's so little that it takes to identify ourselves.

Yeah. And I think especially, right, so credit card spending is interesting because there's certain ways in which we also use it to express ourselves. So the moment that when we think about spending, we typically think about, well, we're going to buy this flashy car, we're going to buy this watch and so on. But we don't typically think of, well, the fact that I now went to the deli on the corner five times a day, probably

also sends a signal that I might not be the most organized and reliable person, right? Like the organized person goes to Costco once, they buy everything that they need for a week. So there's like all of these traces that are much more subtle than the ones that we typically think of. And it makes us, you know, for advertising and marketing, it makes us, you know, the kind of person who is going to be approached in a certain way. Yeah. What about GPS coordinates and mental health?

Yeah, that's another, I think, fascinating topic that also creates a lot of opportunities. So this is some work that we've done in my lab together with Sandrine Müller, who used to be a postdoc. And what she was interested in is essentially, is there a way that we can detect relatively early whether someone might be struggling with something like depression? So the problem with depression, for example, is that once you enter this, like a

full-fledged clinical depression. You're kind of in this valley and it's really difficult to get out, right? First of all, you're probably, oftentimes people are just not looking for support because they're so inward focused and they're just not leaving the house, the bed as much anymore. So is there a way that we can catch you much earlier? So when we just kind of might see that there's differences to how you typically behave and then try and get you connected to

with the help that you need. So when we look at GPS or like other phone sensing data, for example, what we can look at is, well, is it that you don't leave the house as much anymore, right? You're spending a lot more time at home, which is what we can get from your GPS records, because that's where your phone is pretty much every night stationary plugged in. So we get a pretty good sense of where you live. We can also see that there's much less physical activity. So just the extent to which you travel to different

places might be reduced compared to your baseline. You might not be taking and making as many calls anymore because, again, you're focused on the self and you're shutting yourself out from the rest of your social environment. So

Those are all indications that, again, you might be suffering from depression. It's not a diagnostic tool. So you're not getting the accuracy that you would get by going to a therapist and going through all of the diagnostic tools that they have. But it's still interesting because it's passive, right? So

You don't have to go to a therapist to have, let's say, an alert pop up on your phone, especially if you've kind of previously indicated that you want help in tracking your mental health, that says, "Hey, look, it might be nothing. Maybe you're just on vacation. That's why you're not at home. That's why there's much less physical activity. That's why you're not making taking as many calls. But why don't you look into this? Because we see that there seems to be something off. You don't have your typical routines. So why don't you just try and look into this?"

You could even imagine that if you really know that you have a history of mental health problems, that you nominate someone else. So if I, for example, I could nominate my husband and say, look, if there is any indication that I might be sliding into like a depressive episode again, why don't you reach out to him first?

And again, do the same thing. It's not a diagnosis, but why don't you check in with Sandra and see if she's okay? So I think there's many ways in which you could use this early warning system in a way that helps people. In this interview, data privacy expert Sandra Matz discusses how easy it's gotten for tech companies to target us. In revealing their methods, she shares what we gain and what we lose. If you'd like to learn how design impacts our decisions, check out episode 198 with Eric Johnson, author of the book, The Elements of Choice.

He uncovers how designers of everything from restaurant menus to product websites deliberately influence our decision-making. For years, people have been studying decision-making and noticed that depending upon how you pose the question to people, they will actually make different choices. And, you know, this often gets called as a demonstration of irrationality. What choice architecture does is flip that on its head and say, we can actually use the right way of posing a choice to somebody that will help improve their decision-making.

Now let's get back to my interview with Sandra Matz. Talk about your experience wanting to impact the healthcare sector versus using these kinds of tools in corporate marketing.

It's a sad and funny story in some way. So I started my PhD wanting to use all of these insights to impact the healthcare sector. So my idea was essentially, I always wanted to know how do we make it actionable? So my colleagues had at the time that I was starting my PhD, they had already written the first papers on I can take your social media and I can make predictions about who you are. And the immediate question I had was like, what does it mean? Once we can understand someone's psychology, there's a clear potential to use it to

either exploit people or help them make better decisions so my initial um my initial goal was to see well can we help people comply with the medication that they get can we help them to do their health checkups because it seemed so obvious that once you know what someone's motivation needs and preferences are that you could actually use it to to help them do that and that was like a

challenging avenue because the healthcare system was not necessarily excited about me just trying to overhaul it without any real evidence. And in my early twenties, they were not exactly pumped. And so I relatively quickly pivoted to the marketing context. So totally different, right? So marketing, first of all, they have...

ton of resources, any single improvement that you can show them, like improve a marketing campaign by 2% and everybody's excited because typically the budgets are really, really big. So I relatively quickly pivoted from the healthcare space into, can we just have a proof of concept for the idea of psychological targeted? Once you know that,

what someone's personality looks like by predicting it from data? Is there a way to change their behavior? So that's when we started working with companies to see how this plays out in the marketing context.

