TED Radio Hour - How AI is using your data to influence you

Episode Date: May 30, 2025

Your data is used to manipulate you—and AI makes it easier than ever before. But is that so bad? Sandra Matz explains the state of psychological targeting today and how you can protect your privacy.TED Radio Hour+ subscribers now get access to bonus episodes, with more ideas from TED speakers and a behind the scenes look with our producers. A Plus subscription also lets you listen to regular episodes (like this one!) without sponsors. Sign-up at: plus.npr.org/ted See pcm.adswizz.com for information about our collection and use of personal data for sponsorship and to manage your podcast sponsorship preferences.NPR Privacy Policy

This is the TED Radio Hour. Each week, groundbreaking TED Talks. Our job now is to dream big. Delivered at TED conferences. To bring about the future we want to see. Around the world. To understand who we are. From those talks, we bring you speakers and ideas that will surprise you.

You just don't know what you're going to find. Challenge you. We truly have to ask ourselves, like, why is it noteworthy? And even change you. I literally feel like I'm a different person. Yes. Do you feel that way? Ideas worth spreading.

From TED and NPR. I'm Minoosh Zamoroti. On the show today, your data and you. But let's start with my data and me. Sandra, we have never met before, right? That's correct. And we've never spoken before. We've also never spoken before.

Okay. But I asked you here, at least, least in part, because you have psychologically personality profiled me? I have. I did some snooping around your online life yesterday night, which was extremely fun to do. This is Sandra Mats. She is a computational social scientist at the Columbia University Business School. Working at the intersection of psychology, computer science and business. Sandra is an expert on our digital footprints, the data we leave behind that companies can piece together psychological profiles of us, which can then be used to target us with ads to try to influence how we'll behave.

But as Sandra demonstrates, you don't need to be a big corporation to profile people these days. Yeah, so I've asked for support from chatyPD to turn your tweets and tweets, social media posts into a Big Five personality profile. The Big Five are five traits, openness, conscientiousness, extroversion, agreeableness, and neuroticism. So the Big Five personality traits are really what psychologists in a way default back to when we try to explain how people think, feel, and behave about the world. Oh dear.

Okay. Let's do it. What did you find? Let's do it. Here's what chat ShpD thinks about you, just based on what you post on X or former Twitter. The prediction of the model is that you score high on openness to experience. So the idea, what it says here, this person likely thrives in intellectually stimulating environments. They express curiosity about different topics from technology to ethics to personal development.

That sounds fair, but I have to say I haven't been on Twitter, which is now X, in several years. Which is actually an interesting point, because if we believe that personality is this disposition that doesn't really change as much over time, even the traces that you generated a couple of years ago oftentimes give us this relatively accurate understanding of who someone is. But let's continue. Next one up is conscientiousness.

So that's the extent to which you're organized, dependable, reliable. What Chachapiti says, moderate to... to high. And then what it says here is responsible, but also allowing for flexibility. Okay. Yep. Extroversion. Extroversion is predicted to be high. So the extent to which you enjoy talking to other people. And as I went through your tweets, there's a lot of tweets that have like words, all caps. There's lulls. There's emojis. So that's what it's picking up here. That could also be taken as yelling at people, though.

I think it very much depends on what are the words that are capitalized. Yep, yeah. Well, we'll come to agreeableness, I think, is next. Agreeableness. Also predicted to be high. Expressing warmth, empathy through expressions of gratitude, and frequently saying thank you and acknowledging contributions of colleagues. Okay.

And I think the last one is... Ready for the last one? Neuroticism? Oh, God. You know, this one. Low to moderate. Oh, really? It expresses occasional frustrations, but often balances them with humor or constructive insights.

So here's the thing, though, Sandra. Like, I was a tech reporter for a long time. And when people would talk about this idea of measuring someone's, you know, data as a window into our psychology, it always kind of sounded like a horoscope to me. So some of it is essentially, to what extent is it just a horoscope and what people want to hear? The other one is, to what extent is it? it a public persona. And it's a legitimate concern. What's interesting in the context of these predictive models is that it's very predictable. Everybody is pretending to be more extroverted.

