Skip to main content

tv   Big Data and Politics  CSPAN  August 26, 2017 9:15pm-10:28pm EDT

9:15 pm
now, a discussion on how data from social media sites analyze people's words and abilities for business purposes. we will hear from a psychologist from stanford university. this is one hour and 10 minutes. >> tonight we are going to explore one of the most important facts about our present and our future. large scale data collection and analysis has and will continue to profoundly change politics in the developing and most especially, the developer world. increasingly, private firms are assembling profiles of individuals. for example, every eligible voter in the united states, with a depth of information that would've made you a occur hoover we put and the -- j edgar hoover weep with envy. singular purpose. to craft and deliver messages
9:16 pm
that will shape the future behavior of an individual. from buying a particular brand of whitening toothpaste to discouraging a citizen from voting, to choosing one right-sharing service over another to boating one way or another on profound decisions like the u.k. brexit and presidential election. these profiles are assembled from our digital footprints. the traces of our daily lives captured and often sold and analyze. many of us do not realize we're leaving these footprints behind through our use of credit cards, web searches, online purchases, smartphone use, what we listen services, andg what we watch on cable tv. , professoronight
9:17 pm
from stanford university who will talk about facebook likes in pictures. he will help us begin to grasp how assessments like these are being used and could be used to shape our political reality. please welcome me and joining him to the stage. thanked me -- thank you for helping us spirit behind the stage. tell us about what you can help us learn from our likes and facebook pictures.
9:18 pm
>> i am a computational psychologist, which means that i am working mostly with a data. so instead of saving time with research subjects in my or usingy or running surveys, i would look at the digital footprints that you so nicely introduced before that we are leaving the hind while we use digital products and services. it is a great time to be a computational psychologist, it is a great day to be made because you guys -- we all, leave an enormous amount of digital footprints behind. we are, ibm as to my did leaving 50 megabytes of digital footprints a day. this is an enormous amount of data. if you wanted to back it up on
9:19 pm
paper as he rows and ones double-sized,per, font size 12, and you wanted to stock it up and it's just one day's worth of data, the stack of paper would be like from here to the sun four times over. so, well, hopefully you guys -- [laughter] we'll store them in the museum over here. so we're all generating enormous amount of information. and now this information, of course, contains our trail of our behaviors, thoughts, feelings, social interactions, communications, purchases, even the things that we never intended to say.
9:20 pm
i'm not sure you realize that if you type a message on facebook and then you decide, ok, it's 2:00 a.m. and maybe i drank too much wine, i shouldn't probably be sending it and you abandon the message, close the window, guess what. the message is still being saved and analyzed. and now this is not just this one platform. in most cases data is preserved even if you think that you have deleted it. in my research, my main goal is to try to take this data and learn something new about human psychology or human behavior. one of the byproducts of doing that is that i will produce models that take your digital foot prints and will try to predict your future behavior. maybe we'll try to predict your psychological traits such as personality, political views, religionocity, sexual orientation, and so on. well, what was really shocking to me when i started working in
9:21 pm
this field is how accurate those models are. so this is one shocking thing. the other shocking thing in fact, is that those models are also very difficult to -- i know that a computer can predict your future behavior. a computer can reveal or determine your psychological traits from your digital footprints. but it's very difficult for a human scientist to now understand how exactly a computer is doing it, which brings me to this black box problem which basically means that it might be that human psychologists or human scientists will be replaced one day by ai running science. but in the meantime, you basically have those models that we don't really actually understand very well how they do it but they are amazing at predicting your future behavior, your psychological traits, and so on. i worked with facebook likes quite a lot not because facebook likes are the best type of digital footprint we are leaving behind, not at all. in fact, facebook likes are not so revealing.
9:22 pm
why? because liking happens in a public space. so when you like something on facebook, you probably realize that now your friends will see what you have liked. so you wouldn't like anything really embarrassing or maybe something really boring or something that you want to hide from your friends. but now when you use your web browser or you search for something on google or you go and buy something, you basically have much less choice. you kind of will search for things that you would never like on facebook. you would visit websites that you would never like on facebook. and you would buy stuff that you would never like on facebook. like you would buy medicine that is very revealing about your health. and most of us don't really like the medicine we are taking on facebook. which basically means that if someone can get access to your credit card data, your web browsing data, your search data, records from your mobile phone, these digital footprints would be way more revealing than whatever i can do using facebook likes. so whatever findings i'm coming up with, they are just
9:23 pm
conservative estimates of what can be done with more revealing data. and you can actually see that the entire industries, not just one industry. they are moving towards basically building their business models on top of the data we are producing. and my favorite example is credit cards. how many of you guys have actually paid for the credit card recently? ok.
