Sandra Matz: The Personal Data Privacy Crisis

0:00:02 (upbeat music)
0:00:10 – Hi, I’m Guy Kawasaki.
0:00:12 This is the Remarkable People Podcast.
0:00:14 And as you well know,
0:00:16 we’re on a mission to make you remarkable.
0:00:20 And the way we do that is we bring you remarkable guests
0:00:23 who explain why they’re remarkable
0:00:25 and how they’re remarkable and their remarkable work.
0:00:28 And today’s special guest is Sandra Mott.
0:00:31 – She’s a professor at the Columbia Business School
0:00:35 and she’s gonna talk to us about psychological targeting
0:00:37 and changing mindset.
0:00:39 Congratulations.
0:00:43 Shipping a book is a big, big accomplishment.
0:00:45 Trust me, I know this firsthand, so.
0:00:48 – I know, it feels good when it’s out.
0:00:50 Even though I had a great time writing it.
0:00:52 So I think I probably enjoyed it a lot more
0:00:55 than what I was told by other authors.
0:00:57 So I already enjoyed the process.
0:00:59 – I have written 17 books
0:01:02 and I have told people 16 times
0:01:04 that I am not writing another book.
0:01:06 – Good luck with that one.
0:01:07 – Yeah, exactly.
0:01:08 – I’m just waiting for,
0:01:11 what would he be placing the order for number 18,
0:01:12 if that’s the case?
0:01:19 – Alrighty, so first of all, if you don’t mind,
0:01:21 let me just tell you something kind of off the wall
0:01:25 that your story about how you met your husband
0:01:27 at that speaking event,
0:01:29 that was the closest thing to porn
0:01:32 in a business book that I have ever read.
0:01:35 – And I spared you the details.
0:01:37 It’s actually a lot more to the story.
0:01:39 It’s a good one.
0:01:40 – I was reading that.
0:01:44 I said like, man, where is this going?
0:01:47 Like, is she gonna have this great lesson about how to,
0:01:50 you know, tell men to stick it and get out of my face?
0:01:52 And then I keep reading and it says,
0:01:56 oh, and the night went very, very well.
0:01:57 – What?
0:02:02 – It’s such a fun anecdote in my life,
0:02:02 how I met him.
0:02:05 So it was just like conference.
0:02:07 He was late for the people who have read the book
0:02:08 and I was like, what a jerk.
0:02:10 And I kind of had written him off.
0:02:12 And then as the night progressed
0:02:15 and I learned more about him by spying on him
0:02:18 as part of, in his place.
0:02:20 I was like, interesting guy.
0:02:22 I think I’m gonna give him a second chance.
0:02:23 And we’re married.
0:02:25 We have a kid now who’s one year old.
0:02:27 So it all worked out.
0:02:30 – And is he still meticulously neat?
0:02:32 Or, you know, was that just a demo?
0:02:34 And this is the real thing now.
0:02:35 – No, no, no.
0:02:36 So yeah, as part of the story,
0:02:39 it’s like, one of the first things I learned about him
0:02:41 is that I think he’s borderline OCD
0:02:43 ’cause he just sorts everything.
0:02:47 It’s like the person who sorts his socks by color.
0:02:48 We just moved apartments,
0:02:50 which is with a one year old,
0:02:52 not the most fun thing to do in the world.
0:02:54 And they were like boxes everywhere.
0:02:56 You could barely walk around the apartment
0:02:58 and I just opened one of the drawers
0:03:01 and he had put the cutlery, like perfection.
0:03:05 I’m like, there’s a hundred thousand boxes in this place.
0:03:08 I can barely find anything for the baby,
0:03:11 but I’m really glad that you spent at least an hour
0:03:14 perfecting the organization of the cutlery.
0:03:16 So it’s, he’s still.
0:03:18 – I hope your new place has a dishwasher
0:03:22 so he can load the utensils in the tray.
0:03:23 – Tell me about it.
0:03:24 – That’s exactly what happens.
0:03:27 Yeah, I’m not allowed to touch the dishwasher anymore
0:03:28 ’cause I don’t do it perfectly.
0:03:30 So you’re spot on, yeah.
0:03:33 (laughing)
0:03:35 – So you listeners out there, basically,
0:03:39 we have an expert in psychological targeting
0:03:43 and now she’s explaining how she had absolutely no targeting
0:03:46 in meeting her future husband, right?
0:03:48 – I think I nailed it from the beginning
0:03:51 and that his place, I looked at his,
0:03:54 him being put together and it gave me a pretty good
0:03:56 understanding I think of who he was.
0:03:58 I feel like I know what I signed up for.
0:04:02 – Okay, so this is proof that her theories work.
0:04:06 So I’ve already, you know, said this word,
0:04:08 psychological targeting twice.
0:04:10 So I would really like,
0:04:13 this is an easy question to start you off,
0:04:16 not that we got past the porn part of this podcast,
0:04:21 which is from a psychological targeting perspective.
0:04:25 What’s your analysis of the 2024 election?
0:04:26 – It’s a, I mean, interesting one.
0:04:30 So psychological targeting typically looks at individuals.
0:04:32 So it’s trying to see what can we learn
0:04:34 about the psychological makeup of people,
0:04:36 not by asking them questions,
0:04:38 but really observing what they do, right?
0:04:40 You can imagine in the analog world,
0:04:43 I might look at how someone treats other people,
0:04:46 whether they’re organized as my husband is.
0:04:48 And I think you can learn a lot by making these observations.
0:04:49 That’s true in the offline world.
0:04:51 That’s also true in the online world.
0:04:54 And I think if you just look at them,
0:04:57 presidential candidates, the way that they talk,
0:05:00 if Trump writes in all caps all the time
0:05:04 and doesn’t necessarily give it a second thought
0:05:06 before something comes out on Twitter,
0:05:08 I think that is an interesting glimpse
0:05:11 into what might be going on behind the scenes.
0:05:14 – And do you think that his campaign
0:05:18 did like really great psychological targeting
0:05:20 of the undecided in the middle?
0:05:24 Or, you know, like from an academic perspective,
0:05:29 as a case study, how would you say his campaign was run?
0:05:31 – I talk a lot about psychological targeting
0:05:33 ’cause for me, it’s interesting to understand
0:05:35 how the data translates into something
0:05:37 that we can make sense of as humans, right?
0:05:41 So if I get access to all of your social media data,
0:05:43 an algorithm might be very good at understanding
0:05:45 your preferences and motivations
0:05:46 and then play into these preferences.
0:05:48 But I, as a user, as a human,
0:05:51 can’t really make sense of a million data points
0:05:52 at the same time.
0:05:54 If I translate it into something
0:05:55 that tells me whether you’re more impulsive
0:05:58 or more neurotic or more open-minded,
0:06:00 that just kind of goes a long way in saying,
0:06:03 okay, now I know which topics you might be interested in
0:06:04 in the context of an election
0:06:08 or how I might talk to you in a way that most resonates.
0:06:11 Now, politics is an interesting case, right?
0:06:15 ‘Cause ideally a politician would go knock on every door,
0:06:16 have a conversation with you
0:06:18 about the stuff that you care about,
0:06:19 and obviously they don’t have the time.
0:06:23 So there’s, I think, a lot of potential
0:06:26 of using some of these tools to make politics better,
0:06:29 but obviously, I think the way that some of these tools
0:06:32 were introduced in the context of the 2016 election,
0:06:35 which really shows that the more dark side,
0:06:37 and I don’t know if they’re using
0:06:39 any of these tools on the campaign trail.
0:06:41 I think there are many ways in which you can use data
0:06:46 to drive engagement that’s not necessarily based
0:06:49 on predictions of your psychology at the individual level,
0:06:52 but certainly this idea that the more we know about people
0:06:54 and their motivations, their preferences,
0:06:57 dreams, fears, hopes, aspirations, you name it,
0:07:00 the easier it is for us to manipulate them.
0:07:03 – Well, in politics, as well as marketing,
0:07:06 which you bring up in your book,
0:07:09 I kind of got the feeling that what you’re saying is that,
0:07:13 you know, you psychologically target people
0:07:18 with different messages, but you could have the same product.
0:07:21 So in a sense, you’re saying that, you know,
0:07:22 yes, with the same product,
0:07:26 whether it’s Donald Trump or an iPhone or, you know,
0:07:30 whatever, a Prius, you can change your messaging
0:07:34 to make diverse people buy the product.
0:07:36 So did I get that right?
0:07:39 Or am I like imagining something
0:07:42 that’s kind of nefarious, actually?