So give us an example. You worked for a beauty retailer and you did some things there. And then how that may be different from your work with Hilton Hotels. Yeah. So the beauty retailer was one of the first studies that we did. And the idea was to say, well, once I know, so their goal was just to get women to click on a Facebook ad and send them to their website and then buy something.

So we were in a way agnostic to the specific products that they bought, but the idea was like, could we make beauty relevant to women of different personalities? Right. So why is it as an extrovert that you might be buying beauty products?

Probably because you think it makes you stand out from the crowd, makes you the center of attention. It kind of gives you the social capital that you're looking for. But it's not true for introverts. Right. So introverts might buy beauty products because it helps them relax at home. It helps them make the most of their me time and whatever it is. So the idea was really just to kind of see like what makes beauty products relevant and interesting for different personalities.

So what we did is we came up with, um, we focused on extroversion in this case. So we came up with, um,

different ads that were tailored to extroverted and introverted women. And you can imagine that the extroverted ads, there was always a lot happening. So it was like a woman on a dance floor, for example, surrounded by a lot of other people, very saturated colors, a lot happening. The text would say something like, dance like no one's watching, but they totally are. So playing with the need of extroverts to be the center of attention. And then contrast that to the introverted ads that the

team of designers came up with, which was much more quiet and reserved. So it's typically one woman in front of the mirror or at home enjoying some beauty products. And the text would say something like beauty doesn't have to shout. So again, trying to tap into the motivations of extroverted and introverted women. And then we ran a campaign on Facebook where we targeted extroverted and introverted women. And here I should say, and I think that's actually like an important distinction to Hilton when we talk about this later, is that

Facebook doesn't allow you to target personality traits specifically, right? So you can't say on Facebook, I want my ad to be sent to extroverts and introverts. But what we knew, and this is coming back to some of the work that we've discussed already earlier, is I know that certain interests like liking CNN's Facebook page, liking Lady Gaga's Facebook page are related to certain personality traits. And now I can say, well, just

Give me a list of interests that I can target on Facebook that are highly associated with extroversion. And by doing so, you define an audience that on average is more extroverted than the average person. So we're not really taking someone's full behavioral history or Facebook profile and we're making predictions at the individual level. We're just defining audiences.

And so that's a very, very crude way of targeting. And still, we saw an uptake of about 50% when it comes to purchase rates and return on investment for the beauty retailer. It's like a pretty substantial level. I talked earlier about 2% being exciting for marketers. I can imagine they were really happy. And again, this was like a very crude way of targeting people.

and also a very crude way of customizing because the only thing that we changed was essentially the ad that women saw on Facebook once they went on the Beauty Retailers website looked the same for everybody.

Wow. And then with Hilton, did you do something similar? So the reason for why I like the Hilton example is it was much more involved on the consumer side. So the beauty retailer was really just top-down, right? We show you, we passively predict your personality, we show you an ad, and we hope you buy more. Hilton made it part of their value proposition and also gave their users a lot more control, which is one of the reasons for why I like this project so much. So in this case, we just...

Hilton customers. Well, we're building this app that is kind of helping you to find a perfect location based on your personality. What you can do is you can connect with your Facebook profile. We run it through our algorithms actually on the

Our own service back in the day was Cambridge University. So Hilton never even had access to any of the data that participants or like the users, customers generated. So we ran it through the algorithms who predicted, let's say you're more extroverted and open-minded. And now we use that to pick the perfect look

vacation for you. So it was really very much focused on including and involving the user and not doing it behind their backs, but really having their buy-in. So that was like what I really liked about that campaign.

It's so interesting because when you did this work, clearly you learned a lot. You got to see the potential for all of these tools. But then you decided at a certain point in your career that, yes, you could keep doing that and maybe to a certain extent you do. But then you wanted to kind of shift and use psychological targeting for good. And one of those ways is helping people to save more. Can you tell us about this project and how psychological targeting made a significant impact on helping people save?