9:24 pm
we have few people that maybe didn't do their research on line properly. but most of us, including me, we don't pay for credit cards. now, guess what. if you are not paying for something -- think about it for a second. a credit card is just an amazing magical thing that allows you to pay for stuff without carrying cash around. it's a complicated network behind it, computers crunching data and so on. now, we're not paying for it. why? because we're the product. of a credit card company. and when you -- it's not a secret. you can go to the website of a visa or mastercard or any other credit card operator and you will see that they see themselves not as a financial company anymore. they started as a financial company and it was helping to channel payments. now they see themselves as a -- not computer. customer insights company by observing the things you're buying and when you are buying them and how much you're spending.
9:25 pm
on the individual level they can learn a lot about you but also they can see -- extract interesting information on the broader level. when they see that, you know, recently people in san francisco started buying certain things, or going to certain restaurants or what not, this is very valuable information that can be sold. so basically if you're not paying for something, you're most likely a product. so now think about your web browser that you probably didn't pay for, your facebook accounts, your web searching mechanism, and one of the gazillions of apps that you have on your phone. and now think about how much data you're sharing with the others. david: is your use of facebook likes -- i guess at the time initially, a graduate student at the university of cambridge. correct? and at the time, i believe, facebook likes were public. anyone could see your facebook likes. so did that make that kind of dataset available to you since
9:26 pm
it was just public on facebook? is that what led you to use that data? michal: yes. you're pointing out here -- not a reason why -- another reason why i'm using facebook likes, which is that i was very lucky to get a huge dataset of volunteers that donated their facebook likes to me as well as their political views, their personality and other psychological scores, and basically other parts of their facebook profiles. so back in 2006 or 2007, my friend david stillwell, started this online personality questionnaire where you could take a personality -- standard personality test and then you would receive feedback on your scores. it went viral. we had more than six million people that took the test. and half of them generously gave us access to their basic facebook profiles. so when you finish your test, we would ask you if you would be willing in return for us offering you this interesting thing, if you would be willing to give us access to your facebook profile that we would later use for our scientific
9:27 pm
research and more than six million people, in fact, took the test and we got around three million profiles, facebook profiles. at the beginning, in fact -- you know, people like to say, oh, when i graduated from high school, i already planned, you know -- ran this research 20 years later. no, it wasn't the case. in my case, i kind of stumbled by accident, kind of got into
9:28 pm
this research by accident. what happened is i was developing traditional personality questionnaires. and traditional personality questionnaires are composed of questions such as -- i'm always on time at work or i like poetry or i don't care about abstract ideas. and i had this dataset of facebook likes where basically people liked poetry or they like -- i don't like abstract ideas or i don't like to read. and what struck me is that why would we even ask people this question if we can just go to their facebook profile, look at their facebook likes, and just, you know, fill in the questionnaire for them? [laughter] >> so i started running those machine-learning -- simple machine-learning models that would take your facebook likes and try to predict what would be your personality score. and this worked pretty well,
9:29 pm
which actually was pretty disappointing for me because i spent so much time developing those bloody questionnaires and now here a computer can do the same thing in a fraction of a second for millions of people. but then we started -- we had other data in our dataset. and in so we were like,
9:30 pm
him ok, so it can predict and personality. i wonder if you can predict political views, religionocity, sexual orientation, whether your parents were divorced or not. and each time we asked this question, the computer would think for a few seconds and then say of course we can predict it. it's curiousy. it's amazing. in fact, we were pretty suspicious. so at the beginning i would rerun the models with independent pieces of data or rewrite my entire code thinking that i must be doing something wrong given that a computer can look at your facebook likes and predict with very high accuracy, close to perfect, whether you're gay or not. and people don't really like anything obviously gay on facebook. well, some do but it's actually a very small fraction of people. for most of the users running these predictions, this was really based on the movies they watched or books they read. and it looked very counterintuitive to me at the time that you could do it. now i'm a bit older and spent more time running those models. it's actual actually pretty obvious left met illustrate for you why -- maybe let me kind of try to offer you a short introduction to how those models work. it's actually pretty intuitive. look, if i told you that there is this anonymous person and they like hello kitty -- it's a brand, i'm told. [laughter] you would probably be able to figure out, if you know what hello kitty is, that this person is most likely female, young, an iphone user, and you can probably go from here and make some other inferences about
9:31 pm
her. and you will be actually very correct. 99% of people who like hello kitty are women. so you don't need computer algorithm or rocket scientist or even a computer scientist to basically make inferences of this kind. most of your facebook likes or most of your purchases on amazon or most of the locations that you visited with your phone recording it or most of the search queries that you put in google are not so strongly revealing about your intimate traits. but it doesn't mean they are not revealing at all. they are revealing some of them to a very tiny degree but they are still revealing. so the fact that, let's say, you listen to lady gaga 30 times yesterday, it's not only a bit weird, it does also show us something about your musical tastes. but it also reveals to