0:07:44 – I think it depends on how you think about this, right?
0:07:48 ‘Cause the fact that we talk to people
0:07:49 in different ways all the time.
0:07:52 So imagine a kid who wants the same thing.
0:07:54 The kid wants candy.
0:07:58 The kid knows exactly that they should talk to their mom
0:08:01 in one way and that they should talk to their dad
0:08:02 in a different way, right?
0:08:03 So the goal is exactly the same.
0:08:05 The goal is to get the candy,
0:08:08 but we’re so good as humans,
0:08:10 making sense of who’s on the other side,
0:08:12 understanding what makes them tick,
0:08:14 how do I best persuade them to buy something?
0:08:17 And the same is true, I think, in politics and marketing.
0:08:19 The more that we understand where someone is coming from
0:08:21 and where they want to be in the end,
0:08:23 the easier it is for us to sell a product, right?
0:08:28 So products have the benefit that it’s not just what you buy,
0:08:28 right?
0:08:30 A lot of the times we buy products
0:08:32 because they have like this meaning to us.
0:08:33 They help us express ourselves.
0:08:35 They serve a certain purpose.
0:08:38 And if we can figure out what’s the purpose of a camera
0:08:40 for a certain person, what’s the purpose of the iPhone
0:08:43 for a camera, why do people care about immigration?
0:08:46 A certain take, why do people care about climate change?
0:08:48 Is it because they’re concerned about their kids?
0:08:51 Is it because they’re concerned about their property?
0:08:54 Then I think we just have a much easier way
0:08:55 of tapping into some of these needs.
0:08:57 And whether that’s offline,
0:09:00 when we, again, talk to our three-year-old,
0:09:02 not in the same way that we talk to our boss and our spouse,
0:09:05 or whether that’s market is doing that at scale,
0:09:07 it’s really the more you understand about someone,
0:09:11 the more power you have over their behavior.
0:09:14 – So are you saying that at an extreme,
0:09:16 you could say to like a Republican person,
0:09:20 you know, the reason why we have to control the border
0:09:22 is because of physical security,
0:09:25 where to a liberal, you might say, you know,
0:09:27 there’s a different message,
0:09:32 but in both cases, you want to secure the border,
0:09:37 one for maybe job displacement, another for security.
0:09:38 I mean, it would be different,
0:09:41 but the same product in a sense.
0:09:42 – Yeah, or the same, yeah.
0:09:44 So a hundred percent, there’s all of this research,
0:09:46 and this is actually is not my own.
0:09:48 It’s very similar to psychological targeting.
0:09:52 And in that space, it’s usually called moral reframing
0:09:53 or moral framing.
0:09:56 So the idea that once I understand
0:09:58 your set of moral values, right?
0:10:00 So there’s a framework that kind of describes
0:10:02 these five moral values.
0:10:04 The way that we think about what’s right or wrong
0:10:06 in the world, that’s how I think about it myself.
0:10:10 And some of it is loyalty, fairness, care, purity,
0:10:11 and authority.
0:10:14 And what we know is that across the political spectrum,
0:10:15 so from liberal to conservative,
0:10:18 people place different emphasis on some of these values.
0:10:20 So if you take a liberal,
0:10:22 typically they care about care and fairness.
0:10:24 So if you make an argument about immigration again,
0:10:27 that’s, or climate change does matter,
0:10:29 that’s tapping into these values,
0:10:32 you’re more likely to convince someone who’s liberal.
0:10:34 Now, if you take something like loyalty,
0:10:35 authority, or purity, you’re more likely
0:10:37 to convince someone who’s more conservative.
0:10:39 And for me, the interesting part is that,
0:10:42 as humans, we’re so stuck with our own perspective, right?
0:10:46 If I as a liberal try to convince a conservative
0:10:48 that immigration might be a good thing,
0:10:51 I typically make that argument from my own perspective.
0:10:55 So I might be very much focused on fairness and care,
0:10:56 and it’s just not resonating with the other side,
0:10:58 ’cause it’s not what they’re coming from.
0:11:00 And algorithms, because they don’t have an incentive,
0:11:03 they don’t necessarily have their own perspective
0:11:05 on the world that’s driven by ideology.
0:11:07 It’s oftentimes much easier for them to say,
0:11:11 I try and figure out what makes you care about the world,
0:11:13 what makes you think about what’s right or wrong in the world.
0:11:16 And now I’m gonna craft that argument along those lines.
0:11:19 And what’s interesting for me is that,
0:11:20 depending on how you construe it,
0:11:22 it can either be seen as manipulation.
0:11:24 So I’m trying to convince you of something
0:11:25 that you might not otherwise believe,
0:11:27 but it could also be construed as,
0:11:29 I’m really trying to understand
0:11:30 how you think about the world.
0:11:32 But I’m really trying to understand and engage with you
0:11:35 in a way that doesn’t necessarily come from my point of view,
0:11:37 but is trying to take your point of view.
0:11:41 So it really has, for me, these two sides.
0:11:46 – So I could say to a Republican is the reason why
0:11:51 you wanna support the H1B visa program
0:11:53 is because those immigrants have a history
0:11:55 of creating large companies
0:11:58 which will create more jobs for all of us,
0:12:01 which is a very different pitch.
0:12:03 – Yeah, and so in addition to the fact
0:12:05 that we can just tap into people’s psychology,
0:12:07 there’s also this research that I love.
0:12:08 I think it’s mostly done,
0:12:10 I think in the context of climate change,
0:12:12 but it’s looking at what do people think
0:12:14 the solutions to problems are,
0:12:17 and how does that relate to what they believe in anyway?
0:12:20 If I tell you, well, solving climate change
0:12:23 means reducing government influence,
0:12:25 it means reducing taxes.
0:12:26 Then suddenly Republicans are like,
0:12:28 “Oh my God, climate change is a big problem
0:12:30 “because the solutions are very much aligned
0:12:32 “with what I believe in anyway.”
0:12:33 If you tell that to Democrats,
0:12:35 they’re like, “Actually, it’s not such a big deal
0:12:37 “’cause I don’t really believe in the solution.”
0:12:41 So the way I think that we play with people’s psychology
0:12:43 and how they think about the world
0:12:45 and show up in the world just means
0:12:47 that oftentimes it gives us a lot of power
0:12:50 over how they think, feel, and behave.
0:12:54 – Another point that I hope I interpreted correctly
0:12:57 is like, you know, I’ve been trained so long
0:12:58 to understand the difference
0:13:01 between correlation and causation, right?
0:13:05 So like, if you wear a black mock turtleneck,
0:13:06 so did Steve Jobs.
0:13:08 So you should wear a black mock turtleneck
0:13:10 because you’ll be the next Steve Jobs.
0:13:13 Well, didn’t quite work out that way for Elizabeth Holmes,
0:13:16 but I think you take a different direction.
0:13:17 I just want to verify this.
0:13:21 So you don’t really discuss correlation versus causation.
0:13:24 In a sense, what you’re saying is that
0:13:29 there doesn’t need to be a causative relationship
0:13:34 if there is a predictive relationship that you can harness.
0:13:37 So I don’t know, if for some reason people,
0:13:40 we noticed a lot of people with iPhones by German cars,
0:13:41 well, that’s predictive.
0:13:44 I don’t have to understand why that’s true.
0:13:45 – Yeah, no, totally.
0:13:48 And I’ll give you an example that I think is interesting.
0:13:52 So one of the relationships that I still find fascinating
0:13:54 that we observe in the data that I don’t think
0:13:57 I would have intuited even as a psychologist
0:13:59 is the use of first person pronouns,
0:14:01 like people post on social media
0:14:02 about what’s going on in their life.
0:14:05 And I remember being at this conference,
0:14:07 it’s like a room full of psychologists
0:14:09 and this guy who was really like a leading figure,
0:14:12 Jamie Panner Baker in the space of natural language processing,
0:14:14 he comes up and he just asks the audience,
0:14:17 what do you think the use of first person pronouns?
0:14:21 So just using I, me, myself more often than other people,
0:14:23 what do you think this is related to?
0:14:25 And I remember all of us sitting at the table
0:14:27 and we’re like, oh, it’s gotta be narcissism.
0:14:30 If someone talks about themself constantly,
0:14:33 that’s probably a sign that someone is a bit more narcissistic
0:14:35 and self-centered than other people.
0:14:38 Turns out that it’s actually a sign of emotional distress.
0:14:40 So if you talk a lot about yourself,
0:14:43 that makes it more likely that you suffer
0:14:45 from something like depression, for example.