Yeah, happy to. It's funny because I think it almost went from me feeling relatively neutral about it when I was working on the beauty retailer to me feeling absolutely terrified by the technology. And then coming back to that, there's actually something that we can do in terms of making it work for rather than against us. So the savings example is actually it's the flip side.

pretty much of the beauty retailer right so beauty retailer i try to tap into people's psychology to get them to spend more the savings example we teamed up with a fintech company called saver life so there are a company that's trying to help people with really low levels of of saving and help them save more so in our case those were people who had less than a hundred dollars in their savings account so they are really kind of struggling right and the moment that you have

levels of saving that are that low, anything that happens is potentially a lethal event, right? So your car breaks down. Now you can't get it to the shop to repair it. You can't get to work. You lose your job. So those are the people that really, first of all, struggle saving but also need help the most.

So what we did, again, in this case, we surveyed actually their users. So we gave them a personality questionnaire to get a sense of whether, again, they're more extroverted, introverted, agreeable, disagreeable, open-minded, or neurotic. And then we tailored over the course of four weeks. It was part of what they call Race to 100. So over the course of four weeks,

the users are encouraged to save at least $100, which is a lot, right? It might not sound that much to some of the listeners, but it's essentially doubling someone saving over the course of four weeks. And if they manage to do so, they are put into a lottery and they have a chance to win $2,000. So we kind of tapped into that

into that race to see, okay, there's a version of the messaging that Save a Life had been using for a long time. They had been trying to optimize and see how can they maximize the number of people that actually managed to save these $100. And what we provided was a separate category where we said, okay, we

by understanding people's psychology, can we craft messages that are even better? So let's take someone who's agreeable. So that's the personality trait of people who care about their social relationships. They're very empathetic, very trusting. They're not necessarily motivated by just putting an extra dollar in their savings account. What they're motivated by is by making sure that their loved ones are protected now and in the future, right? So that's the messaging that we centered around then. If someone was more critical and competitive, which is the

opposite side of agreeableness, we would kind of tell them, well, if you put some money to the side right now, you kind of get ahead of the competition, you're going to make and manage to win this lottery. So there's like different messaging. And what we saw, again, the same way with the beauty retailer, there was about a 60% increase in the number of people that managed to save those additional $100.

So the pattern that keeps coming up, whether it's for corporate or whether it's for this example, is you determine where someone is personality-wise, and then you couple the messaging to the type of messaging that that particular personality would resonate the most with them in terms of motivating them or helping them see what's possible. It's pretty fascinating.

It's like the way that I think about it, it feels like almost like this magic, right? Oftentimes when you look at the media, how they talk about it, but it's so fundamentally human. So every offline conversation that you have is by nature customized. So we're so good at not talking to a three-year-old the same way that we talk to our mom or a colleague or a friend.

that we don't even think about the fact that any kind of type of communication that we have is customized. So on some level, we're just really taking this away from like a small scale analog world to like the online digital world where we can do it for millions of people at the same time. Privacy and data. You know, you're getting at issues of privacy here that most of us brush away. How do we often respond to worries about privacy and our data? How should we think about them?

Privacy, I think, is front and center, right? So when we think about how data is being abused, oftentimes we think of privacy. And there's, like, I give these talks quite a lot, and I work with a lot of students. I have a class on the ethics of personal data. And when I talk about how intrusive some of the data that we generate really are, there's often someone who either says, well, I don't care about it because the perks that I'm getting are so good, or I don't care about it because I have nothing to hide.

And I think there's a bunch of problems. So first of all, I can somewhat relate to these comments. I think I for a long time was like, I don't see how we're ever going to get back privacy. If that's the case, shouldn't we just move on? But I do think that there's challenges with this notion. So the first one

that I think is important is that not worrying about your data is a real privilege that I think most people don't realize, right? Just because you don't have to worry about your data being out there doesn't mean that the same privilege is granted to everybody. And it's also not true that this is going to be the same privilege forever because

that privilege can be, can be pretty fleeting. So think about the domain that one example that I use, which just always gives me goosebumps because it's like so terrible is like the history of my own country that I grew up in Germany. That was a democracy for a long time. And at some point it wasn't right. So leadership shifted.