Everybody is pretending to be a bit more neurotic. So you only get this parallel shift. And it's quite rare that someone who is extremely neurotic, kind of suddenly becomes someone who is extremely emotionally stable and overtakes someone who was on this lower end. And you could argue, it's also part of your identity, right? If I figure out, here's the person that you want to be. Think of marketing again. Knowing that and knowing here's the direction that you want to take your life in, it's as helpful when it comes to influencing your behavior, perhaps,

as it is knowing here's the person that you actually are. In all of these cases, you and people in your field can apply these discoveries to crafting messages that can target us, all online based on what you think will resonate with us emotionally. You can play into what will make people feel what you want them to feel, whether that's to get them to buy something or potentially even vote a certain way. Yeah. So the way that I think about psychological targeting is really the ability of algorithms to use the data that we generate anywhere from your social media posts, your credit card spending, the data that gets captured by your phone, to understand who

you are as a person in terms of your psychology, your needs, personality, values, and so on, and then use these insights into your psychology to potentially change the way that you think about the world, feel about the world, and also behave in the world. By now, most of us have left behind years' worth of data as we've searched, posted, and clicked our way across the internet. And despite warnings of losing our digital privacy, little has been done to legally protect us users. But is that a bad thing?

What are the real dangers and benefits of psychological targeting? And how effective is it really? On the show today, the author of Mindmasters, the data-driven science of predicting and changing human behavior, computational social scientist Sandra Matz explains the latest on what's being collected about you, how AI is accelerating this field, and the future of trying to influence what people do. So you started your career in the early 2010s by pitching psychological targeting to companies saying, I can help you understand your clients better. I mean, we were just talking about one person, me, but you're talking about, you know, entire slices of the population that a company would want to sell to better, right?

Can you tell me about that? So some of the early studies that we did was we teamed up with a beauty retailer in the UK, and their goal was to say, well, we want women to click on an ad on Facebook, go to the website, and then buy a beauty product. So they were agnostic as to which products women bought. but they were just trying to drive traffic and then eventually purchases and profits. And the idea was, why do women buy beauty products? That probably depends on something like your personality.

If you're more extroverted, maybe you're using it to stand out from the crowd to kind of go out on a weekend night and be the center of attention on the dance floor, which is how we eventually ended up framing beauty products for when we targeted extroverted women, as opposed to women who are much more introverted, that they might use purity products to make the most of the time that they have for themselves, right? In this case, the ads would show like one woman in a very quiet setting. And maybe the marketing message said, like, beauty doesn't have to shout. So for the extroverts, I think the message said 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, like a woman on the dance floor. And what we did is

we ran these campaigns on Facebook, I think over the course of a week. targeting based on prior research, right? If you go to Facebook, there's no way that they explicitly allow you to target personality trade. So you can't say, I want my ads to be sent out to people who are extroverted or introverted or neurotic or emotionally stable. By the way, not because Facebook does know how to do it. So Facebook filed a patent in 2015 on how to predict personality from their internal data. But they don't allow, they don't offer. this option to advertisers. But because we knew that certain interests and certain language was associated empirically with being extroverted and introverted, we could create these different

audiences on Facebook. And now, for the beauty retailer, the idea was if we target our extroverted ads at extroverted women, tapping into their motivation, kind of really playing into their needs, they would be more likely to buy the products. And that's exactly what we found. So about a 50 percent higher likelihood of women actually buying some of these products when the message matched their psychological traits. And super simple manipulation if you think about it. It's like it's really this one ad that they saw on Facebook once they went to the website. It looked at the same for everybody. But still in the mind of the different audiences, it must have instilled something that says, yeah, no, this is actually, it's fulfilling some of the needs that I have as a person

who's extroverted or introverted. This has been going on for a long time. We've been hearing about it. But the reason why we wanted to talk to you about it now, obviously you have a new book about it too, but it's because we've sort of reached a tipping point with this ability to collect this data with what we can do with it. Yeah, I think what's changed a lot.

And this research has been going on for almost 10 years now. But what has really changed is the fact that any one can make these predictions now. So when I started doing this research back in the day, what you needed was, first of all, a dataset that allowed you to connect someone's psychological profiles based on questionnaire responses to the digital traces that you could get your hands on, whether that's again, social media posts, fredic hard spending. And then you had to train a model that learns over time how to translate these digital traces into psychological profiles, which meant that only very few people had access to some of these predictions.