0:14:49 And now taking a step back, it actually makes sense, right?
0:14:50 If you think back to the last time
0:14:53 that you felt blue or sad or down,
0:14:54 you probably were not thinking about
0:14:56 how to fix the Southern border
0:14:58 or how to solve climate change.
0:14:59 What you were thinking about is like,
0:15:00 why am I feeling so bad?
0:15:02 Am I ever gonna get better?
0:15:03 What can I do to get better?
0:15:05 And this inner monologue that we have with ourselves
0:15:07 just creeps into the language that we use
0:15:10 as we express ourselves on these social platforms.
0:15:14 Now, the causal link is not entirely clear, right?
0:15:16 It could be that I’m just using
0:15:18 a lot more first person pronouns
0:15:19 because I have this inner monologue.
0:15:21 What you see in the language of people
0:15:23 who are suffering from emotional distress
0:15:25 is all of these physical symptoms.
0:15:28 So just being sick, having body aches.
0:15:30 And again, it’s not entirely clear
0:15:32 if maybe you’re having a hard time mentally
0:15:33 because you’re physically sick,
0:15:35 but also maybe you’re physically sick
0:15:37 ’cause you’re having like a hard time
0:15:39 with the problems that you’re dealing with.
0:15:43 So on some level, I don’t even care that much, right?
0:15:45 If I’m just trying to understand and say,
0:15:47 is there someone who might be suffering
0:15:49 from something like depression
0:15:50 who’s currently having a hard time
0:15:52 regulating their emotions?
0:15:54 I don’t necessarily care if it’s going from
0:15:57 physical symptoms to mental health problems
0:15:58 or the other way.
0:16:01 What I care about is if I see these words popping up
0:16:03 or if I see some of these topics popping up,
0:16:06 that’s an increase in the likelihood
0:16:08 that someone is actually having a hard time right now.
0:16:11 Now, I think what is interesting is that
0:16:13 the more causal these explanations get
0:16:14 and these relationships get,
0:16:17 oftentimes they’re a lot more stable.
0:16:20 So it could be that if it’s like a causal mechanism,
0:16:22 and first of all, it allows us to understand
0:16:23 something about interventions,
0:16:26 like how do we actually then help people get better?
0:16:30 And they’re also oftentimes the ones that last for longer
0:16:32 because it’s not something just the fluke and the data
0:16:34 that maybe goes this direction or the other,
0:16:36 but it’s something that is really driving it
0:16:37 on a more fundamental level.
0:16:39 So you’re absolutely right in that,
0:16:41 oftentimes when we think of prediction,
0:16:44 we don’t need to understand which direction it goes in.
0:16:49 It’s still helpful to know if you think of interventions.
0:16:51 – So at a very simplistic level,
0:16:54 could you make the case to a pharmaceutical company?
0:16:57 You know, look at a person’s social media
0:16:59 and if the person is saying aye a lot,
0:17:04 sell them some lorazeprame or some anti-anxiety drugs,
0:17:08 is it that simple?
0:17:10 – I personally probably not go to the pharma companies
0:17:13 and make that proposition, but it is that simple.
0:17:15 And again, one of the points that I make in the book
0:17:17 that is super important to me
0:17:20 is that those are all predictions with a lot of error, right?
0:17:24 So it means that on average, if you use these words more,
0:17:28 you’re more likely to suffer from emotional distress.
0:17:30 That doesn’t mean that it’s terministic.
0:17:32 There’s a lot of error at the individual level.
0:17:36 So if I’m a pharma company and I wanna sell these products,
0:17:38 yeah, on average, I might do better
0:17:39 by targeting these people,
0:17:42 but it still means that we’re not always going to get it right.
0:17:44 And then on the other side, what is interesting for me
0:17:46 is if you think about it,
0:17:48 not from the perspective of a pharma company,
0:17:50 but from the perspective of an individual,
0:17:52 I think there’s ways in which we can acknowledge
0:17:54 the fact that it’s not always perfect, right?
0:17:56 You could have this early warning system
0:17:58 for people who know, for example,
0:18:00 that they have a history of mental health conditions
0:18:02 and they know that it’s really difficult
0:18:05 once they’re at this valley of depression to get out.
0:18:07 So they could have something on their phone
0:18:08 that just tracks their GPS record,
0:18:11 sees that they’re not leaving the house as much anymore,
0:18:15 less physical activity, more user first person pronouns.
0:18:17 And it almost has this early warning system.
0:18:20 It just puts a flag out and says, “It might be nothing.
0:18:22 It’s not a diagnostic tool. There’s a lot of error,
0:18:25 but we see that there’s some deviations from your baseline.
0:18:26 Why don’t you look into this?”
0:18:29 And for me, those are the interesting use cases
0:18:31 where we involve the individual,
0:18:33 acknowledging that there’s mistakes that we make
0:18:35 and the predictions,
0:18:36 but we’re using it to just help them
0:18:39 accomplish some of the goals that they have for themselves.
0:18:48 So speaking of interesting use cases,
0:18:50 would you do the audience a favor
0:18:53 and explain how you help the hotel chain
0:18:57 optimize their offering? ‘Cause I love that example.
0:19:00 – It was one of the first projects
0:19:01 and industry collaborations that we did
0:19:04 when I was still doing my PhD.
0:19:06 And there’s many reasons for why I actually liked the example.
0:19:09 But the idea was that we were approached by Hilton
0:19:11 and we worked with a PR company.
0:19:13 And the idea of Hilton was,
0:19:16 can we use something like psychological targeting?
0:19:19 So really tapping into people’s psychological motivations,
0:19:21 what makes them tick,
0:19:23 what makes them care about vacations and so on
0:19:25 to make their campaigns more engaging
0:19:28 and then also sell vacations
0:19:30 that really resonated with people.
0:19:33 And what I like about the example is that Hilton didn’t say,
0:19:36 well, we’re just gonna run a campaign on Facebook and Google
0:19:39 where we just passively predict people’s psychology
0:19:42 and then we try to sell them more stuff.
0:19:46 They turned it into this mutual two-way conversation
0:19:47 where they said, hey,
0:19:49 we wanna understand your traveler profile.
0:19:51 And for us to be able to do that,
0:19:53 if you connect with your social media profile,
0:19:55 we can run it through this algorithm
0:19:57 that actually we don’t control.
0:19:59 It’s the University of Cambridge is doing it.
0:20:01 We don’t even get to see the data.
0:20:05 But what we can do is we can spit out this traveler profile
0:20:07 and then make recommendations
0:20:09 that really tap into that profile.
0:20:11 So it was this campaign.
0:20:15 And you can imagine that doing that increased like engagement.
0:20:18 People were excited about sharing it with friends.
0:20:21 It was essentially good for the business bottom line.
0:20:24 But it also gave, I think, users the feeling
0:20:27 that it’s a genuine value proposition.
0:20:30 So there was a company that operated first of all
0:20:32 with consent ’cause it was all,
0:20:35 it’s up to you whether you wanna share the data or not.
0:20:37 Here’s like, how does this works behind the scenes?
0:20:39 Here’s what we give you in return.
0:20:42 So it was very transparent with the entire process.
0:20:44 And it was also transparent in terms of
0:20:46 here’s what we have to offer, right?
0:20:48 It’s by understanding your traveler profile.
0:20:50 We can just make your vacation a lot better.
0:20:53 So that’s one of the reasons why I like this example a lot.
0:20:59 – Now, just as a point of clarification,
0:21:01 you said the University of Cambridge, right?
0:21:02 – Yeah.
0:21:05 – Which has nothing to do with Facebook
0:21:08 and Cambridge associates, right?
0:21:10 – With Cambridge Analytica has nothing to do at all.
0:21:13 It was funny ’cause I get mixed up with them all the time.
0:21:16 Not surprising ’cause I got my PhD there on the same topic.
0:21:20 And there was like, I mean, the idea originated there, right?
0:21:23 The idea that we could take someone’s social media profile,
0:21:25 predict things about their psychology,
0:21:29 originated at Cambridge and that’s where it was taken from.
0:21:31 But we were involved and for me,
0:21:33 it’s almost like a point of pride
0:21:36 and like a point that made me think about the ethics a lot
0:21:39 is we helped the journalists break the story.
0:21:41 So when the journalists in, first in Switzerland,
0:21:43 were working on trying to see what happened
0:21:45 behind the scenes of Cambridge Analytica,
0:21:47 we just helped them understand the science.
0:21:49 How can you get all of the data?
0:21:51 How do you translate it into profile?
