you know how we can find out more about me how about that you Bing yourself oh wait no that's Bing no no no then that's not going to work no because Bing will actually just give you some basic information but uh there's a surprise if we look me up on Google you ready to be surprised I am not ah there we go we anyone see the surprise this is the this is the smartest search engine in the world do you see any problem there anything a little off how did how did it to give you the top oh I always get top rting for but that's not the problem you don't see no one sees a little like some a little problem there maybe say it louder Kate picture is not you that's not me that's right well it could be it could be you in a very much earlier instantiation I suppose no no this guy is a I I figured out who he was he's a contractor in New York okay and he builds buildings and that's what he does that's not me this is the smartest search engine in the world okay they've been running uh people who are not me on my bio uh for as long as I can remember do you think maybe I pissed them off or something did I did I irritate someone there laed a photo you think could me yeah I'm sure it's innocent right everything they is innocent right it's it's all the it's always the algorithm right it's never the people there are no mischievous programmers right just because I've been a programmer since I was 13 and and I've been mous my whole life doing playing pranks like that my whole that doesn't mean any of those people work for Google it's impossible who maintains the Google Blacklist well I have an article coming out on that so I I know I know more about their Blacklist than they do at this point you know I have here's my thing right here see there are my slides it's 54 slides okay no slides nothing up my sleeve I'm just going to talk to you as people it's a small group maybe I a few of you I've already chatted with a few names I even remember this is jeene CU this is Kate uh let's see one of you guys well that's Dennis obviously that's easy one of you guys is Mike there Mike and uh who's the one who heard the lecture last night from Nate Percy Mike and oh others too did Nate mention me by any chance he did how nice yeah we spoke together last week at the at the National Academy um expressing uh some concerns I guess you could say all right let me tell you why I'm just going to talk to you as people as human beings and tell you kind of a little journey of research I've been on okay let me start out by saying that I'm the most skeptical person I know and it seems to me all scientists are skeptical people they're supposed to be that's the idea and what I'm saying is on a scale of 1 to 10 I'm a 10 okay you never want to have a journal send a paper to me for a review never I'm one of those people uh a professor of mine very famous historian of science at Harvard many years ago I Bernard Cohen any of you know of Cohen he founded the history of science department there extraordinary man what a character I mean even he used to complain about how personnic I was okay I I'm very skeptical so that's one thing you need to know second thing is and this you probably already know is that researchers tend to exaggerate the importance of their research anyone know anyone like that anyone ever maybe done that and is willing to admit it I have especially when I was working with pigeons okay I definitely look back on that and I think what the hell did I say because but you know we do that when people who are researchers and I've been a researcher for almost 40 years um sometimes we do it just because we're enthusiastic about our work and that's I think fine sometimes we do it because we're trying to impress somebody a journalist a Dean uh who else you need to impress well a granting agency a funding agency right uh with me sometimes it was a girlfriend hey you got to see the research I'm doing want to see my etchings my research etchings I mean there reasons why people exaggerate and I have I've exaggerated the importance of my work what I'm about to tell you about the search engine manipulation effect okay I I find myself in the opposite position I started exploring this a few years ago uh I thought I would find something and it would be very really really small but I thought that would be worth something I thought that would still be important to find this really small thing and I'll give you details shortly that's what I thought I'd find that's not what I found in fact I didn't believe what I found and now I find myself you know a few years later in this very awkward situation because the more I have learned about this the more concerned I've become and I don't even have a way to communicate this to people generally speaking I don't even have a way to communicate it because uh I don't I don't even want to believe it so it's odd I've never been in this position before in my life and that's why I'm just talking to you as a person as a as a fellow human being if you're if you're a parent I've talked to you as a fellow parent I have kids of my own and two stepchildren and you know when you're a parent especially you're you you automatically start to think about you know 20 years from now and 50 years from now and so on right and I I swear I don't I don't have a way of explaining how odd and I think important you know this work is and it's and and I think it's bigger than I even think it is if you see what I mean now let me explain