Sometimes those were researchers, sometimes those were companies. I think Cambridge Analytica is the one that's top of mind for people. But now, fast forward with the introduction and like really pretty rapid spread of generative AI, we now have these models that were never explicitly trained to take someone's digital footprints and turn them into predictions of psychological traits. And yet, because they've read the entire internet and they understand psychology on a really fundamental level, I can simply take your social media posts as I did with you yesterday and just ask chat GPT, right, an off-the-shelf language model that anyone has access to. Tell me, based on these tweets, based on what this person has written about, who do you think she is when it comes to the Big Five personality, traits, political ideology, and whatever else you might be interested in learning about that person.

When we come back, what regular people can and can't do to. protect their data. More from Sandra Mats and the world of psychological targeting. Your data and you. It's the TED Radio Hour from NPR. I'm Manushe Zamoroti. Stay with us. It's the TED Radio Hour from NPR. I'm Manushe Zamorodi. On the show today, we are covering the past, present, and future of psychological targeting. How our digital footprints are used to try and influence our behavior. We're talking to the computational social scientist Sandra Mats. Cambridge Analytica drilled deep, looking for a trove of social media. Many of us, the first time we heard about psychological targeting, was in 2018 with the Cambridge

Analytica scandal. We were able to get upwards of 50 million plus Facebook record. This was a political consulting firm that used the Facebook data of tens of millions of voters to target them with bespoke political ads. That was part of their work for a major Republican donor. Just to remind people, this was the alleged targeting of people on Facebook based on their personality types with messaging that would persuade them to vote a certain way, particularly people in swing states where that vote really mattered. But, you know, I remember covering that and there was never any proof that it had actually worked. Any evidence.

Yeah, no evidence. So where does that stand now in terms? terms of knowing whether this stuff worked. I mean, I guess if you're, the companies you were working for, they could track. Somebody click the link. They put it in their cart and paid. But, but when it comes to other things, what do we know? Yeah. And I kind of argued right from the beginning is we don't know what they actually did. We have no way of tracking where they actually worked. Even if they did something, they were probably not convincing the diehard Hillary Clinton voted to suddenly vote for Trump. Right. Oftentimes it's just like the people on the margins that we're pushing around.

But we do have like decades of scientific evidence that kind of what we call personalized persuasion. The idea that the more you know about who's on the other side and you tap into these motivations, the more persuasive you can be in terms of trying to change their behavior. And for me the most, so by now I think there's lots of evidence across different contexts that doing this kind of this exercise or going through this exercise of. making predictions of who's on the other side and then tapping into these motivations actually does have an impact on how people think and feel but and also behave. And for me, the interesting kind of thought experiment in a way is this is what we do all the time in our daily interaction with other people. Like kids figure out really, really quickly that they have to ask mom for candy in a different way than they ask your dad for candy to be successful. And as a doctor,

Or else, we don't talk to a three-year-old the same way that we talk to our boss, that we talk to our spouse, and really any conversation. Code switching. And it's fundamentally human, but it's kind of trying to say I'm adjusting both what I talk about and how I talk to the person who's on the other side. Because that's what makes communication effective. That's what makes it a lot more likely that I'm going to get what I want at the end of the day. And on some level, the way that I think about this notion of psychological targeting in the context of data is like replicating some of that at scale. And of course, then that comes with a whole bunch of like ethical questions because it's no longer bidirectional. Right. We're not doing it to one another and we're not doing it in a very transparent way. But it's just like there's people collecting this data and trying to exercise this influence in a way that's really opaque and we don't have any control over. So in terms of the ability to actually persuade someone to change their behavior, are there new ways of measuring that accuracy?

Yeah, it very much depends on the context. So if I was trying to influence someone's health behavior, for example, I would probably go with a psychological dimension that's much more closely related to what I'm trying to influence. So in the context of health, there's something that we oftentimes use, which is promoting. motion versus prevention. Are people trying to get better and healthier and exercise more, or are they just trying to prevent their health from deteriorating, and they're really trying to mitigate risk? So oftentimes what the psychologists actually add to the computer science-based predictions are like an understanding of here's the dimensions that might be relevant in a given context. But I think by now, really, there's a lot of evidence across different kind of context. So I told you about the beauty retailer example, which is trying to get people to spend more.

Some of the research that we've done in our lab is taking like an understanding of your psychology, of your big five personality traits to help people save. And there we see pretty much exactly the same pattern. So in this case, it was actually a study done with households with very low levels of savings. So in our case, it was people were less than $100 in savings. and we try to get them to save an additional $100 over the course of four weeks. And again, we're not doing magic. We're not suddenly kind of making everybody save those $100 because that's impossible.