9:32 pm
some tiny me extend, in your religion, tastes, traits, behavior. the amount of information that is there is really tiny. so for a human being, this is useless. it now, what a computer aggregatecan do is information over thousands of digital footprints you are leaving behind two arrive at a very accurate prediction of what your profilers. this is basically the paper i published in 2013 and i was very excited about the promises of the technology and i am still excited about the promises of improve oursed to lives and so many ways. if you do not leave me, think about what if i or facebook news
9:33 pm
feed, which is so engaging that people spend two hours a day if i remember correctly looking at them. they don't look at it because it is so boring and unpleasant. they look at it because the ai behind it made a very accurate prediction about what your character is in adjusted the message to make it engaging. downsides are also which i'm sure we will be talking more about today. basically, the paper i published some pressled, quite coverage. but most of the press coverage was so cool you can predict whether someone is, i don't know republican from their facebook likes. i was like, no, wait, there are tremendous consequences for
9:34 pm
the future. and you're like, no, no, no. likes but thatit is as far as we go. howresting, this is policymakers and companies took notice. two weeks after those were published, facebook changed the preview policies and such a way that likes word no longer public. before 2013, to go as marge roy published the paper, before that likes where public for everyone to see. so i should not even have to be your friend on facebook to see what you like. but our work shows that by seeing what you like, i can also determine your sexual orientation, political views, and other intimate traits that people are not happy to share. i think it is a great thing that facebook took notice to preserve your privacy and switch that
9:35 pm
off. governments took notice and they started working on changing the legislation to protect their citizens from some of the shortcomings of this phenomenon. to talk about some of those things. about to hear you talk how private firms are using this data analytics. to shift voting results in one way or another in micro-targeting messages defined by its intended persuasion rather than by its accuracy. one of these firms is cambridge analytic, part of an interconnected set of private
9:36 pm
because much of what we know about this political story is due to investigative journalism, especially by the guardian newspaper, without i went provide our audience a quick review. alytica until august 13 had steve bannon as its secretary. one of the most successful quantitative hedge fund managers was also, and a manager of the trump campaign. steve bannon left his positions when he became manager of the trump campaign and of course is now chief strategist to president trump. analytica employed
9:37 pm
social media data mining to develop a dossier on every u.s. voter. first used by the ted cruz campaign and later by the trump campaign to micro-target their messaging and influence voters. relatedly, a canadian firm been aaggregate iq has central consultant for this kind of thing with a few organizations that pushed for the u.k. exit vote. it appears to be the owner of us time magazine reported yesterday that u.s. congressional investigators were currently analytica cambridge in the course of their investigation of russian activity because they may have
9:38 pm
used techniques like those used analytica in their investigation of american data. how should we think about cambridge analytica claims and terms of their trump campaign research? michael: no is a lot there. first of all, we do not know how ticactive cambridge analy was. when you look at them themselves, they start by saying how amazing and efficient they were but then when they realized that governments are getting interested and maybe when they realized that some things they not be anould
9:39 pm
entirely legal, they someday change there should be all and now they say, it did not work massive scale. outrage,emember any especially on the left side of the political spectrum.
9:40 pm
spenty clinton not only three times more money thing on donald trump on doing personalized targeting on social media, but also hired way smarter people, in my opinion. yes, she lost, but she did not lose because trump was using some kind of magical methods. the difference in the outcome was mostly caused by something house. me, can datae ask analytics and personal marketing win the election, the answer is yes and no. fact of lifet is a when you are running a clinical campaign like television spots and offensearticles and putting ads in the paper. using it.eryone is
9:41 pm
it is not giving anyone any unfair advantage and the only unfair advantage i can think of is of barack obama who used a nonny massive scale. think people, because we as human kind of like to talk about the negative. it is great we focus on the negative because it is clearly a great psychological trait because it allowed us to be pretty successful as a species to some degree, but let's put aside focusing on the recent. let's think about advantages of politicians being able to personalize their message. there is still some advantage people have an notice. youis, if i can talk with
9:42 pm
guys one-on-one, because i'd is what i can do. i can use algorithms to talk with you one-on-one about things that are most relevant to. rhythms toese out try to understand your character, your interests, your dreams, and your fears to make the message more interesting and relevant to you. which, first of all, has one outcome. messages became more important. in the past, i could say, yes we can. we can spend gazillions of dollars running it on every stations. not have theld ability to communicate with people one-on-one. so, i had to settle down on a message that was a common denominator. a message that would reach the broadest range of the population, yet was not particularly aimed at anyone.