okay I I still am directing about a dozen research projects so this is a very small takes up very little of my time okay take my time divided by 12 it doesn't take up that much of my time next year I'll have coming out the biggest uh sexual orientation study ever conducted more than 300,000 people more than 90 countries and and it's and it's a good study I mean it's it's you there's actually some some discoveries in study and I do research on Stress Management and creativity parenting love whatever and Google to me was just a cool toy it's like the coolest toy ever that's all it was to me nothing else the beginning of 2012 gee what day was it oh could it have been ah yes January 1st New Year's Day how could I forget I I got about six or seven emails uh all to me all from Google saying that um my website was being blocked okay I okay so okay I've been hacked everyone's hacked right uh but what I but then I started there were there were technical aspects of this which as a programmer I I got very curious about so so uh first of all the fact that Google even has a blacklist was news to me and then as I looked into it more and found out that they are actually have on any given day millions of websites on their Blacklist that was news to me who just said yep and who are you I'm Carl H again Carl Carl hi Carl hi Robert Epstein that's what I that's not your picture that's pict I don't know maybe that's the real guy maybe he maybe we need the contractor here so I didn't know that and I started looking into it and then but then again I got kind of curious because I realized okay they got crawlers blah blah blah they can find malware blah blah now they're going to make mistakes yeah whatever and and of course when they find malware it makes sense that the on their search engine they're going to put a warning so I was okay with all of that I it made sense to me I just had to find the malware get rid of it blah blah blah some aspects didn't make sense the fact that you know it takes a split to get on The Blacklist and it can take you weeks to get off The Blacklist doesn't still doesn't make sense to me you know fact that they have no customer service department who can help you you know there's some aspects of it that were like I thought were a little odd but what I thought was really odd because I'm a programmer was that when I tried to access uh my main website or or the kind of the you know the ancillary websites that that have psychological tests using Firefox or Ari I found I couldn't get to them and I thought what that doesn't make sense Firefox is not Google it's not Chrome I don't get it Safari is certainly not it's that's a competing company and Firefox is a nonprofit and I and I and I said wait a minute how could Firefox what's the mechanism what is the mechanism by which this is occurring so I have an article coming out that's all about Google's Blacklist it's long long piece it's got all kinds of things in there and even even their financial motive for uh using The Blacklist uh is in there which no one's ever revealed before so but I'm not going to tell you about that I'm just going to tell you that because this happened which was a very minor incident I started looking at Google a little differently with a with a more critical eye like I look at everything in life I started looking at Google that way and you I found things that were genuinely disturbing of course I I quickly got up to speed on their business model which still to this day I astounds me that how few people know this I'm sure everyone in this room knows that Google is actually an advertising agency that's that's that's all they are they're an advertising agency with a brilliant uh method of data collection that's all the method of data collection of course is you give people lots of free stuff like a really cool search engine and then you take the information and you basically leverage that with companies who are trying who want to sell products and services it's brilliant it's a brilliant model and people don't see what's going on you know so I wrote a piece for Time Magazine about that which is called uh Google's dance about this kind of OD so I and I just kept kind of looking at different aspects of the company and I thought wow this one after another I was seeing things which I thought were just true genius and as I understand it uh the guys who created this uh did so not too far from here yeah so you know so this should be especially meaningful to you because I want to make it clear that I think that when these these two guys came up with the search engine it certainly wasn't the first one it was probably the 21st roughly Eugene yeah so uh I don't think they set out to create a monster and I don't think they even set out to create an advertising agency I think that came later because there's a there's a guy who actually left there named um Whitaker I think his name is who actually who was a pretty high ranking guy who was there at the beginning who then actually left the company and published a piece online saying why I left Google and he basically said in the beginning it was just this cool corporate anomaly and then it became this advertising agency he said and so he left now of course most people don't leave Google it's very you know people don't leave because there's a lot of money at stake but okay toward the end of 2012 uh you know thinking about various issues related to Google uh which again I I