Oftentimes it's just like the people on the margins that we're pushing around. But even in the context of saving, what we managed to do was by tapping into people's psychology, getting, I think, 60% more people to hit that saving goal that they set for themselves in the beginning. You demonstrated what anyone could do using chat GPT or any AI based on publicly available data. But can you drill down a little bit more in terms of the data that marketers can buy? Because I feel like we've kind of stopped having that conversation, at least here in the U.S. It feels like, and this pains me, having done very deep reporting and trying to get people very interested in their digital privacy. but it does feel like that ship has sailed.

So just can you walk us through what data can be collected, compiled, diced and sliced if you really tried to go get it on someone. And I think it's such an important and interesting question because oftentimes I think what comes to people's minds when we think about data and we think about potentially violations of privacy, people gravitate towards social media. when I give talks, there's oftentimes at least one person who says, well, I'm so glad I'm off Facebook and I'm off Twitter because now you can't track me anymore. And I'm like, I wish that was true. But it's clearly not, right?

Because it's just one of these data points that are so salient because the media talks about them all the time. But it kind of almost the data points that we just generate without thinking and without oftentimes being aware of are all the ones ranging from, again, your Google searches, your browsing histories, the cookies, your credit card swipes. There's almost no one now paying with cash, which is, you know, there was a reason for why the mafia is using cash because we just can't track it as easily, but nobody uses cash anymore. So whether that's your credit card or your phone. Yeah, I don't even carry a wallet anymore. Yeah, I know. And it's such an intimate data point because it means I know exactly where you are. I know your routines. I know what you're

spending your money on. Then take this a step further. Your phone is essentially, she a stalker 24-7, right? Because it is someone in your pocket who knows exactly where you are at any given point in time. And even if you've turned off location tracking, right? Like, to function, it has to know where you are. Exactly. It connects to a cell tower. And that also means that it's very easy for me to know, for example, where you live, because that's where your phone is at night when you're home. It also makes it very easy for me to figure out who's connected to whom? Because if there's two phones showing up in the same location multiple times, well, that's probably a good indication that you are connected. We also oftentimes so mindlessly sign away

access to these signals on the phone to apps when we download them and you download the weather app and that app wants to tap into your microphone and your photo gallery and your GPS records. And you're like, they clearly don't need all of this data. And yet, because we don't have the bandwidth to really go through all of these permissions and read through all of the terms, and conditions. We all too often just sign away that data. And then kind of take it even a step further, you could argue, well, worst case, if I don't want to be tracked, I'm just going to leave my phone at home. I'm not going to put on my Fitbit. I'm not going to put on that smart watch. I'm not going to spend any money with my credit card. But there's cameras pretty much anywhere.

So in New York, you can't go from lower Manhattan to up to Columbia campus, which is where I'm located. And without being seen by a camera, unless you swim through. the Hudson, which I really don't recommend. So combine that with facial recognition. And there's someone who knows what you do at any given point in time. And for me, this comparison to the offline equivalence of, again, like a stalker walking behind you, if you think about messaging, this is the mailman, opening your mail, reading that mail, or if you think about just the conversations that we have about, like, relate

a little bit more to, say, mental health, for example, that's a therapist just exposing. the conversations that they have. All of these people in a traditional offline context would go to jail. And yet somehow because the digital equivalents are much less obvious and they're not as observable, they're not as in our faces, we don't think about them in the same way. Legislation in the U.S., I mean, it's been a complete failure. Nothing has happened, right? Is that still the case?

Is Europe still way ahead of the United States? Yeah. So United States is an interesting case because it happens at the state level. If you look to California, the CCPA, which is the Consumer Protection Act, is actually very similar in the basic principles as to the European Union, to the general data protection regulation, which still is the most progressive type of regulation. And so you see differences across this different states in the U.S.

For me, the interesting part with regulation is that, and I've changed my mind about this quite a bit over the years, is they all focus on transparency and control, which those principles sound great in theory, right? The idea that I just have to tell you what's happening with your data and then I have to give you the control to delete it, to take it somewhere else, to kind of exercise what you want to do with that data. That's really great in theory. The problem is it feels much more of a burden and a responsibility that was just not equipped to take. take on. And it, in a way, lets companies and entities benefiting from our data off the hook, because it says, well, we just like the consumer had the ability to manage their data. They didn't do it and use it in a certain way. And I think it's just not a fair battle, right? Even if you understand

everything about technology, everything about data, which is already a Herculean task. So I think about this topic pretty much 24-7, and I have a hard time keeping up. But even if you did understand it, It would be a full-time job. Imagine carefully reading all of the terms and conditions, deciphering the legalese, going through your permissions. I'd much rather spend time with my friends and family than doing that. It would be a full-time job. So for me, this notion of transparency and control that these regulations are pushing

is absolutely necessary, but it's nowhere near sufficient. I mean, I'm probably the only person, well, maybe not, who when that box comes up that says, except all cookies, only necessary. Like, I am the person who digs in and turns off all. But it annoys me. I have to do it like 20 times a day. It's a such a waste of time.