9:43 pm
now, i can talk to you about things that are relevant to you. and someone else with something relevant to them. which means the content of the political program became more important. this, in turn, has two important outcomes. if i make a message more relevant to you, you became more engaged in politics. this is great when we have more voters interested in the messaging they are getting. we just have more people engaged in politics and this is just great i believe. second of all, it also makes politicians think, what is important for you, david. in the past i could say yes, we can. now i had to hard about what is important to you and taking -- thinking about it would perhaps make me update my beliefs about what my political agenda should be. specifically when it comes to minorities, if i can broadcast my message on a tv, i focus on
9:44 pm
the majority interest because this is the logical broadcasting. if i can talk to people one on one, i can develop interest as a politician in what they have to say and what their interest and in.hat they are interested more engagement and more importance of the message over emotional slogan. the second change -- we have seen it in the recent election, politicians like donald trump but also bernie sanders bringing in people into the political process that traditionally were not so interested, disengagement politics because they thought there was nothing in it for them. i believe even if you disagree with those new people that were brought to the political process, you still recognize it's great for democracy when people are engaged in politics. the other outcome of personalized marketing is that if i can talk to you about
9:45 pm
things that are relevant to you, i don't need to show you this slogan 20 times, i can just have one conversation with you through social media. i'm exaggerating now but what i'm saying is i don't need to spend so much money on communicating because i can communicate with you using more relevant messages, which is another great outcome for democracy, which is decreasing the entry cost into politics. we have seen it in american elections, we have seen with donald trump and bernie sanders, who were not establishment candidates. they didn't have much money compared to other politicians in the race. i also didn't have experience in running big scale presidential campaigns. they had to basically do it on a budget, in a new way. they started talking to people directly about the issues that people cared about, which means at the end of the day we have those two candidates, and you
9:46 pm
may be stuck with one or both of them, but i believe it is great for democracy that you don't need a lot of money and backing from the industry and establishment to be a part of the political process. we have seen the same with brexit. i'm not a big fan of brexit myself. it seems even people campaigning for brexit are not fans. but at the time of the campaign, these ragtag political militia with no experience or money but they were able to enter the political process. david: i'd like to challenge you a little bit on this characterization. saying that this micro-targeting is producing messages that are
9:47 pm
more relevant, the relevance of these messages is they are persuasive quality, it has nothing to do with their accuracy, nothing to do with their positive contribution to civic discourse. the be-all and end-all of the messaging is persuasion. what i see as the great danger in this sort of work is actually an assault on consensus reality. if there is pervasive and endemic and ceaseless micro-targeting, especially of messages that are intended to be persuasive politically, you will end up -- a result of that, i fear, is that people will end up living in essentially different informational worlds and will no longer have the ability to agree on what reality is, so how can you have a democracy if you cannot agree on what is real or not real? [applause]
9:48 pm
david: that is not against you. michal: i am glad you guys are clapping, because you know where there was a perfect consensual reality, you know where there was perfect consensus for reality, soviet russia. perfect, one tv station government approved, one pravda, the truth newspaper government approved. if you knew something that other people did not know, you probably ended up in the concentration camp. so i am glad you guys clapped, because it was a bit ironic to me. i'm sorry. let me actually -- why? because, thank you. [applause] michal: it is actually, no, i reacted too aggressively, but maybe because i was born behind the iron curtain in a country where we had perfect information bubble.
9:49 pm
you know, the bubble was exactly the same for everyone. america luckily, tried to break the information bubble by dropping printing presses into poland so the underground press could produce and spread some, some fake news, as the communist regime would call it. so yes, basically we have the myth now, which i believed for the longest time, everybody is telling you we are living in echo chambers and information bubbles and systems are giving us information crafted for us that somehow is making society worse, because we now know different things about the bubbles not really overlapping. i was thinking, this is very bad because everybody says it is
9:50 pm
bad, but then i actually in the context of the findings started thinking about it. look, i strongly believe, and i did some research on that as well, that it is not bad at all. first of all, there is absolutely no evidence -- let me start differently. humans are just, we are just destined to live in an echo chamber. it is called confirmation bias in psychology and it basically means that if you have a worldview and get new information, and it contradicts your worldview and is not compatible, you will have a tendency to reject it. you don't always reject it, but you like information that confirms your worldview a bit more. it is not a bad trait to have. if you did not have it, you would be crazy because any new information you got, you would think -- ok, interesting. lizard people running the world,
9:51 pm
very interesting. no, luckily we have this, like our brains like to have consistent mental structures there. so one of the effects is we will have a preference for information that confirms our views. and the systems recognize that, so they will give us more information that will confirm our views. so we are all kind of living in echo chambers. but those echo chambers today are way larger than they have ever been in the history of humankind. let me give you an extreme example. if we are born in a little town anywhere in the world, let's say america, 100 years ago. you only knew what your priest or your teacher or your community told you. guess what? 99% of that was fake news. now you are living in a society where there is not only other people like teachers, and your friends at work, social networks
9:52 pm
are much larger, journalists are trying to constantly give you some additional information, get you out of your bubble. and algorithms are entering the race and try to give you personalized views of the world. what happens is the bubble is expanding. in the process of the bubble expanding, they also stop perfectly overlapping. if you live in your bubble, by the way it is larger than whatever bubble a human has occupied ever before now. i live in my bubble, my growing bubble, and our bubbles grow apart, which makes me aware, he is in a different information bubble. overlap, overlap, tragedy, tragedy. if you live in soviet russia, you are never able to look at people with different information bubbles. so you never thought about it.