0:21:55 So yeah, not related to Cambridge Analytica in any way,
0:21:57 other than trying to take them down.
0:21:59 – Okay, so I misspoke.
0:22:02 I said Cambridge associates, not Analytica.
0:22:04 So if you were for Cambridge associates,
0:22:07 if there’s such a thing out there, I’d correct myself.
0:22:08 (laughing)
0:22:09 – I’m not sure.
0:22:13 – So listen, in the United States,
0:22:15 this is a very broad question,
0:22:19 but in the United States, who owns my data?
0:22:21 Me or the companies?
0:22:23 – Well, as you might have imagined,
0:22:24 it’s typically not you.
0:22:27 So the US is an interesting case
0:22:29 ’cause it very much depends on the state that you live in.
0:22:32 So Europe, I would say has the strictest
0:22:34 data protection regulations.
0:22:37 So they very much try to operate on these principles
0:22:39 of transparency and control
0:22:42 and giving you at least the ability to request your own data
0:22:44 to delete it and so on and so forth.
0:22:46 In the US, California is the closest.
0:22:48 So California’s CCPA,
0:22:51 which is the Consumer Protection Act,
0:22:53 I can’t remember the exact name,
0:22:57 but this is like very close to the European Union principles
0:23:00 where you as a producer of data
0:23:03 and even though companies also can hold a copy,
0:23:04 you at least get to request your own data.
0:23:05 In most parts of the US,
0:23:07 the data that you generate,
0:23:09 you don’t even have a shot at getting it
0:23:10 because it sits with the companies
0:23:12 and you don’t even have the right to request it.
0:23:16 So I think we’re a very long way from this idea
0:23:18 that you’re not just the owner of the data,
0:23:23 but it’s also you can limit who else has control to add.
0:23:25 – So I live in California.
0:23:28 So you’re telling me there’s a way that I could go to Meta
0:23:31 or Apple or Google and say, I want my data
0:23:33 and I don’t want you selling it.
0:23:34 – That’s a great question.
0:23:36 So what you can do is you can request a copy of your data.
0:23:37 That’s one thing.
0:23:39 In many states, you can’t even do that.
0:23:42 You might generate a lot of medical data,
0:23:43 social media data and you,
0:23:45 even though you generated it,
0:23:47 you can’t even request a copy.
0:23:50 Now what you can do is you can go to Meta request a copy
0:23:53 and you can also request it to be deleted
0:23:55 or to be transferred for it somewhere else.
0:23:58 Now it’s still really hard to say,
0:24:00 I want to use a service and product.
0:24:02 And this is one of the things that I think makes it really
0:24:05 challenges for people to manage the data properly.
0:24:07 Because it’s a binary choice,
0:24:09 you can say, yeah, I want you to delete my data
0:24:11 and I’m not going to use the service anymore.
0:24:12 But then you also can’t be part of Facebook.
0:24:14 And yes, there are certain permissions
0:24:15 that you can play with.
0:24:16 What is public?
0:24:17 What is not public?
0:24:19 You can even play around with here’s some of the traces
0:24:22 that I don’t want you to use in marketing.
0:24:24 But typically, and this is true for,
0:24:26 I think still Meta and other companies,
0:24:28 it’s usually a binary choice.
0:24:30 Either you use our product with most of your data
0:24:33 being tracked and most of your data being commercialized
0:24:36 in a way that you might not always benefit from.
0:24:38 But you get to use the product for free
0:24:39 or you don’t use it at all.
0:24:41 And I think that’s the dichotomy
0:24:43 that’s really hard for the brain to deal with.
0:24:45 ‘Cause if the trade-off that we have to make as humans
0:24:48 is service, convenience,
0:24:51 the ability to connect with other people in an easy way,
0:24:53 that’s what we’re going to choose over privacy
0:24:55 and maybe a risk of data breaches in the future
0:24:58 and maybe a risk of us not being able
0:24:59 to make our own choices.
0:25:01 So I think there’s now ways
0:25:03 in which you can somehow eliminate that trade-off.
0:25:06 ‘Cause I think if that’s what we’re dealing with,
0:25:08 it’s an uphill battle.
0:25:11 – I need to go dark for a little bit here.
0:25:14 I read in your book about the example of Nazis.
0:25:18 And I just want to know like today,
0:25:22 could the Nazis go to Facebook, Apple and Google
0:25:25 and get enough information from the breadcrumbs
0:25:29 that we leave to track down where all the Jewish people are?
0:25:31 Would that be easy today?
0:25:33 – I think it would be incredibly easy.
0:25:35 And it’s one of these examples in the book
0:25:37 that I think is hard to process
0:25:39 and that’s why it’s so powerful.
0:25:41 I teach this class on the ethics of data
0:25:43 and there’s always a couple of people who say,
0:25:44 “Well, I don’t care about my privacy
0:25:47 ’cause I have nothing to hide and the perks that I get,
0:25:50 they’re so great that I’m willing to give up my privacy.”
0:25:53 And what I’m trying to say is that it’s a risky gamble.
0:25:55 But first of all, it’s a very privileged position
0:25:57 ’cause just because you don’t have to worry
0:25:58 about your data being out there,
0:26:01 doesn’t mean that that doesn’t apply to other people.
0:26:02 So I think in the US,
0:26:05 even the role versus way it’s a Supreme Court decision
0:26:08 to meddle with abortion rights,
0:26:12 I think overnight essentially made women across the US
0:26:14 realize, “Hey, my data being out there
0:26:15 in terms of the Google searches that I make,
0:26:18 my GPS records showing where I go,
0:26:20 me using some period tracking apps,
0:26:21 it’s incredibly intimate.”
0:26:24 And it could overnight be used totally against me.
0:26:28 So the example that you mentioned about Nazi Germany
0:26:30 is such a powerful one
0:26:33 ’cause it shows that leadership can change overnight.
0:26:34 And I care so much about it
0:26:36 ’cause I obviously grew up in Germany.
0:26:38 So it was a democracy in 1938.
0:26:40 And then the next year it wasn’t.
0:26:42 And what we know is that atrocities
0:26:44 within the Jewish community across Europe
0:26:47 totally dependent on whether religious affiliation
0:26:48 was part of the census.
0:26:50 So you can imagine if you have a country
0:26:52 where whether you’re Jewish or not
0:26:54 is written in the census,
0:26:57 all that Nazi Germany had to do is go to city hall,
0:26:58 get hold of that census data
0:27:00 and find the members of the Jewish community
0:27:03 made it incredibly easy to track them down.
0:27:06 But of course you don’t even need that census data anymore
0:27:09 ’cause you can now have all of this data that’s out there
0:27:10 that allows us to make these predictions
0:27:13 about anything from political ideology,
0:27:15 sexual orientation, religious affiliation,
0:27:17 just based on what you talk about on Facebook.
0:27:20 And even you could make the argument
0:27:22 that maybe it’s the leaders of those companies
0:27:25 handing over the data voluntarily.
0:27:28 And I think we’ve even seen in the last couple of days
0:27:31 how there’s like this political shifts in leadership
0:27:33 when it comes to the big tech companies.
0:27:35 But even if they weren’t playing the game,
0:27:38 it would have been easy for a government to just replace
0:27:41 that C-suite executives with new ones
0:27:43 that are probably much more tolerant to
0:27:44 some of the requests that they have.
0:27:46 And I think it’s terrifying.
0:27:47 And I think it’s a good example
0:27:50 for why we should care about personal data.
0:27:55 – Okay, so what you’re saying is,
0:27:57 if I look at pictures of the inauguration
0:28:02 and I see Apple, Google, Meta, Amazon up on stage.
0:28:07 And so now the government can say,
0:28:10 you know, according to Apple,
0:28:13 you were in Austin and then you landed in an SFO.
0:28:16 And then according to your visa statement,
0:28:17 you know, you purchased this.
0:28:19 And according to your phone’s GPS,
0:28:21 you went to a Planned Parenthood
0:28:23 in San Francisco, California.
0:28:26 So we suspect you of going out of state
0:28:27 to getting an abortion.
0:28:30 So we’re opening up an investigation of you.
0:28:33 That’s all easy today.
0:28:34 – I think it’s very easy.
0:28:37 And again, I’m not saying that the leaders
0:28:40 of those big tech companies are sharing the data right now,
0:28:41 but it’s certainly possible.
0:28:44 And for me, there’s like this thing that I have in the book
0:28:46 is data is permanent and leadership is.
0:28:48 And right, so once your data is out there,
0:28:50 it’s almost impossible to get it back.