And what we know, and this I think is just like such a powerful example, is that when Nazi Germany was trying to invade other countries, trying to find members of the Jewish community,

is some countries had religious affiliation as part of the census. So it was very easy for the Nazis to come in, go to City Hall, find the records, find the members of the Jewish community. And there were some countries that didn't have that. Now, atrocities in the countries where the data was available were far, far higher. Now, just

Fast forward to today, they don't need City Hall anymore. They can tap into all of the data that's out there. They can make predictions about pretty much anything they want. It doesn't have to be religious affiliation tomorrow. It could be something completely different. It could be that we're suddenly intending to discriminate against fans of Manchester United.

just something trivial, it's out there. And the way that I like to kind of talk to this about my students is that data is permanent, but leadership isn't. So there's a good reason to be worried about your data being out there, even if right now you have the privilege to feel like you have nothing to hide. So that's kind of

one of the critical things when it comes to privacy. And then I think there's a second one that goes a bit deeper when it comes to agency and self-determination. Happy to talk about that as well. Yeah, please do. Please do.

So that's the, it's funny because I think when I talk about privacy, people kind of get it on a conceptual level, but it just doesn't feel as close to home, right? So it's something that is like a hypothetical in the future. And we know that the brain has a hard time dealing with these hypotheticals because technically you have to give up something in the here and now for maybe protecting your privacy and it being abused tomorrow. But the one thing that is very tangible in the here and now is the fact that

the moment that you lose your privacy and your data is being out there and people can make predictions about your psychology, also lose the ability to make your own choices. So think back to the beauty retailer example. Think back to the savings example. It's essentially once I understand your needs, preferences, motivations, fears, hopes, aspirations, and so on, I can essentially use it to push your behavior in a certain direction. Now, is it the end of the world if I manage to get you to buy a mascara or some beauty products that you otherwise wouldn't have bought?

Maybe not. Is it potentially the end of the world if I get you to vote in a different way, to think about the world in a different way, to kind of take a loan that you don't need? There are many, many decisions that we make that are...

I think we kind of care about ownership a lot more than these small ones. And this is kind of in a way goes out the window. So the moment that you lose control over your data and your privacy, you also lose control over your agency. And again, the ability to make your own choices and to control your own life. And I think that's something that people care a lot more about than just giving up privacy.

Well, and I think, Sandra, something you also write about in your book as part of this is that we can think to ourselves, well, how different is this from being exposed to certain books or certain literature? And the reality is it's quite different because the scale is so much greater and it's a very targeted personalization.

Yeah, I think that it's the combination of the two, right? So we've always had propaganda, right? So radio and mass TV was essentially kind of broadcasting to everybody, trying to convince everybody with the same message. Now,

what we have now, or then you had like the small ones, which is like village community that essentially is kind of spreading gossip by talking to each and every individual person telling their own story, but you didn't have the combination. So the problem that we have right now is we can hyper-personalize. It can be very kind of,

directly tap into your preferences and your motivations, but we can do that at scale. And that means that visibility just goes out the window, right? So when we talk about traditional propaganda, it's very effective, but it was also at least visible to everybody. So there was a way for

people to get together and say, hey, this is like something that just doesn't make sense. So there was like a way to form a resistance, if you want, which is much, much harder today, because what you see might be totally different to what I see. I might not see at all what you see. Maybe it's a different version. So this collective oversight that we used to have over something like propaganda has really gone out the window, which is one of the reasons why we woke up so late to the reality of potentially like third parties interfering with national elections. It's because just not visible.

If we want to solve the data privacy problem, you argue that one of the things we could do is to create a better data ecosystem. What do you mean? What would this look like?

Yeah. So the way that, and I've actually changed my mind on this quite a bit. So I think when I started researching this and thinking about potential solutions, I was very much focused on how do we empower people? How do we educate people? And you see this reflected actually in regulation. So most of the regulation changes.

like in the European Union, for example, or in California, mandate transparency and control. So just tell people what's happening with their data and give them control to manage it themselves. I've

kind of changed my mind on this quite a bit. I do think we need control, right? So we should be kind of, we should be able to take our data, our data and so on and so forth. But I do think it's almost like this impossible task for us to manage, right? So I do this for a living. I think about this all the time. I have a hard time keeping up with,

with the latest technology. And I also, as I hope most people, we hopefully have better stuff to do than just managing all the terms and conditions and the legal league for like 24-7, right? If you really wanted to get a sense of what exactly is it that all of the products and services that you're using, how are they tracking the data? What are they capturing? How are they using it? That would be a full-time job. You would have to kind of do this 24-7. So the