9:53 pm
you just kept living in a small, tiny echo chamber. which brings me to a related topic of fake news. i did not see any evidence, and i encourage you to send me some if you have any, that we are somehow surrounded by more fake news today than we were surrounded by 20 years ago. quite the opposite! [laughter] michal: i will take that. quite the opposite, the amount of valid information that we have, not only as a society but also each individual person in the room, is again away larger than whatever people had in the past. and one of the outcomes of us having so much valid information, and also being able to access any information humanity has ever produced, including sensitive data, just with one click of the mouse on on the internet, what happens is if you have fake news or you heard something, you heard a story, you can quickly debunk it. what happens in the present society is we have gotten so quick at debunking stories that you basically hear the phrase fake news, fake news, all the time, which creates an impression of -- wow, we are surrounded by fake news. of course we are, there is a lot of fake information around, but it is way less than ever in the history in the past. if someone has evidence to the contrary, i would love to receive it. please send it to me.
9:54 pm
david: i think, from my perspective having those bubbles overlap is a really important thing. but let's turn, let's pivot and talk about some of your very interesting work that is going on right now working with photographs of people's faces. particularly facebook profile pictures. in recent talks you say that these neural networks, the machine learning algorithms you employ, can identify a man as straight or gay by pictures of his face, or identify a woman as extroverted or introverted by a picture of her face. how is that possible? michal: yes, in my recent work i got interested in seeing what the algorithms can predict, if they can determine intimate traits based on a still image of their faces. when i mention it, especially when i mention it to colleagues in academia they say, oh my god. we lost you. michal: and it is very bad.
9:55 pm
9:56 pm
but what i think they keep forgetting is in fact humans are great at predicting intimate traits, determining intimate traits of other people based on their faces. we are so good at it that we kind of do not realize we are doing it all the time. let me give you an example. gender is a pretty intimate ttroublet you have no gender by a quick look at their face. emotions are intimate psychological traits. we can very quickly detect other people's emotions just buy a quick glimpse on their face, even when they are trying to
9:57 pm
hide that emotion we are still able to detect it. think about race. think about genetic issues, like there are certain disorders we can recognize by looking at some -- someone's face. another genetic expression of genes is recognizing that somebody is the child of somebody else. when you say, you look like your father, you are actually saying i see through your face your genome. i see it is similar to the genome of this other guy, and he seems to be your father. it is like a magical skill. now, humans are good at predicting those kind of intimate traits, but we are not good at predicting other intimate traits like personality or sexual orientation, or political views. this might mean that there is no information about your political views on your face, but it also could mean that your brain has not evolved or do not learn to extract this information from a human face. it might be that sexual orientation is as clear, clearly displayed on your face, as your gender, is just the human brain does not do well at revealing
9:58 pm
it. guess what, it seems that computer algorithms can basically do it pretty well. in terms of sexual orientation, more than 90% distinction between straight and gay men. and personality predictions, those are comparable with a short personality questioner. how to computers do that? well, actually it is no easy answer, because it is a black box model that i would love to understand, but my brain is basically too weak to do it. it seems computers are great at looking at giveaways that are on our faces, then putting together hundreds of little pieces of signals into an accurate prediction of a given trait.