0:28:52 And you don’t know what’s gonna happen tomorrow.
0:28:56 Even if Zuckerberg is not willing to share the data,
0:28:59 there could be a completely new CEO tomorrow
0:29:01 who might be a lot more willing to do that.
0:29:05 So I think that the notion that we don’t have to worry
0:29:07 in the here and now about our data being out there
0:29:10 is just a very short-sighted notion.
0:29:12 And ideally we can find a system.
0:29:15 And I think there are ways now in which we can get
0:29:17 some of these perks and some of the benefits
0:29:20 and they come from using data without us necessarily having
0:29:23 to collect the data in a central server.
0:29:24 – Okay, so if I’m listening to this
0:29:27 and I’m scared stiff because, you know,
0:29:29 yes, you could look at what I do.
0:29:31 You could look at, I went to the synagogue
0:29:34 or I went to the, you know, temple or whatever.
0:29:36 So yeah, and you’re right.
0:29:39 Any of those people could replace and who knows.
0:29:41 So then what do I do?
0:29:46 – I do think that people should be to some extent scared.
0:29:48 So I’m really trying to not say that technology,
0:29:52 it’s all, like we’re all doomed because the data is out
0:29:53 there and technology can be used too many.
0:29:55 But I think there’s like many good use cases,
0:29:58 but I do think we should be changing the system
0:30:00 in a way that protects us from these abuses.
0:30:03 And the one thing that I describe in the book,
0:30:06 which I think we’re actually seeing a lot more of,
0:30:08 but just not that many people know of,
0:30:11 are these technologies that allow us to benefit from data
0:30:13 without necessarily running the risk
0:30:15 of a company collecting it centrally.
0:30:17 So what I mean is, and there’s a technology
0:30:19 that’s called federated learning.
0:30:22 And you can imagine the example that I give
0:30:24 is take medical data.
0:30:27 So if we wanna better understand disease
0:30:30 and we wanna find treatment that work for all of us,
0:30:32 not just the majority of people who usually
0:30:34 the pharma companies collect data of,
0:30:37 but like we wanna know, given my medical history,
0:30:39 given my genetic data, here’s what I should be doing
0:30:42 to make sure that I don’t get sick in the first place
0:30:44 or I can treat a disease that’s either rare
0:30:46 or not as easily understood,
0:30:49 we would all benefit from pooling data
0:30:51 and better understanding disease.
0:30:52 Now there’s a way in which you can say,
0:30:56 instead of me sending all of this data to a central server
0:30:59 and now this entity that collects all of the data,
0:31:00 they have to safeguard it.
0:31:03 Same way that Facebook is supposed to safeguard your data
0:31:05 against intrusion from the government.
0:31:08 Instead of having to sit in the central server,
0:31:10 what we can do is we can make use of the fact
0:31:12 that we all have supercomputers,
0:31:14 but that might be your smartphone.
0:31:16 Your smartphone is so much more powerful
0:31:18 than the computers that we used to launch rockets to space
0:31:19 a few decades ago.
0:31:22 So what this entity that’s trying to understand disease
0:31:25 could do is they could essentially send the intelligence
0:31:28 to my phone or ask questions from my data
0:31:30 and say, okay, here’s like how we’re tracking your symptoms.
0:31:32 Here’s what we know about your medical history,
0:31:35 but that data lives on my phone.
0:31:37 And all I’m doing is I’m sending intelligence
0:31:40 to the central entity to better understand the disease.
0:31:42 Apple Siri, for example, is trained that way.
0:31:44 So instead of Apple going in
0:31:46 and capturing all of your speech data
0:31:49 and centrally collecting it right now,
0:31:51 Apple would be one of these companies
0:31:54 who has to protect it now and tomorrow.
0:31:56 And they just send the model to your phone.
0:31:58 So they send Siri’s intelligence to your phone.
0:32:00 It listens to what you say.
0:32:03 It gets better at understanding, gets better at responding.
0:32:05 And instead of you sending the data,
0:32:07 it essentially just sends back a better model.
0:32:10 It learns, it updates, sends back the model to Siri.
0:32:13 And now everybody benefits ’cause we have a better speech.
0:32:14 And that’s a totally different system
0:32:16 ’cause we don’t have to collect the data
0:32:18 in a central spot and then protected.
0:32:22 – But Sandra, I mean, the point that you just made
0:32:27 is that, yeah, Tim Cook may be saying that to us now.
0:32:29 We’re only sending you the model
0:32:31 and all your data is staying on your phone,
0:32:34 but tomorrow’s Apple CEO
0:32:36 could have a very different attitude, right?
0:32:38 So how do we know if they’re only still
0:32:41 sending the model right now?
0:32:42 – So I think it’s a great question.
0:32:44 And it’s funny that you mentioned Apple in that space
0:32:46 ’cause I think they’re thinking about it this way.
0:32:50 So again, I would much rather have Tim say,
0:32:53 we’re only gonna locally process on your phone
0:32:55 and that even if they change it tomorrow,
0:32:57 what I’m mostly worried about
0:33:00 is that they collect my data today under Tim Cook
0:33:02 with the intention of making my experience better.
0:33:05 They collect it today and then tomorrow there’s a new CEO
0:33:08 ’cause now that CEO can just go back into the existing data
0:33:11 and make all of these inferences that we talked about
0:33:13 that are very intrusive and we don’t want to be out there.
0:33:15 At least even if Apple decides tomorrow
0:33:18 to shift from that model to a new one,
0:33:20 that’s gonna be publicly out there.
0:33:23 So if that happens, at least people can start from scratch
0:33:25 and decide whether they still want to use Apple products
0:33:26 or not.
0:33:29 My main concern is that all the data that gets collected
0:33:31 and now leadership changes.
0:33:32 – Wow.
0:33:35 Okay, speaking of collected data,
0:33:39 you mentioned an example of a guy who applied to a store
0:33:41 and he took a personality test
0:33:46 and the personality test yielded let’s say undesirable traits.
0:33:50 And so he didn’t get that job
0:33:52 and that personality test stuck with him
0:33:56 and kind of hurt his employment in the future too.
0:33:58 So what’s the advice?
0:34:02 Don’t take the personality test or lie on the personality test.
0:34:04 What’s the guy supposed to do
0:34:07 if he’s required to take a personality test
0:34:08 to apply for a job?
0:34:11 – Yeah, and you’re really going to other dark places
0:34:13 but I think which I think is important
0:34:15 ’cause for me, this example
0:34:18 and this one is not even using predictive technology, right?
0:34:20 So this one is a guy sitting down
0:34:22 and admitting that I think in his case,
0:34:24 he was like suffering from bipolar disorder.
0:34:27 So kind of sends the score on neuroticism
0:34:29 which is one of the personality traits
0:34:32 that kind of says how you regulate emotions through the roof.
0:34:34 And because he admitted to that,
0:34:38 he was essentially almost discarded from all of the jobs
0:34:40 that had like a customer facing interface
0:34:42 ’cause companies were worried that he wouldn’t be dealing
0:34:45 and well with people who come and complain.
0:34:48 Now, the reason for why I think this example is important
0:34:52 is it just means who other people think we are
0:34:55 kind of closes some doors in our lives, right?
0:34:56 So sometimes it opens doors.
0:34:59 If someone thinks that you’re the most amazing person
0:35:01 and you absolutely deserve a loan,
0:35:04 maybe you have opportunities that other people don’t have
0:35:07 but oftentimes that the danger comes in
0:35:10 when someone thinks that we have certain traits
0:35:13 that then would lead to behavior that we don’t wanna see.
0:35:16 And now in the context of self-reported personality tests,
0:35:19 at least you have like some say over what that image is.
0:35:23 If you take it to an automated prediction of an algorithm
0:35:24 and coming back to this notion
0:35:26 that those algorithms are pretty good
0:35:28 at understanding of psychology,
0:35:29 but they’re certainly not perfect.
0:35:31 So now you suddenly live in a world
0:35:33 where someone make a prediction about you
0:35:35 based on the data that you generate,
0:35:37 you never even touched that prediction
0:35:39 ’cause you don’t even get to see it.
0:35:40 They predict that you’re neurotic
0:35:42 and maybe they even get it wrong.
0:35:44 Maybe you’re one of the people where the algorithm
0:35:46 makes a mistake and gets it wrong.
0:35:47 And now you suddenly you’re excluded
0:35:50 from all of these opportunities for jobs, loans and so on.