And who knows if it even works. I don't even know. I know. And then it's also, they don't make it easy because I'm also one of these people who tries. And then you already spent two minutes on figuring out, okay, what should I say yes to, whatnot. Yes, exactly.

And then you get to the bottom and the button that's blinking bright red is like, accept all. And then the one that's like you can barely see, you don't even notice that it's there is like, accept my preferences. And you're like, oh, like I failed. The user design. It wants you to fail. Exactly. It wants you to fail. And it shouldn't be, it shouldn't be a trade-off.

It shouldn't be the fact that we have to go through this every single time. There should be a base level of protection where we both get the benefits without necessarily having to sign away our data. I mean, I think for people who maybe don't remember a time before they were tracked, the reasoning stands. Like, on the one hand, well, some of this tracking actually is extremely helpful to me. It's the reason why I'm wearing these pants that I have on right now, and they're my absolute favorite pair. I'm so glad that that company pushed an ad to me on Instagram. And then the other thing people say is, well, you know what? If you have something to hide, then you shouldn't be on these platforms to be.

begin with. And I know we've heard both of those for a long time, and I just want an update. Where are we now in the world in terms of those two reasons, making people feel better about the amount of data that they give away on a daily basis? Yeah. It's such a great question because I teach this class on the ethics of data in the business school. And there's always a couple of people who give me exactly those arguments, right? Well, there's all of the benefits that I get. And really, I'm not too worried because I have nothing to hide. And to the first one is like, yeah, obviously you get benefits, but that's only partially true.

So sometimes you, like Google Maps, right? Yeah, it's helpful to get from A to B faster because Google Maps knows where you are. There's many, many, many other instances where you don't really benefit from the data immediately that gets captured by companies and also then use. to make you behave in a certain way. There's this one really funny, like also terrifying example of a woman whose picture was leaked online

and it shows her with her pants down sitting on the toilet. And she couldn't figure out for her life like how did that picture end up, first of all, online? And who even took it? And it turns out that it was her Roomba. So the autonomous vacuum cleaner

capturing that picture. She had signed up to some kind of better trial. The Rumba was trying to optimize the computer vision. Pictures were sent to a contractor in Venezuela. They leaked the picture. So there's just so many instances. It's like a terrifying story.

But it's like it's funny to tell. But it's like imagine you're that person how it just feels like such an intrusion into your privacy because you didn't, she didn't even understand how the picture could have been taken. And then finding out that it's your vacuum cleaner, putting it. those pictures out there is just terrifying. So I think it's so easy for us to come up with these instances where data is being helpful that we oftentimes forget about. There are there's so, so many instances where it's not helpful. And yeah, we should be using those instances where

we benefit, but in a way that protects us from all of these other ones. And the second argument, I think, is even worse, which is the I have nothing to hide argument. Yeah. So you're still hearing that argument from your students. Yeah, absolutely. Because it's a very privileged position, first of all. So if you don't have to worry about your data being out there, that just means that you're currently in a really good spot because it doesn't apply to everybody. So if I can take your data and predict your religious affiliation, your sexual orientation, in many parts of the world, that's potentially lethal. So if you're not worried, you're in a currently in a good spot, and that can change very, very quickly. I think that's what people oftentimes forget.

even if you have nothing to worry about today, you have no idea what this is going to look like tomorrow. Because the data, once you put it out there, is accessible. And leadership might change very rapidly. And in the U.S., I think one of the best examples was Roe versus Wade. Like overnight, like half the population in the U.S., somehow I had to be worried about their data being out there. If I can track your Google searches and see are there certain things that you're looking for, Or maybe you kind of talk about some of the pregnancy-related topics. Maybe you search for information about abortion.