9:59 pm
much like facebook likes. and there is some evidence of that. one example is if you take one picture of an extroverted person, you would not know if they are extroverted or not, but if you take 100 pictures of extroverted people and you overlay them, make an average face, suddenly a human being is able to very accurately distinct between an extrovert and introvert. it seems the information is there, but it is too weak for a human being to detect. but for a computer it is pretty easy. david: with the networks you are using to do this classification work, they are explicitly and fundamentally like non-explanatory. they do things, right? you train a classifier with pictures and you tell it, this, yes or no, yes or no, and then it starts to be able to identify cats, pictures or whatever. it is just a trained classifier, so whether it is picking up in
10:00 pm
the case of your work like on straight or gay male faces, is it detecting subtle social and cultural information in the photography, or is it picking up, you know, something that is actually of the face? there is no way to know. right? -- is it detecting subtle social and cultural information in the photography, or is it picking up, you know, something that is actually of the face? there is no way to know. right? michal: we can test it by
10:01 pm
changing morphological -- by taking a picture in photoshop in some elements like making your nose longer or broader or whatnot, and see what the computer does, if it changes its mind about you. by doing that, you can basically experiment with the computer and reveal what features of the face are predictive of being extroverted or introverted, gay or straight. at the end of the day, we could have a separate conversation just about that, what makes your face look extroverted or gay, but at the end of the day it does not matter whether those are morphological features or whether it is something cultural or environmental that allows the prediction, because at the end of the day we see that the prediction is possible. when people say, what if i try to maybe change my hairdo and fool the algorithm. guess what, as soon as people change their hairdos, the computer would move on to you something else in the prediction. what i think is crucial and key in those studies showing you can take a human face and predict their traits is the following. we can talk about introducing
10:02 pm
good policies and introducing good technologies that will protect our data from being used to reveal our intimate traits. but at the end of the day, it is difficult to imagine this working out. why? first of all, it is difficult to police data. data can move across borders really quickly and can be encrypted, stolen, without a person even noticing. organizations such as nsa cannot reliably defend data, so how can we expect that we as individuals can protect our own data. which basically means in the future there will be more data about us out there. algorithms are also getting better at turning the data into accurate predictions of our intimate traits, our future behavior, and so on. moreover, even if you gave me full control over my data, there is still a lot i want to share.
10:03 pm
i want the discussion to be available to people after, i want my blog to be read by people, i want my twitter feed to be public, and most importantly i want to walk around without covering my face. [laughter] basically, my conclusion is, going forward it is going to be no privacy whatsoever. the sooner we realize that, the sooner we can start talking about how to make sure that the privacy world is still habitable and safe and nice to live in. now when people say, how can you say that we should work on technology to continue protecting our privacy, but yes let's do it, but realize it is a distraction, because it makes people believe that we can have privacy in the future.
10:04 pm
this believe, -- the disbelief, -- this belief, i believe now, this belief is completely wrong. to stress the importance of that, we are having a conversation about how invasion of privacy, how revealing our intimate traits can be used to let say manipulate us to buy products or maybe vote for political candidates. well, creepy, it makes me feel uneasy and so on. but we have to realize that outside of our western liberal free bubble, losing privacy can be a matter of life or death. think about countries where homosexuality is punished by death. if a security officer there can take a smartphone, take a photo of somebody's face, and reveal their sexual orientation, this is really, really bad news for the gay men and women all around
10:05 pm
the world. now think about political views. think about other intimate traits like intelligence and so on. basically, the sooner that we make sure that we are acting, or we think -- we basically take an assumption that there will be no privacy in the world, how do we change the international politics? how do we make sure that lives of minorities, religious minorities, sexual minorities, political minorities in other countries are preserved, even in in the post privacy world? it is a question we should be talking about now and not how to change the policy to protect our privacy better, because guess what -- you already lost. we can win the battle, but we completely lost the war already. david: we can really talk a lot
10:06 pm
about that, because in preparing for this and reading about this very position, you know, it is a post privacy world, deal with it. i really truly wondered, is democracy possible in a post privacy world. does that make the effort to create laws that mitigate or prevent such a condition, does that make it sort of an existential question? it's how my thinking went. i do not think we can solve that in the time available. i want to turn to some questions from the audience. apparently, lots of people had this question. can you change your digital profile by liking items you actually do not like, or by searching misleading things to sort of cover your tracks?
10:07 pm
[laughter] michal: that is a great question. great question. my answer would be, no you cannot. because maybe you can like some random things, but first of all it would make you look pretty weird. to your friends. people would not do it, right? imagine posting like 100 status updates and only one of them is real and the rest is some kind of random gibberish. [laughter] nobody would do it, not to mention a computer would have no trouble figuring out which one of those 100 is an actual status update and which one is not. so, no, you cannot do it because it not only would make all of those platforms like facebook and google in netflix completely not fun anymore, but it also would make your life not fun. imagine with your credit card,
10:08 pm
you have to buy 100 random things -- [laughter] before you bought whatever you want to buy. it is not possible. let me give you -- even with facebook likes it is not possible. let me give you an anecdote here. one of the results i wrote about in my study, and this was the result that people with high intelligence -- people who like curly fries on facebook tend to have higher intelligence. the correlation is actually very tiny, because the fact that you like curly fries does not make you smart. but there is a tiny correlation that when you combine it with other likes, you can reveal a person has a higher iq. there was a test dog when this fact was mentioned and the next day the popularity of curly fries on facebook went through
10:09 pm
the roof. everybody liked them on facebook. guess what, algorithms quickly discovered that. so this is why we call it machine learning, because it realized, ok, curly fries is no longer diagnostic of high intelligence. [laughter] david: does that suggest that it is like the pattern of data points rather than any particular data point that is where you have, or you can find a signal for those traits? obviously, even though i love curly fries -- of course, as we all are. everyone in this room. [laughter] it is nonsensical that that could have anything to do except by some baroque argument that there is some cultural signal in liking curly fries. i do not see it. does that suggest -- not a particular data point, but it is patterns?