0:35:53 And so I think for me, this notion that there’s someone
0:35:55 who passively tries to understand who you are
0:35:59 and then takes action that again, sometimes open doors,
0:36:01 sometimes it’s incredibly helpful
0:36:03 because maybe we connect you with mental health support
0:36:07 but at other times it might also close doors
0:36:09 in a way that you don’t even have insights to.
0:36:10 And for me, that’s the scary part
0:36:12 where I feel like we’re losing control
0:36:15 over essentially our lives.
0:36:18 – Wait, but are you saying that you should refuse
0:36:22 to take the personality test or you should lie?
0:36:26 – So in the case of the personality test,
0:36:27 first of all, it’s not a good practice.
0:36:29 So as a personality psychologist,
0:36:31 the way that we think of these personality tests
0:36:34 is that it shouldn’t be an exclusion criteria.
0:36:37 So I think that what they’re meant to do
0:36:40 is to say, here’s certain professions
0:36:43 that you might just be more suited for.
0:36:44 ‘Cause if you’re an introvert
0:36:46 who kind of hates dealing with other people
0:36:49 and you’re constantly at the forefront of like a sales pitch,
0:36:51 you’re probably not gonna enjoy it as much.
0:36:53 They were never really meant to say,
0:36:55 you got a low score on conscientiousness
0:36:56 and we’re gonna exclude you.
0:36:58 It’s also very short-sighted
0:37:01 because technically what makes a company successful
0:37:04 and what makes team successful is to have many people
0:37:05 who think about the world differently.
0:37:08 So I have this recent research that’s still very preliminary,
0:37:11 but it’s looking at startups
0:37:13 and it just looks at how quickly do they manage
0:37:15 to hire people with all of these different traits.
0:37:17 So you can come together and you can say, well,
0:37:19 but I think this way and then you think this way
0:37:21 and we all bring a different perspective to the table.
0:37:23 And they’re usually more successful.
0:37:25 So this notion that companies just say,
0:37:27 here’s a trait that we don’t wanna see.
0:37:29 It is very short-sighted.
0:37:30 What we do know, and this is,
0:37:33 I promise coming back to your question,
0:37:35 is that saying that you don’t wanna respond
0:37:38 to a questionnaire is typically seen as the worst sign.
0:37:41 So there was this study where they looked at things
0:37:42 that people don’t like to admit to, right?
0:37:44 I think it was like stuff about health,
0:37:46 stuff about people’s sexual preferences
0:37:50 and saying, I don’t wanna answer the question is worst
0:37:53 and hitting the worst option on the menu.
0:37:55 So I absolutely agree that in that case,
0:37:58 the guy essentially didn’t have a shot,
0:38:00 but the problem is once it’s recorded,
0:38:02 he didn’t even get to take the test again
0:38:04 because the results were just shared
0:38:06 from company to company.
0:38:09 – So what I hear you say is lie.
0:38:14 – In this case, frankly, if it had been me,
0:38:15 I probably would have lied.
0:38:17 If I had known that this is,
0:38:19 if the company is making the mistake
0:38:22 of using the test in that way,
0:38:26 what I would recommend to people taking the test is,
0:38:29 yeah, like think about what the company wants to hear.
0:38:30 – Okay.
0:38:33 – Which is harder to do with data, by the way.
0:38:36 It’s funny ’cause oftentimes when we think of predictions
0:38:39 of our psychology based on our digital lives,
0:38:40 we think of social media and it’s always,
0:38:43 but I can to some extent manipulate
0:38:45 how I portray myself on social media.
0:38:47 That’s true for some of these explicit identity claims
0:38:50 that we think about and have control over.
0:38:51 There’s so many other traces.
0:38:53 Take your phone again.
0:38:58 The fact like my thing is that I’m not the most organized
0:38:59 person even though I’m German.
0:39:02 So I think I was expelled for a reason.
0:39:06 And I don’t organize my cutlery the way that my husband does.
0:39:09 And would I admit to this happily on a personality test
0:39:11 that like in the context of an assessment center,
0:39:12 probably not, right?
0:39:14 If someone gives me the question there
0:39:16 that says I make a mess of things,
0:39:18 would I be inclined to say I strongly agree?
0:39:20 Maybe not ’cause I understand
0:39:21 that’s probably not what they want to hear.
0:39:23 Now, if they tap into my data,
0:39:26 they see that my phone is constantly running out of battery,
0:39:28 which is like one of these strong predictors
0:39:30 of you not being super conscientious.
0:39:34 I constantly, I go to the deli on the corner five times a day
0:39:36 ’cause I can’t even plan ahead for the next meal.
0:39:37 And I constantly run to the bus.
0:39:40 So if someone was tapping into my data,
0:39:42 they would understand 100%
0:39:44 that I’m not the most organized person.
0:39:46 So there’s something about this data world
0:39:48 and all of these traces that we generate,
0:39:52 which are in a way much harder to manipulate
0:39:54 than a question on a questionnaire.
0:39:57 – Well, and now people listening to this podcast
0:40:02 are thinking, how many times did I use the pronoun I?
0:40:06 Oh my God, I’m telling people that I have, you know,
0:40:07 depression and stuff.
0:40:10 – And again, it’s not deterministic.
0:40:13 So you might be using a lot of I
0:40:15 because something happened that you want to share.
0:40:18 It’s just like on average, it increases your likelihood.
0:40:21 – Up next on Remarkable People.
0:40:24 – If I wanted to get a portfolio, a data portfolio,
0:40:25 on most of the people,
0:40:27 I would be able to get it really cheaply.
0:40:29 And that’s something that, again,
0:40:32 I think most of us or all of us should be worried about.
0:40:35 And you do see use cases where policymakers
0:40:37 are actually waking up to this reality.
0:40:40 There was this case of a judge actually across the bridge
0:40:41 from here in New Jersey,
0:40:44 whose son was murdered by someone
0:40:46 that she litigated against in the past.
0:40:48 They found her data online from data brokers,
0:40:51 tracked her down, and in this case, killed her son.
0:40:55 (gentle music)
0:40:59 – Thank you to all our regular podcast listeners.
0:41:02 It’s our pleasure and honor to make the show for you.
0:41:04 If you find our show valuable,
0:41:08 please do us a favor and subscribe, rate, and review it.
0:41:11 Even better, forward it to a friend,
0:41:13 a big mahalo to you for doing this.
0:41:18 – Welcome back to Remarkable People with Guy Kawasaki.
0:41:22 So you had a great section about how,
0:41:25 by looking at what people have searched Google for,
0:41:27 you can tell a lot about a person
0:41:30 or at least draw conclusions.
0:41:35 So do you think prompts will have the same effect?
0:41:37 Like, you know, what I asked chat,
0:41:41 GPT is a very good window into what I am.
0:41:42 – I think so, right?
0:41:44 And I don’t necessarily, I think it’s prompts.
0:41:46 I think it’s questions that we have.
0:41:47 And if you think about Google,
0:41:51 there’s questions that I type into the Google search bar
0:41:54 that I wouldn’t feel comfortable asking my friends
0:41:55 or even sharing with my spouse.
0:41:57 So it’s like this very intimate window
0:41:59 into what is top of mind for us
0:42:02 that we might not feel comfortable sharing with others.
0:42:03 Yeah, so I was actually,
0:42:04 which I thought was so interesting
0:42:06 ’cause I was part of this.
0:42:09 It was like a documentary about artistic
0:42:11 and what they did is they invited a person.
0:42:13 So they found a person online.
0:42:15 They looked at all of her Google searches
0:42:19 and then they recreated her life all the way from,
0:42:20 here’s the job that she took,
0:42:22 kind of suffered from anxiety
0:42:24 and the feeling that she wasn’t good enough
0:42:26 in the space that she was working in,
0:42:29 all the way to her becoming pregnant
0:42:30 and then having a miscarriage.
0:42:33 And they kind of recreated her life with an actress.
0:42:35 And then at some point bring in the real person
0:42:37 and the person watches the movie
0:42:39 and you can see how just over time,
0:42:43 she realizes just how intimate those Google searches are
0:42:46 ’cause what the documentary team had created,
0:42:49 the life that they had recreated was so close
0:42:50 to her actual experience.
0:42:52 And again, just by looking at their data.
0:42:54 So for me, it was a nice way of showcasing
0:42:57 that it’s really not just this one data point
0:42:58 or a collection of data points,
0:43:02 but it’s a window into our lives and our psychology.
0:43:05 – And not to get too dark,
0:43:08 but the CEO of Google was on the stage, right?