What I talk about when I talk about changing the ecosystem is essentially trying to say, how do we make it for people to do the right thing? So it can't be that it's just a burden is placed on us and everybody gets to benefit from our data. How do we make it more costly for companies to collect our data and use our data? That could be as simple as a data tax. Right now, like a company has almost every incentive to collect as much data as they want.

because it's very cheap to collect and store. Maybe they commercialize it and sell it on. Maybe they just collect it because they might need it in the future, even if they don't need it here and now. And it's very hard for us users to manage it effectively, right? Because most of it is an opt-out. So unless you actively say you don't want your data to be tracked, it's being tracked continuously. So we should flip the script and say, well, we should actively sign up for data tracking and companies should pay a

price for using the data because they profit from it. So they should pay a price for collecting it. So those are just two ways in which you can make the ecosystem a lot better for people. If they paid a price, what would you see them having to do with that money? Where would that money go? Would that go to the individual who agrees to allow the company to use their data? Would the tax go? What would happen with that? Do you see?

So I think there's different ways in which companies can actually pay a price. So some of it, like take data brokers, for example. So data brokers, the only business is to collect your data and sell it on to third parties. So in this case, there are models now that try and have you be part of that value exchange. So just as you suggested, the moment that a data broker sells your data onto someone else, you might be able to get a cut.

Now, I also think that sometimes companies might just not want to collect the data in the first place. But if you have to pay for a data point that you don't need, it's much better to simply not collect it. So that's also what I think about when I talk about this cost. So it's just like cost of collecting data that you don't need and don't create value from. And as far as a tax goes, if we did put a tax in, what would you see happening to that tax money?

So there's many different ways in which you could use it, right? So you could use it to try and, for example, invest it in research that is looking at how you can protect someone's privacy while also providing a service. And we might talk about this later on. You could directly give it back to citizens. There's many different ways. It's just essentially trying to see how do you create value for the collective by the fact that the

Data is like a collective good in a way. Data becomes valuable because we're collecting it from many different people, so we should also benefit as a collective. You talk too about data co-ops, and you reference electricians, credit unions. Say a little bit more about that.

Yeah, so that's my, I think my favorite go-to solution these days, because when we talk about regulation, and I do think regulation is important because it almost sets the stage, right? So you need certain control to be able to manage your data. It also sends an important signal as a society of what we care about, but it's typically

typically mitigating risks. So regulation is typically aiming at making sure that the average person is somewhat protected, but it doesn't really focus on how does each individual maximize the value or the utility that they get from their data. So data co-ops are these much more dynamic and fluid systems of support. So idea here is that you come together as a community of people with shared interests in data. So my go-to example that is top of mind is expecting moms.

As someone who is pregnant, you want to know pretty much every step along the way, how am I doing? How's the baby doing? What should I be doing to be healthier myself to make sure that the baby is healthy? What risk factors do I have, like from a genetic point of view, medical history, lifestyle?

and so on. Now, that's clearly not how the system is set up. You go to your doctor every four weeks, you get like a quick 20 minute visit, they see baby's okay, that's it. There's no recommendation of here's what you should be doing. Now, imagine we had like this community of

pregnant women who pull something like the genetic data, medical records, could be anything from like biometric data, so Fitbit data, lifestyle data. And we now have this data set sitting there where we can really see, okay, a woman who was born in a certain region, has a certain lifestyle, certain genetic makeup, this medical history, what she should really take care of is X.

And so it's like a pool of data that could be incredibly valuable and trying to figure out what each and every woman should be doing to make sure that she's healthy and the baby is healthy. Which means that we can generate these collective insights, but we also give direct advice to each and every member. So data co-ops is just a legal entity or like a governance structure that says, let's create...

a union that or a co-op that has fiduciary responsibility it's owned by the members so all of the members own the entity it has fiduciary responsibility so it needs to legally act in the best interest of its members and now because you have many people first of all you can get much better insights right if i have my own genetic data and my own medical history

There's nothing I can learn from this. But it also means that now because we're coming together, we actually have the resources to put management in place

who knows what they're doing and who's able to not only protect our data, but also kind of make the most of it. So there could be some kind of, let's say collaborations with startups or even established companies that try and help you figure out how you should be doing this and that build models, but that the expertise lies within management and they can sort it out for their members. So it's like just one way in which you can benefit from the collective, but also have someone help you, shepherd you through this experience.