10:10 pm
michal: the power is in numbers. the more data you have about someone, the more accurate the prediction would be. even when you look at things like signal theory. i didn't want to bring it up, but look at signal theory -- if you have a real signal, but you add a lot of random noise around it, you just need a slightly more of a signal and you are still going to discover what was the truth there. so unfortunately, this is a great idea, but it is not going to work. [laughter] david: here is another representative question i think also many people asked. you argue data analytics can be used to engage more people in politics, to which the writer of the question agrees, however could it not also be used to dissuade voters from voting? some argue certain political parties benefit from lower voter
10:11 pm
turnout. may incentivize to try to achieve this through this sort of thing we were talking about. michal: good point. no question that trying to discourage people from voting is an awful antidemocratic thing to do. but, you guys have to realize it is not the fault of social media that this is possible, it is the fault of the cultural and legal environment in this democracy, where it is ok to put negative adverts on tv, telling bad things about the other politicians. so if you are concerned with people being discouraged from voting, i think that it will be, again, it will probably be impossible to come up with regulation that would prevent politicians from actually doing
10:12 pm
it somehow. because maybe they will not do it themselves, they will ask friends in other countries to leash out some bots that will do the job for them. at the end of the day, also when i am discouraging you from voting i am not saying do not go and vote, i make up a story and tell you about a pizza place -- [laughter] so there are other ways, ways that are difficult to police. if you are concerned about people not going to vote, think about a smarter way of solving the problem than regulating traditional media or social media, like for instance -- and i am not an expert on political processes, but one thing that comes to mind is left -- let's make voting an obligatory thing, but also the duty that we have as citizens. [applause] david: the french example. [laughter] i think this is a very interesting question.
10:13 pm
can or maybe how can the sort of technologies and techniques we have been discussing be used to allow individuals to understand how they are being profiled? what they look like in these kinds of big data views? michal: great question. i see quite a few apps out there that are basically aimed at doing this. you can go and open such apps and they will tell you how you could potentially be seen by advertisers or platforms. one of such apps is hosted by my previous academic institution. cambridge university. it is called and you can go there and basically the platform will take your facebook likes and tell you
10:14 pm
what predictions can be made based, what predictions, what can be determined from your facebook likes. i would argue a simpler way of checking how you are being perceived by the platforms. open facebook in the morning. arestories you see there basically the stories that facebook ai believe you want to see. ufos, see stories about you probably, according to facebook, mike's computer seat theories. -- mike conspiracy theories. david: i think you should find this to be an encouraging question. how do i get into the field of research? [laughter] what skills do i need? michal: that is a great
10:15 pm
question. i love this question. toould encourage everyone get into this field of research and to continue studying, because it is fun. past, i dropped out of my high school because i was so excited about running my startup , and then i dropped out for my college three times because i was so excited about running my startup. it was great. then, i discovered science by accident, to some extent. then, it is just the best thing in the world. i have the best job there is, and i would gladly pay to do it. actually, people do. they study for years. to work inthey pay academia. another angle to it, which is, i would strongly encourage every
10:16 pm
single person in this room, whatever you do in your life, you should try to learn programming. programming is fun. like a computer game. you don't shoot at people, but you build droids into virtual space. it changes your thinking. you start thinking in an organized way, start thinking with procedures and with loops. [laughter] i'm sure you would not lose your current way of thinking, but it will enrich your thinking. i think this is important, not only because it is fun and there are great jobs out there, and becomingbecause ai is more important in the society, which means we are surrounded by products and services that are run by software.
10:17 pm
understanding the language of software will enable you to understand those entities of our ai overload better, but it will also allow you to understand how successful people think. among the most successful people recently, there is a huge overrepresentation of nerds and geeks who program. those guys are shaping not only the products you are using, but also the society you live in. now, being able to think like them and being able to understand them more will or could empower anyone. by the way, i am a social scientist. i am not a software engineer. i learned software engineering. code usingriting my google.