0:43:12 So what happens when generative AI takes over
0:43:17 and the AI is drafting my email,
0:43:21 drafting my responses and to take an even further step,
0:43:25 what happens when it’s my agent answering for me?
0:43:28 Then is it still as predictive
0:43:31 or will the agent reflect who I really am
0:43:32 or it throws everything off
0:43:36 because it’s not guy answering anymore?
0:43:37 – So to me, that’s a super interesting question.
0:43:40 First of all, in a way like generative AI
0:43:42 democratized the entire process.
0:43:44 So when I started this research,
0:43:48 we had to get a data set that takes your digital traces.
0:43:50 Let’s say what you post on social media
0:43:52 and maybe a self-report of your personality.
0:43:55 And then we train a model that gets from social media
0:43:56 to your personality.
0:43:59 Now I can just ask chat GPT and say,
0:44:01 hey, here are guys Google searches.
0:44:02 Here’s what he bought on Amazon.
0:44:05 Here’s what we talked about on Facebook.
0:44:07 What do you think is his big five personality traits?
0:44:10 What do you think are his moral values?
0:44:11 What do you think is again,
0:44:12 like some of these very intimate traits
0:44:14 that we don’t want to share?
0:44:15 And it does a remarkable job.
0:44:17 It’s never been trained to do that,
0:44:18 but because it’s read the entire internet,
0:44:21 it has to understand so much about psychology.
0:44:23 And then obviously taking it to the next level,
0:44:25 it’s not just understanding,
0:44:28 but also replicating your behavior.
0:44:32 And the one thing that I’m most concerned about,
0:44:33 aside from like manipulative,
0:44:35 it’s just that it’s going to make us so boring.
0:44:37 If these language models,
0:44:39 they’re very good at coming up with an answer
0:44:43 that works reasonably well, like 80%.
0:44:45 But it’s very unlikely that it comes up with something
0:44:47 like super unique that we’ve never thought about,
0:44:49 that makes us different from other people.
0:44:51 So I think what happens is that we’re just going to see
0:44:54 more and more of who the AI believes we are.
0:44:57 ‘Cause it’s essentially almost like the solidified system
0:44:59 of here’s who I think guy is,
0:45:01 and now I’m just optimizing.
0:45:02 And in the way that humans learn,
0:45:05 there’s this trade off between exploitation.
0:45:08 So that is doing the stuff that you know is good for you.
0:45:11 So if you think about restaurant choices,
0:45:14 you can either go to the same restaurant time and again,
0:45:15 because you know that you like it.
0:45:17 So there’s not going to be any surprise.
0:45:19 It’s going to be a good experience.
0:45:21 But the second part of human learning
0:45:23 and experience is the exploration part.
0:45:25 And it exposes you to risk,
0:45:26 because maybe you go to a restaurant
0:45:28 and it turns out to be not great
0:45:29 and you would have been better off
0:45:31 going to your typical choice.
0:45:33 But maybe you actually also stumble on a restaurant
0:45:35 that you love.
0:45:36 And for that, you had to take the risk
0:45:38 and explore something new.
0:45:40 And my worry with these AI systems
0:45:42 and most types of personalization
0:45:45 is that they very much focus on exploitation.
0:45:47 They take what you’ve done in the past,
0:45:47 who they think you are,
0:45:50 and they try to give you more of that.
0:45:52 But you don’t get like the fun parts of exploring.
0:45:54 It’s like Google Maps is amazing
0:45:57 at getting you from A to B most efficiently,
0:46:00 but you also never stumble upon these cute little coffee shops
0:46:01 that you didn’t know were there before
0:46:03 because you got lost.
0:46:05 And for me, that’s in a way the danger
0:46:07 of having these systems replaces.
0:46:10 Is that just gonna make us basic and boring?
0:46:15 – What if I ask the opposite question,
0:46:19 which is I want to help companies be more accurate
0:46:23 in predicting my choices, right?
0:46:25 So I wanna tell Google,
0:46:29 stop sending me world wrestling news and Google news
0:46:32 and stop telling me about the Pittsburgh Steelers
0:46:36 and stop sending me ads for trucks
0:46:38 ’cause I don’t want a truck and I don’t want a Tesla.
0:46:42 And I wanna make a case that what if you want companies
0:46:44 to understand you better, then what do you do?
0:46:47 – First of all, I think it should be an option, right?
0:46:49 So there should be two different modes for you guys
0:46:51 that says right now I’m trying to explore.
0:46:53 Right now I just wanna see something
0:46:55 that’s different to what I typically want.
0:46:57 But also now I’m in this mode
0:46:59 where I just want you to know exactly what I’m looking for.
0:47:01 And I don’t want you to send me the camera
0:47:03 even though I was not interested in the camera
0:47:05 for the last three weeks.
0:47:08 So in this case, I think what companies can do,
0:47:11 which is what they I think oftentimes don’t do enough of.
0:47:14 So it’s like having a conversation with you
0:47:17 that kind of allows you to interact with the profile.
0:47:19 Most of the time they just passively say,
0:47:20 here’s who I think guy is
0:47:23 and now we’re optimizing for their profile.
0:47:24 But if they get it wrong,
0:47:26 there’s no way for you to say no, no, no,
0:47:28 why don’t you just take out this prediction
0:47:30 that you’ve made ’cause it’s not accurate,
0:47:32 which is annoying for me ’cause now as you said,
0:47:34 you get like ads for wrestling
0:47:36 that you might not be interested in at all.
0:47:37 And it’s also bad for business
0:47:39 ’cause now they’re optimizing for something
0:47:41 that is not who you are.
0:47:43 So I think first of all, give people the choice
0:47:45 whether they wanna be in an explorer mode
0:47:47 or an exploitation mode.
0:47:50 And then second part is even within the exploitation mode
0:47:51 where we’re just trying to optimize
0:47:53 for who we think you are,
0:47:56 give people the choice and say, no, you’re wrong.
0:47:57 I wanna correct that.
0:47:59 It’s good for the user and it’s good for the company.
0:48:03 – Well, if anybody out there is listening
0:48:05 and embraces this idea,
0:48:09 I suggest you not call it exploitation mode,
0:48:12 maybe optimization mode might be a more pleasant marketing.
0:48:14 – Personalization mode, yeah, that’s true, that’s true.
0:48:16 – Personalization mode, yeah.
0:48:19 Okay, so some three short,
0:48:21 tactical and practical questions.
0:48:23 So knowing all that you know,
0:48:26 and I think we went dark a few times
0:48:28 and show people the risk here.
0:48:33 So do you use email, messages, WhatsApp or signal?
0:48:36 What do you use personally?
0:48:37 – I mostly use WhatsApp.
0:48:38 First of all, it’s encrypted,
0:48:41 but then it also just what everybody in Europe uses.
0:48:44 So I wouldn’t even give myself any credit for that.
0:48:46 And it’s funny ’cause I think the fact
0:48:48 that I’ve become a lot more pessimistic over the years
0:48:50 has to do with my own behavior.
0:48:53 So I know that we can be tracked all the time
0:48:56 and I still mindlessly say yes to all of the permissions
0:48:57 and so on and so forth.
0:48:59 So I think we just don’t have the time
0:49:02 and the mental capacity to do it all by ourselves.
0:49:04 There’s only 24/7 in a day.
0:49:06 And I’d much rather spend a meal with my family
0:49:08 than going through all the terms and conditions and permission.
0:49:11 So I think if it’s just up to us,
0:49:13 it’s an unfair battle that we don’t stand a chance.
0:49:16 – And why, of all people in the world,
0:49:18 would you not default to signal
0:49:21 because it’s encrypted both the message
0:49:24 and the meta information?
0:49:26 – It’s mostly because not that many of my friends
0:49:27 are using it.
0:49:29 So again, in this case, it would be a trade-off
0:49:31 between I get protected more,
0:49:33 but there’s also like a downside
0:49:35 because I can’t reach out to the people
0:49:36 that I want to reach out.
0:49:38 And I feel like if that’s the trade-off,
0:49:40 the brains of most people will gravitate to,
0:49:43 I’m just gonna get all of the convenience that I want.
0:49:46 – Okay, second short question is,
0:49:49 when you use social media,
0:49:52 do you use it like read only and you don’t post,
0:49:54 you don’t comment and don’t like
0:49:57 or like are you all in on social media
0:50:00 and dropping breadcrumbs all over the place?
0:50:02 – I think even if you don’t use social media,
0:50:05 even if I was completely absent from social media,
0:50:07 I would still be generating breadcrumbs all the time
0:50:09 ’cause I have a credit card and I have a smartphone
0:50:11 and there’s facial recognition.