Yeah, it's so sophisticated. When I think of social media, you can join various groups, right? But it's pretty much images and it's language, it's verbal, it's sharing information. But what you're doing is you're saying, yeah, and let's pool our data as well, which will take things to a whole other level. Yeah, that's fascinating. There are always two questions that I wrap up the interview with, Sandra. So the first one has to do with the theme of the podcast, which is curiosity. What are you most curious about today?

What I'm most curious about today? First of all, it's like I constantly get asked, like, what do I think is one of the most important qualities in leaders? And my answer is always curiosity because I think it's, first of all, keeps life interesting and makes you more empathetic to other people. I mean, it sounds like such a trite answer, but I'm super curious to see where AI is going to take us next. So I think for a long time, there was like, we could see like some advances that are done. And I'm curious to see,

If it's really that game changing technology that we currently think it is. Mostly with this focus on actually just my son growing up. I think I was always like, I have a sense of like how technology develops. I can see maybe in 30 years where this is going. But right now, like I just have like a much wider perspective.

longer time horizon. So that's what I think about a lot these days. Is there anything you're hoping for with AI? Is there something that's been on your wish list as you think out 30 years of what you'd love to see?

Yeah, so I mean, the one thing that I would love to see is, first of all, like us having control in a sense that we can manage what we've just talked about, right? There's so many ways in which having a personal AI could be incredibly helpful. There's so many limitations, like my husband and I established a center that we call the Center for Advanced Technology and Human Performance.

And the idea here is that for like centuries now, we've built all of this technology to help us overcome our physical limitations, right? So we have airplanes that get us from A to B faster. We have heaters so that we can survive in the cold and ACs that we can survive in the heat and so on. We have very few tools like this for our mental space, right? So

We have not really done much to help us deal with the fact that living in today's reality is so much more complex than what it used to be when our brains developed. So for me, this idea that we can build something that helps us accomplish the goals that we set for ourselves is

rather than just someone else taking over and deciding for us how we should live life. I think that's a very exciting prospect. And the last question is, is there anything I haven't asked? There is so much in your book, we couldn't cover it all. Is there one thing you want to leave us with or a topic that you want to talk about before we wrap up?

There's, I mean, there's maybe just one because it also ties to what I've just said. I think there's oftentimes a question of like, how do we get there, right? You can either have privacy and self-determination and agency and control, or you give us all of your data and you can have the service and convenience and personalization and all the good perks that come with it.

My hope is that we can break down this dichotomy because if it's an ease of author question, everybody's going to choose the personalization, convenience and service, right? That's how the human brain works.

I do think there's technology now that allows us to do the opposite, to actually break it off and say we have personalization, convenience and service, but also privacy and self-determination. And that's what we oftentimes call distributed learning or federated learning. So it's a way in which you don't have to send all of your data to a company to get service. So take Siri, for example, on iPhone, which is the speech recognition. So the way that Siri is trained is instead of

you sending all of your data to Apple and they process it on a central server, you have to trust them that they're not going to use your data now and in the future. What they do

They can just send the intelligence that they have developed with their language models to your phone because your phone is incredibly powerful. Your phone is so many times more powerful than the computers that we used to launch rockets into space with a few decades ago. So what they can say is they can send the intelligence to your phone. You update locally. It learns how you speech, learns how to translate that, learns how to respond. And then it sends the intelligence back to Apple.

So now what happens is that your data never leaves its safe harbor. So you can still have your privacy protected. You can make choices of what you want to send back to Apple in terms of intelligence. But you also get the service and convenience of your iPhone now being able to interpret your voice and so on and so forth. So I think it's this dichotomy that we need to break down if we really want to have the best of those two worlds.

It's a fantastic solution. And of course, one that has a lot of nuance to it, which is sometimes it doesn't allow us to shortcut. That's a great thing to leave us with. Thank you so much, Sandra. It's been such a pleasure to speak with you. Thank you so much.

Curious Minds at Work is made possible through a partnership with the Innovator Circle, an executive coaching firm for innovative leaders. A special thank you to producer and editor Rob Mangabelli for leading the amazing behind-the-scenes team that makes it all happen. Each episode, we give a shout-out to something that's feeding our curiosity. This week, it's a well-known book from 1937, Agatha Christie's Death on the Nile. It's a book that's been written for a long time,

It's a book I'd always wanted to read, and it did not disappoint. I was seeking comforting distraction, and this book gave me all that and more. She was a genius of detective fiction.