think again Larry and Sergey never intended I don't think they intended the ad thing initially and I I don't think they intended these other things that have been creeping up on me and that I'm learning more about and going I don't think they intended any of this stuff but anyway toward the end of 2012 um I was looking at the literature it's it's a literature studying consumer behavior and uh it's it's pretty extensive and I was looking at that literature on how search rankings affect consumer choices and it's fascinating and I found to be fascinating I mean for for one thing uh you know people pay a lot of attention to what's at the top of the list on the first page and much less attention to what's below there eye tracking studies showing for example that uh you know they're going to look at what's at the top of the list even if very obviously there's something much better that's down in position seven or eight the I might glance down when it comes right back up to the top and ultimately 50% of all clicks go to the top two slots 95% % of all clicks go to the first page okay this is this has been shown repeatedly so higher is better right and that's why a lot of companies spend a lot of money on SEO search engine optimization because they're trying to go up another notch because you can literally calculate for any given industry how much money that's worth to you if you can just jump up one notch do you know how much Jus uh no no you should know that Justin is one of my sons and he works in internet marketing in the Bay Area and there he is yes ma'am what is your name I'm an Ze and you realize there's so few people here I can actually get to know all your names an well I've run a big home school website for 17 years and I used to be one of Google aden's top 500 Publishers and I was making around $200,000 a year on website and then um uh Panda and you were on the first page I was on the first page of course you were I I bated around the first three and um the interesting thing is when Panda hit AdSense did not know it was coming those guys were all being paid on commissions by how well they helped their Publishers well their Publishers did and they were Furious they were totally caught the GU they could not talk to the search people they had the whole Department wrote a white paper to complain about what the search people had how they search had ruined their income stream basically so I I found that interesting that Google people don't talk to each other well there's a wonderful piece which I recommend to all of you if you're not familiar with it written by the head of of Springer in in Germany which is the largest publishing conglomerate in Europe called fear of Google and it's all about what's happened to search trank of various subsidiaries of his company and how a lot of CEOs just live in absolute fear momentto moment Terror of what Google might do to their search rankings because if you if you suddenly we did went down to a quarter of that yeah overnight so so just just a clarification Panda was a revision of the search algorithm it was a major rision uh I'm February 28th 1913 20 2013 it it was overnight all sudden I'm like what the hell is going on oh I dropped I had been on 450,000 Google Pages me and my research and I dropped to 60,000 but with me they didn't stop so now I'm down to you'll have to look it up yourself well now what they do because it's an education site they know it's important in August and September and then it like falls off map it it's is sort of weird it's so they've prejudged my usefulness of my site even though it's useful all year round well keep this in mind what Anna saying is very important because what she's saying is you know that there are decisions that are made right well actually not that far from here which direction is it I don't know I'm disoriented but they're decisions that that that impact our lives and we have no idea how these decisions are made there's no transparency we have no control over what's happening but they affect our lives I am going to tell you about something that is impacting most of the world right now and in ways that go far beyond elections that's what I've been realizing so we saw the power of these search rankings on consumer behavior and we thought about it and we said yeah I wonder if if people really believe that what's higher is better maybe they also believe that what's higher is truer and maybe that could impact uh you know their opinions on something that they're uncertain about so uh we ran an experiment in uh this is about January 2013 and uh we created our own version of Google which is called kadoodle but it works exactly like Google does but it but it isolates people from the full internet okay so they're trapped in our in our world our search engine world and we thought let's see if we could use a search engine to influence voter preferences that was the idea uh we used as our election a real election the 2010 uh election for prime minister of Australia Julia Gallard uh versus uh what's his name he's the current prime minister Abbott yeah and so we use that election why because we're going to use our subjects were all going to be Americans who probably would know very little about these people uh and that's good because that makes them by definition undecided and that's what we wanted to start with people who are undecided and we said up uh I'm which I'm very proud of we set up an extremely well-designed