0:50:12 I just don’t want people to think
0:50:15 that social media is the only way to produce traces.
0:50:17 Now I don’t actively use it as much,
0:50:21 but not because I know that I shouldn’t be doing it.
0:50:22 It’s just because it’s so much work.
0:50:26 I feel like I much rather have interesting offline conversations
0:50:29 that thinking about what I should post on X
0:50:31 and some of the other ones.
0:50:35 So it’s a different reason than worries about privacy.
0:50:39 – Okay, now is the logic that, yes,
0:50:41 Google knows something, Apple knows something,
0:50:44 Meta knows something, X knows something,
0:50:45 everybody knows something,
0:50:47 but nobody knows everything.
0:50:51 So the fact that it’s all sort of siloed
0:50:55 keeps me safe or is that a delusion?
0:50:56 – I think it’s probably a delusion.
0:51:00 So my argument would be that they have most of these traces.
0:51:03 So if you think of applications, again,
0:51:05 like when you download Facebook here,
0:51:07 it asks you to tap into your GPS records,
0:51:10 into your microphone, into your photo gallery.
0:51:12 You use Facebook to log into most of the services
0:51:14 that you’re using elsewhere.
0:51:17 So they have a really holistic picture
0:51:20 of what your life looks like across all of these dimensions.
0:51:21 And by the way, they also have it for users
0:51:24 who don’t use Facebook because it’s so cheap now
0:51:27 to buy these data points from data brokers
0:51:31 that if I wanted to get a portfolio, a data portfolio,
0:51:32 on most of the people,
0:51:34 I would be able to get it really cheaply.
0:51:36 And that’s something that, again,
0:51:39 I think most of us or all of us should be worried about.
0:51:42 And you do see use cases where policymakers
0:51:43 are actually waking up to this reality.
0:51:47 There was this case of a judge actually across the bridge
0:51:49 from here in New Jersey, whose son was murdered
0:51:52 by someone that she litigated against in the past.
0:51:55 They found her data online from data brokers,
0:51:58 tracked her down, and in this case, killed her son,
0:52:00 Biden signed something into a fact
0:52:03 that now protects judges from having their data out there
0:52:05 with data brokers, which makes me think
0:52:07 if we do this for judges and we’re concerned
0:52:09 that we can easily buy data about judges,
0:52:12 why not protect everybody else?
0:52:16 I think there’s a good point to be made that data on us
0:52:18 is so cheap and available from different sources
0:52:21 that even if you don’t use social media,
0:52:23 it’s easy to get your hands on.
0:52:27 – You introduced the concept in the last part of your book,
0:52:29 which I don’t quite understand.
0:52:33 So please explain what a data co-op does.
0:52:36 – Yeah, it’s one of my favorite parts of the book, actually,
0:52:38 ’cause it thinks of how do you help people
0:52:39 make the most of their data, right?
0:52:43 So we’ve talked a lot about the dark sides,
0:52:44 and I think regulation is needed
0:52:48 if we wanna try to prevent the most egregious abuses,
0:52:50 but it doesn’t really give you a way of,
0:52:53 first of all, managing your data in the absence of regulation,
0:52:54 and it also doesn’t give you a way
0:52:56 to make the most of it in a positive way.
0:53:01 So data co-ops are essentially these member-owned entities
0:53:04 that help people who have a shared interest in using data
0:53:06 to both protect it and make the most of it.
0:53:09 So my favorite example is one in Switzerland
0:53:11 that’s called MyData, and they’re focused on the medical space.
0:53:14 So one of the applications that they have
0:53:17 is working with MS patients.
0:53:19 So patients who suffer from multiple sclerosis,
0:53:20 which is one of these diseases
0:53:22 that, again, is so poorly understood
0:53:24 ’cause it’s determined by genetics,
0:53:25 and it’s determined by your medical history,
0:53:27 by your environment, and what they do
0:53:30 is they have a co-op of people.
0:53:34 So patients who suffer from MS and healthy controls
0:53:36 that own the data together.
0:53:39 So it’s a little bit similar to the financial space
0:53:41 where you oftentimes have entities
0:53:43 that have fiduciary responsibilities.
0:53:47 So they’re legally obligated to act in your best interest.
0:53:50 So data co-ops are entities that are owned by the members.
0:53:53 They are legally obligated to act in their best interest,
0:53:56 and now you can imagine, in the case of the MS patients,
0:53:57 they can pool the data,
0:53:59 they can learn something about the disease,
0:54:01 and they can also then, in this case,
0:54:03 work with doctors of the patients
0:54:05 and say, here’s something that we’ve learned from the data.
0:54:08 This treatment might be particularly promising
0:54:10 for a patient at this stage with these symptoms.
0:54:12 Why don’t you try this?
0:54:15 So the people benefit immediately,
0:54:16 and also because they’re now together,
0:54:20 they can hire experts that help them manage their data,
0:54:23 think about, well, here’s maybe some of the companies
0:54:24 that we wanna share the data with,
0:54:26 but maybe we do it in a secure place
0:54:28 that doesn’t require us to send all of the data.
0:54:30 So these data co-ops, for me,
0:54:33 is just like a new form of data governance
0:54:35 that gives us, I think of it as allies.
0:54:38 So if we have a way that we wanna use data,
0:54:40 we need other people with a similar goal
0:54:42 so that we make data, first of all, more valuable,
0:54:45 ’cause if I have my data, my medical history
0:54:47 and my genetic data as an MS patients,
0:54:49 doesn’t help me at all, I need these other people,
0:54:53 but it’s not coming together as a pharma company
0:54:56 that’s grabbing all of this data and then making profits,
0:54:58 but it’s coming together as a community
0:55:00 and benefiting directly.
0:55:02 So that’s what data co-ops are.
0:55:06 But a data co-op doesn’t exactly solve the problem
0:55:09 of all my breadcrumbs on social media and Apple
0:55:11 and all the other stuff, right?
0:55:14 This is for a very specific set of data.
0:55:17 – Agreed, so it’s not necessarily a specific set of data.
0:55:19 You could imagine in the European Union
0:55:21 where you’re allowed to pull your data,
0:55:24 you could have a data co-op of people
0:55:26 who just pull together their Facebook data
0:55:27 and now they go to Facebook and say,
0:55:31 “Hey, look, we’re all gonna leave if there’s no way,
0:55:33 “if you’re not putting in, let’s say,
0:55:34 “technology like federated learning
0:55:36 “to protect our privacy a bit more.”
0:55:38 So I do think that there is also ways
0:55:40 in which people can come together
0:55:42 and get just a lot more negotiation power at the table.
0:55:44 Then if you go to Facebook and say,
0:55:48 “Hey, I’m Guy, I wanna force you to do something different,”
0:55:49 not sure if they’re gonna listen.
0:55:52 If you suddenly have 10 million people doing that,
0:55:54 you are in a better spot.
0:55:57 – Okay, I like this idea.
0:55:59 Okay, now I understand it better.
0:56:01 Thank you very much.
0:56:04 Listen, I like to end my podcast with one question
0:56:06 that I ask all the remarkable people
0:56:08 and clearly you’ve proven you’re remarkable
0:56:10 with this interview.
0:56:13 And that would be stepping aside, stepping back,
0:56:17 stepping up whatever direction you wanna use.
0:56:20 Like, what’s the most important piece of advice
0:56:24 you can give to people who wanna be remarkable?
0:56:26 – I think it’s don’t take yourself too seriously.
0:56:28 I think some humility and the way
0:56:32 that you approach yourself and others goes a long, long way.
0:56:33 – Alrighty.
0:56:35 This is a great episode.
0:56:36 Thank you so much.
0:56:38 And I hope I didn’t go too dark for you,
0:56:40 but this is a dark subject, actually.
0:56:41 – I do think it is.
0:56:44 And I think there’s a lot of room for improvement.
0:56:46 That’s why I care about the topic so much.
0:56:48 – Alrighty, so Sandra Matz.
0:56:50 Thank you very much for being a guest.
0:56:52 This has been Remarkable People.
0:56:55 I’m Guy Kawasaki, and I hope we helped you
0:56:57 be a little bit more remarkable today.
0:57:00 So my thanks to Matt as a Nizmer, the producer,
0:57:04 Tessa Nizmer, our researcher, Jeff C. and Shannon Hernandez,
0:57:05 who make it sound so great.
0:57:08 So this is the Remarkable People podcast.
0:57:12 Until next time, mahalo and aloha.
0:57:18 – This is Remarkable People.