10:18 pm
a psychologist can do it, i'm sure everyone in this room can do it as well. david: we only have a few moments left. want to, i just use my prerogative as an interview getting in the last question. it was another provocative thing i saw in your work in recent talks, which is a suggestion these classification algorithms could be used in the context of employment and employment decision making. interesting,uld be i believe this is already at play insociety -- society. if you can tell us your concerns about that and what you see is hopeful about that. michal: that is a great question. thank you for bringing this topic up. i probably don't really have to talk about the concerns, because
10:19 pm
we are all concerned. humans focus on the negative. i want to focus on the positive for second. you guys remember the stories about the first cars being introduced in cities. they had to have people running with a red flag in front of the car and warning everyone that a car is coming. want to openhat i this conversation is, we had examples in the past where new technologies scare people a lot and they focused on the negative and had a tendency to overlook the positive changes the technology is bringing. you probably have more examples been just cars. it is the same with the use of algorithms and hiring. shortcomingsnty of of our rhythms and hiring, but you have to look at those shortcomings in the context. is, how we go about
10:20 pm
hiring people at the moment. we are doing it in the old way. why? because we are being unfair and racist and sexist and a just. -- and ageist. even a bad out rhythm be left sexist and racist than a good recruiter. algorithm would be less sexist and racist than a good recruiter. if you don't think you are, you are dangerous. [laughter] you still are, you're just not taking steps to prevent yourself from damaging others and hurting other people and being unfair to other people because of those negative traits that all humans says. when we go about hiring, how do we go about hiring people? we interview people. interviews are great, especially for the interviewers to feel , buttant and empowered
10:21 pm
there is a lot of scientific evidence, one of the most well-established facts in industrial organizational psychology is that interviews have close to zero predictive validity of how well you perform at work. if you are going to be a spokesperson, then it's are innt how likable you an interview. for most of the other jobs, this doesn't matter. what happens is that when you what the people, interview is doing is measuring how much you like a person. you like people that like you. tattoo and there tattoo,ng person with a
10:22 pm
you will like her left, because it is something unusual, new, strange, song. -- strange, so on. algorithms are going to bring more fair hiring and better hiring and recruitment decisions. important, letting people out of prisons, deciding the length of sentences. it makes me feel uneasy when you are telling me that algorithms should decide whether to put someone in prison or not. but guess what, we're doing it already. says still a judge that you are to be released on your parole, but the judges making a decision in many places in the u.s. it is already happen -- happening at the judge will get a printout from an algorithm telling her how likely the
10:23 pm
person is likely to reoffend. the judge will follow, because that is what people do, what the algorithm tells him or her. it makes me uneasy, but when you look at the effect of these policies. two times more people from prison while keeping the reoffending rates stable. i am hopeful, because it means in the future we will have more free people out there. we will have less people in prison while having a safe society, or safer. it means in the future we will have people being hired in a and fair way than today being more fulfilled by their jobs. i am hopeful looking to the human-ai the interaction. david: it made me think of the phrase i heard recently, a
10:24 pm
mirror-tocracy. i guess it is important that we make a good algorithms. we have come to the end of our time this evening. thank you very much for a fascinating conversation. [applause] michal: thank you, everyone. [applause] david: thank you all very much. [applause] [captions copyright national cable satellite corp. 2017] [captioning performed by the national captioning institute, which is responsible for its caption content and accuracy. visit] journal"'s "washington live everyday with news and policy issues that impact you. david up sunday morning, with the foundation for defense and phyllis
10:25 pm
discussed the trump administration's foreign-policy and the president recent speech on afghanistan. officerformer naval know at the hudson institute center for american seapower talks about the recent glitches involving u.s. warships. be sure to watch "washington journal" live at 7:00 a.m. eastern sunday mornings. join the discussion. sunday night on "q&a" the lives of within churchill and george orwell. we talk with thomas about his book. >> churchill and or will never met, for example, but the hero was winston. churchill read it twice and loved it. admiredingly, orwell churchill across the -- clinical cousin. -- across the political chasm.
10:26 pm
>> sunday night on c-span's "q&a." this might be the only get -- only government class you ever take. andare going to be a voter your forever. i need to give you tools that are going to help you for the rest of your life. >> tuesday night, a high school teacher discusses how current events affect lessons on history, politics, and government. >> this is a chance for them to learn about their story. it doesn't start when they are born. it starts of people who come -- have come long before them who have shaped the world. to realize this doesn't just start and end with me but what i contribute, where i am coming from, in that way alone them to take in other people's opinions, a prospective -- a perspective.
10:27 pm
,t gives them a chance to think this is how i see the world, but why? how can i expand that? tuesday at 8:00 p.m. eastern on c-span,, and listen using the free c-span radio app. now, a discussion on the strategy for afghanistan that was announced this week by president trump. from washington journal, this is 30 minutes. host: welcome back. michael rubin is with us now, resident scholar at the american enterprise institute to talk about president trump's approach in afghanistan. what you see as the most significant shift in president trump's approached afghanistan compared to the prior administration? guest: president trump was talking about leaving afghani stan throughout his candidacy. president trumpnn


info Stream Only

Uploaded by TV Archive on