Then I kind of pick it up from your GPS signal that you're traveling to a certain clinic. Maybe you're traveling across state. Maybe to Google searches stop after then. Suddenly, I think people woke up, like women woke up to this idea that, no, we don't know what leadership is going to look like tomorrow. And do you even see this, I think currently with the government, like people being spooked? about Elon Musk and his team having access to all of the personal data of federal agents, which is they were probably not as worried before.

And suddenly there's this change in leadership, which means that the data might be used in a very different way. And I think for me, that's the best reason to say, even if I feel like I have nothing to hide in the here and now, I just don't know what tomorrow is going to bring. When we return, forget smartphones, try neural insights. plants, the technology being developed for the future, and how that will up the ante on what kind of data companies can collect about us and what they can do with it. Today on the show,

your data and you with computational social scientist Sandra Mats. I'm Manusia Zamorodi. You're listening to the TED Radio Hour from NPR. We'll be right back. It's the TED Radio Hour from NPR. I'm Manusia Zamoroti. And on the show today, we have been talking to computational social scientist Sandra Mottz. She is the author of the book Mindmasters, the data-driven science of predicting and changing human behavior. Sandra is an expert in psychological targeting, how businesses and governments can collect and analyze our data to profile us. Can we talk about the future of data collection and what kind of data will be out there and available? We know, we talk to a lot of technologists here on the program.

And, you know, we're already seeing this, that there are apps, mental illness apps that record people's voices and use AI to analyze them to predict when they might be going into a depressive spiral. There are all kinds of new neurotech being developed that measure brainwaves. our bodies are starting to just generate crazy amounts of data that what? How will these potentially be sold and used? Yeah. And it's funny because for me there's almost these two directions.

One is to what extent does data get more intimate if you want? Because you're absolutely right. Right now we're looking at people from the outside. I can look at your social media to get a sense of what do you care about, what is happening in your daily life, get your credit card spending. But it's really trying to penetrate the mind from the outside of making an educated guess. Right. And soon the information will be coming from the inside generated by people's bodies and brains. Yeah. And I think the direction where this is heading is that there's going to be these microbots and body, like essentially

integrated into our bloodstream, which measure what's going on in terms of like our physical health, potentially mental health because that's related to kind of indicators and like hormones and so on. And then all the way to the frontier of can we directly read from the brain and write into the brain? And there's a lot of money going into this kind of research. So Elon Musk's neural link is trying to do exactly that because we, and my husband is a neuroscientist, so I think about this world a lot because we know how to speak the language of the brain. So we know how to kind of understand what's happening, and we also know how to talk the language such that we can actually change the way that people think and behave. A lot of this work right now is done in animals, right, in rats. But we know exactly how to transfer a memory from one rat to another. So if you really kind of think about it in a dystopian way, this is, I think, where we're headed. It's like instead of just listening in from the outside, can we get to the source? And for me, the second part, that I think in this world becomes even more important when I think about data collection is right now the main argument for why we need to collect all of this data is, yeah, the only way that we can provide you with service, with convenience, with personalization, is to collect the data.

But that's no longer true. So we have these technologies now that allow you to reap the benefits of machine intelligence without necessarily having a common. company, grab all of the data and store it in a central place. You're talking about the fact that certain platforms don't take your data. They actually leave it with you, like with Google's keyboard. They're not reading what you're writing, but they have a way of anonymizing your behavior and still helping you with WordPredict, for example. It's more private. Yeah. If you're an iPhone user, for example, the way that Apple trains Siri, So the way that the speech recognition works in Apple is they don't necessarily grab all of your speech data, kind of send it to a central server that Apple operates, where they build the models.

What they do is they say, well, you all have supercomputers in your hands. Your phone is so much more powerful than the computers that we used to launch records to space with just a few decades ago. So you have the supercomputer. And what we can do is we can send the model. So our intelligence for how to recognize words in a sentence and then respond, we send this to your phone. We locally update. So your speech data lives on the phone. We kind of update the models and prove them.

So we better do a better job at recognizing your voice. But we also send the intelligence back to Apple. So everybody benefits. But you don't send the data. So what you send is the model itself. And to me, that's a total game changer. If people have the choice between, yeah, no, you can either use this convenience, but then you have to give up your data, most people will gravitate to the immediate reward, right? The brain is not really good at saying, okay, I'm happy to give up on all of the perks of Alexa and Siri.