uh experiment so welld designed that experiments with this design eventually made it into pnas just just in August because they're randomized they're controlled they're counterbalanced they're double blind I mean absolutely the gold standard of research so that our results would be unambiguous and the basic design is very simple people who are randomly assigned to group a when they're doing research on the candidates for this election uh they end up seeing search results which are biased in favor of candidate a by my in favor I mean that if a search result is higher it's going to have more praise for that candidate and possibly more criticism of the other so and people in group b they're going to have the same thing but for supporting the other candidate candidate B and people in group C they're getting an alter alternated search search results so in other words they're getting search rankings that don't favor either candidate very simple very clean design and I thought that if we you know we first ask people some general questions about you know who would who you know here here's a brief description of the candidates uh you know who would you vote for who do you most trust who do you like the best you know a whole bunch of questions like that uh and and sure enough we get the in the right right at the outset before they do a search we get numbers like right smack in the middle we get 50/50 pretty much uh uh you know for all statistical purposes that's what we get no preference which is what we want right we thought okay if we Happ do a search maybe again the people in these bias groups they're going to shift a little bit at least on their answers to these some of these questions we thought they'd shift and I thought frankly that we could boost the proportion of people favoring either candidate I predicted we could boost that proportion by 2 to 3% two to 3% and of course you never know what you're going to get right but that was my I always make predictions um sometimes they actually work in that first experiment we boosted the proportion of people favoring either candidate by 48% in a search lasting only 10 to 15 minutes 48% that's 2% yes if you want to think of it as a 100 people and I've got 50 for each and I'm boosting it by 50% that means 25 or 20 to be precise 24 more people so now basically I've got three4 of my population uh after one search and it wasn't just the proportion of people saying they would vote liking went up trust went up overall impression every single measure we had we had five different measures all moved dramatically dramatically in the predicted Direction now there was something else that was disturbing because I can tell you these were blatantly biased search results blatantly okay there's no possible doubt and we asked people whether they were aware basically that there was you know something odd about the search rankings we couldn't use the word biased obviously because that's a leading question the point is it appeared that 75% of our participants in that first question seemed completely oblivious to the fact that they were seeing bias search rankings that they were being manipulated so naturally the next thing to do with another group was you know we thought about it we said maybe we can mask the effect so instead of having you know uh you know Gard Gallard Gallard Gard Gard you know all the way down the first page what if in the fourth position we just put in one search result that favors the other candidate by the way these are all real search results connecting to real web pages and they can roam around as they please just as they would on any search engine so they can read whatever they want and spend as much time as they want on any you know maximum 15 minutes so now we've switched the fourth position the percentage we still got a dramatic shift in fact the shift in this case which we call the BMP for vote manipulation power wasn't 48% it was 63% these are Big Numbers uh but now 85% of the participants seem to be oblivious to the manipulation unaware so we said what if we just move that in the next experiment to the third position because right higher you are the more power you have over people just moved to the third position all did again huge shift and 100% of our participants were now oblivious to the manipulation that's all we did is switched the third position okay so this was this was not what I was expecting and I was concerned we had gone to Great Lengths in picking subjects this we did not use college sophomores okay we spent money we did you know advertisements in the community we went to nursing homes I mean we did whatever it took we wanted to have a you know groups that were representative in as many demographic uh categories as possible of the general voting population so very very diverse groups in these experiments but they were small and everyone we had even though these were very diverse groups was basically living in the San Diego area fourth experiment we go Nationwide online more than 2,000 subjects from all 50 states again very diverse group we get a shift of 39% and again very few people aware that they're being manipulated there's something wrong with the rankings now in this case we've got a nationwide study we've got a lot more subjects so now we can start to look at demographics we can look at subgroups right so obviously what's one subgroup you want to look at you want to look at the people who were aware that