I'm in a rush. I need it. And I'm not going to think about, well, my privacy could be violated in five years' time. So what you need is I think you need to break that dichotomy where people have to make these. trade-offs. And if you can have both, back to your point of like people saying, but I benefit so much, yeah, you benefit, but wouldn't you prefer benefiting getting the same perks, getting the same personalization and convenience, but also protecting your data? Most people would probably say yes, in an ideal world, we would have it all. And for me, these new types of AI and machine learning,

which is, it's called federated learning. And I think we're going to see a lot more of this in a couple of years. For me, it's, it kind of breaks this dichotomy. I can either have this or that. And it replaces trust with a somewhat inherently trustworthy system because I don't need to trust that company anymore to keep my data safe because they never actually get to see the raw data. They don't even need it. Yeah. So this, I think this is the ultimate goal where it oftentimes we think of it as like sovereign identity. To what extent do you manage, right, the way that we have a passport to go from place to place. And it's our. and we can take it.

Do you think this is going to happen? I'm remembering talking to Sir Tim Berners-Lee, the inventor of the World Wide Web, that he was creating this sort of a decentralized version where your data could exist, where you had control over it and existed separately from the companies that were actually using it.

I think that will take some time. There's already suggestions of how you could scale this more quickly, right? The most common one is, can we have something like credit unions be the initial starting point? It's like an institution that has people's trust already. They already deal with some personal data. Could they be some of the ones pushing for this change? But you do see, at least in some instances, so Google, for example, is using a lot of federated learning. And they make a lot of the infrastructure publicly available.

So when they develop new algorithms or new ways of setting this up, they often, have it open source so that smaller companies who don't have the resources to develop all of this from scratch can actually tap into that. And my argument is always now, because it's a good question of like, if there's no benefit to companies doing that, it's probably never going to happen. But I do think there is a big benefit for companies. So unless you're in the business of selling data, right, if you're Facebook, you're probably not excited about federated learning because you want that data so that you can monetize it. But for most companies, you're in a much better spot if you can offer the same service,

the same product, same convenience without sitting on this pile of gold that you now have to protect. So if you look at the number of data breaches, the number of hacks, the amount of money that companies have to spend on mitigating some of these hacks and the reputational costs, it's absolutely insane. So if I am a company that can say, I'm going to offer you exactly, if I'm a bank and I can offer you exactly the same service, is if I'm a mental health application that's trying to help you figure out again if you might be suffering from depression early, I'm much better of giving you these insights without sitting on this pile that other people are trying to steal. So I do think that there's really good reasons and incentives for businesses to shifting to a model where we say, no, we offer the same kind of outcome and this. same service without collecting the data itself.

I mean, that is very promising. But I think it's also kind of confusing because Google, for example, had said it was going to stop using cookies, stop tracking people as they made their way across the web. You know, you can turn off cookies on a website, but that didn't stop Chrome from seeing every website you go to. But now Google has backtracked. They don't want to give that data up. Yeah.

So, I mean, so Google has different, Google is an interesting example because it's one of these examples where you actually do see quite a bit of regression. Yeah, I would say in the progress that we've made, which comes back to my point of like, yeah, data is permanent, but leadership isn't. I recently pulled up this t-shirt that I have that says, don't be evil. We've got this covered just because Google has been really taking back so many of the things that they said they were. going to do, which I thought was actually promising. So you're absolutely right in that Google is on the one hand pushing for some of these federated learning technologies. And again, even if they're not using it in all of their products, they're still making it available to some of the other companies. And then there's also Apple. So Apple was the first one to say, we're going to get rid of some of the

third party tracking. Now, you can still make the argument of, is this just because they want to have a positive impact on the world or because they want to be the only. ones holding onto that data. So I'm not saying what it was... Or point of difference to the consumer. Yeah. So I'm not saying that it's necessarily driven. I do think, I mean, it's both, right?

So if you think about what is it, it protects the consumer on some level, but it also gives Apple a competitive edge. So I think that both of them can be true. But I do think that there's at least ways in which you can set this up now. And that wasn't true 10, 15 years ago. So even though it's probably not propagated as much as we would hope for, at least the option is there. And I think if there's a strong enough incentive for companies to do that and an easy enough way to implement it. I like this meta.

Like if you want to change behavior, it's like launching a rocket into space. You need thrust. You need motivation. So that needs companies need to be motivated to make that change. And I think there's good reasons now. But you also need to reduce friction. And that's this kind of infrastructure piece where even if you have a company that's extremely motivated to use these new types of learning and intelligence, it needs to be easy for them to implement it. Because if it's not, then it's just like this barrier is way too high.

And even with the strongest motivation, they're never going to get there. I think people now they get an ad or, gosh, so many notifications. and it's pretty blatant. Do you feel like people have just decided this is how commerce works now? I have to say I'm always surprised by how little people actually think about this stuff. So there's still so many times when I give talks and people don't even fully understand how personalized their Google search results are or how personalized. their news feeds are. For some reason, because we're never explicitly to train to think about influence in that way,

most people still don't really understand that, yeah, their newsfeed looks so, so different from the newsfeed of someone else. Or like what they see as a Google search result is so different to what I see. So even though there's like so much talk about this in the media and once in a while you kind of realize that, oh yeah, I just talked about like a trip to the Philippines. and now suddenly I get ads plastered all over the internet. Or didn't even talk about it. That's the weird thing.

Like, thought about it. Yeah. So I think those are the moments where people kind of wake up and they kind of understand that there's like this infinite. But we're so quick to forget. And we're so quick to move on with our life, which in a way is like it's a very healthy way of the brain to deal with stuff that we don't necessarily, that we think are threatening. If the only thing that we did was constantly run around, worry about how other people are influencing us, the data that was out there, that would probably chip away very quickly from our mental health. So there's something healthy about us forgetting about these instances. But it also means that we oftentimes just don't take the action that would be needed to change anything.

So I do think people understand it. I don't think it's top of mind and front and center in terms of here's something that I want to change. Let me just do one last question, which is like, you know, I always feel like people are like, well, it sounds like there's nothing much I can do about this. And, you know, I hate leaving people feeling that way. Is there one thing we can do to empower people? Obviously, it's not reading the terms of service. But is it turning, like making sure if you download an app, just turn off the microphone and location settings and all those things. I think so for me and I've become sadly a lot more pessimistic over the years just

by observing my own behavior. So I care a lot about the topic and I don't do a great job managing my data across all of the products and services that I'm using. But the one thing that I am, I think a lot more mindful of is really the phone. So I do go through the permissions when I download an app because it's much simpler than dealing with some of the other terms and conditions and most apps really don't need to tap into your pictures and into your microphone and into your GPS record. So for me, the simplest way of protecting at least some of your privacy is managing your phone a bit better. That was Sandra Mats. She is a computational social scientist and associate professor at Columbia University's Business School.

Her book is called Mind Matters, the data-driven science of predicting and changing human behavior. you can see her talk at ted.com. And I just want to leave you with a few words of specific advice. Hopefully my conversation with Sandra inspired you to dig into your settings if you haven't already. The top line is when you're using technology, I think, ask yourself two questions. Do I trust this company? And two, do I need or want this service so much that I am willing to share my personal

information. If the answer is yes, there are still some steps you can take to be a little bit more cautious and live life a little more privately. Go through your apps, turn off unnecessary access to your microphone, location, photos, or camera. To search online without being tracked, there are privacy first search engines like DuckDuck Go or Brave. If you want to text more securely, the Signal app encrypts messages end to end between signal users. Just don't invite any journalists to your group chat by accident. If you're using AI, I am on the hunt for the most privacy respecting model out there. For now, I am sticking with Claude from Anthropic, which says it doesn't use your queries to train its models unless you opt in.

And by the way, if you have shared really personal data like your DNA with the genetic testing company, you can request that your data be fully deleted. For example, 23 and me recently went bankrupt, and it is unclear what will happen to people's information. As for hiding your location, as long as your phone is on and sending cellular signals, there's no real way to do that. If you truly want to get off the grid,

leave your phone at home and head to the library. You can use the public computers there. Many allow anonymous browsing, Or better yet, grab a good old-fashioned book. Just don't check it out with your library card. Thank you so much for listening to our show. This episode was produced by James Delahousie. It was edited by Sanaas, Mesh Kempore, and me.

Our production staff at NPR also includes Rachel Faulkner White, Katie Montalione, Fiona, Giron, Harsha, Nahaada, Matthew Cloutier, and Kai McNamee. Our executive producer is Irene Noguchi. Our audio engineer was Jimmy Keeley. The music was written by Rompeteen R. Bluie. Our partners at TED are Chris Anderson, Roxanne Highlash, Alejandra Salazar, and Daniela Bala Razzo. I'm Manusse Zamorodi, and you have been listening to the TED Radio Hour from NPR.