they were being manipulated because presumably those people don't shift right so this is this was now big enough so we could look at those people and we found that they were very much aware because we let them not only you know check a box but they could type up a paragraph and just you know complaining about what was wrong with the search rankings these people were very much aware that these were highly by a search rankings but they didn't shift by 39% they shifted I Believe by 47 something like that percent they shifted even more than people who were not aware of the manipulation me if I were aware that 90% of the articles I found said U Bob Dylan is a great singer I might just because 90% said he was a big good singer be more convinced that if I saw alternate he's a good singer he's a bad singer so if I look at that as truth behind the differences I would be more convinced that truth was the number of articles with on that side but remember they have access to all the pages well but I I look at the page and I see 90% saying Bob Dylan is a great singer that convinces and you're lazy like everyone else is so you don't go beyond the first page is what you're saying yeah all right that's fine yeah no I'll I'll buy that I'll buy that I have a question ma'am right up maximum and how many pages were there and how many 30 pages I I would urge you all if you want to see the uh the the the details please click on that you know the website for this series of talks and then just click and you'll actually have the pnas article and all the supplements so you're saying 15 minutes 30 pages up yes so they have up to 15 minutes and there they have 30 pages which are spread out over well in real life you have more than 15 minutes so this oh no you don't no in real life searches are are very on average are very very short uh and uh in real life there are a lot more pages available right there are billions but in real life in fact 95% of EV of everybody never leaves the first page and again 50% of all clicks go to the top two items so you know uh in fact our are the if you look at the time distributions and the click distributions which is all in that article you'll find that we are just dead on consistent with what's been found in consumer research just dead on our our subjects behave in in these experiments exactly like everyone behaves and in fact until real recently a fair number of searches were single word searches and same thing happened in libraries as well too when computer systems entered libraries yeah yeah yeah yeah that's a good point okay so I'll buy what Mike said uh you know I I I think I think you could say that if you're if you're aware of uh again that that that not just this search result and the corresponding web page favors a candidate but all of these search results seem to favor the candidate that's a kind of social proof you could call it maybe you're even thinking ah the search engine itself the algorithm obviously prefers Donald Trump so that that's an you know and I trust the search engine who knows but the point is being aware of the bias does not protect people from it not at all it makes them flip even more so now we had four experiments and we've learned a lot the fourth experiment was also large enough so we could look at other demographic groups because and again what did I predict I predicted there would be some de differences among demographic groups but they would be small I was wrong again okay some demographic groups were highly vulnerable to this effect which we eventually called the search engine man manipulation effect or SE sem some demographic groups were highly vulnerable others far less vulnerable the most vulnerable demographic group that we found was moderate Republicans what percentage of those people did we shift anyone want to guess 100% 80% 80% toward any either candidate 80% uh the next one after that was moderate Libertarians that's consistent that's consistent I mean I think they shifted by over 70% so you know we have the ability now to look at demographic the specific numbers for these demographic groups don't matter all that matters is the finding that there are dramatic differences among demographic groups in terms of their vulnerability to seam that's very important from a practical perspective because if you know that and you know and you have massive profiles on people you not only know you know everything about their demographic characteristics it puts you in a much much better position to change people's opinions yes what is your name CHR Len Len hello if I shift the context uh just for the moment and I say okay I'm going to show some moderate Republicans uh stories on Fox News if I if the same sort of thing happens reading newspapers or looking at TV news as happen with the search engine I would expect a shift even though people know that Fox News is biased I would expect a shift to occur among the people who listen to Fox News even if they come to it with the thought I know that this is a bias yes it's called The Fox News effect it's been studied empirically uh it's been Quantified uh it's been studied uh in the context of markets in which Fox News is just entering that's how it was studied and so you you watch and see what happens whether you can shift votes to the right and in fact they shift predictively to the right it's very small shifts so these shifts are about 1% or a little under 1% they're not big shifts but they're enough to flip close elections the same research also shows that in those markets when another competing uh um you know media Source comes in that's not consistent with Fox's politics then the effect starts to get smaller so so yes now you're talking about a world by the way which is the normal political world in which there's competition among different uh you know view viewpoints I'm talking about something completely different I'm talking about a a a media uh media a medium in which there is virtually no competition you see so here you are you're undecided and now you go to this particular medium which you have a high degree of trust uh for and there is no competition and don't tell me that Bing is competition because it's not and Yahoo just signed a deal with Google so that Google Google search results are now going to be incorporated into Yahoo Believe It or Not they've just given up uh and not only that most countries in the world certainly all through Europe 90% of all search is done on Google uh my wife and I just lived in the Fiji islands for a while I was appointed a professor at the University of the South Pacific in Fiji okay crazy thing to do uh even in somebody asked to do it yeah uh but even in Fiji even in Fiji uh 90% of all search is done on Google so yeah so you you you in the experiments that you did you provided a ranking and ran experiments against that ranking but the ranking was really uh pretty much uh the same for every for every participant in the study in the real world now uh Google and others have um a uh an understanding of the individuals could choose a ranking which would uh move them in some specific Direction based on their particular prui that's right and that would seem to me to be a very interesting experiment the this is um essentially the AdWords model uh in a different context yes well we've been doing that I I I you know we we've looked at that issue the the personalization issue and yes if you personalize then guess what you can have a bigger impact so what we call the VMP that shift that percent shift uh gets larger and there another advantage to personalizing by the way because if you personalize then no regulator can ever see you doing anything because The Regulators they never see those those personalized rankings that are going on to people they never see them fifth experiment so at this point you know everything we've done is disturbing we're getting big numbers but you know these are all eligible voters it's true but you know it's still you know we're using the Australian election with Americans so we went to India early 2014 uh we basic basically set set up our our mock search engine uh and recruited more than 2,000 2,100 people to be precise uh all throughout India almost every Indian State uh the only requirement is they had to be eligible voters they must not have voted yet because the vote in India is actually spread out over a few weeks and uh they let's see eligible eligible voters must not have voted yet oh and and they had to be fluent in English high high level of English fluency which is and English is actually the one of the two official languages in India so it's easy to find people who speak English quite well and we set it up so now people were randomly assigned to one of three groups since there were three major candidates uh Modi who ultimately won kedal and Gandhi and uh I predicted you're probably figuring out now that my predictions suck okay um I predicted that we would get a shift I thought of between two and 3% or nothing because these are real people in the middle of a real election where they're being bombarded especially India the campaign is much more intense than it is in the US they're just bombarded uh oh and they had to be undecided they had to be undecided they haven't voted yet they're undecided and so on and uh we compensated people by making a donation to a big children's charity in India some people were paid but most people actually we they did it in order to activate the uh the donation I was wrong again okay what we found and again you want to look at the pns paper for the details but the bottom line is you can very easily shift people even in India even in the middle of a big election by at least 20% 22% you'll see that figure in there there are other figures too because this is a more complicated situation than the ones running in the US uh but I I stand by that you can easily shift people by at least 20% in the middle of a real election and notice we're doing this with one hit one experience what would happen if people were just kind of doing their everyday searches not just on candidates but on issues that are election related and they're doing searches on immigration on abortion whatever it may be but you know they end up with search results that are that favor one candidate what happens if they do that day after day for weeks or months see again I think all all of our numbers are are are they're low yeah yes sir your name I didn't oh you didn't oh oh I'm sorry holding my hand like oh well what's your name anyway David David hello that's my middle name any did was there a hand up over here or was I imagining that okay just wasn't David's hand he was holding his head up okay so also 99.5% of the people in the study showed no awareness of the manipulation and again huge variations among demographic groups uh and some shifts uh as over that were over 60% in fact one shift in one group was over 70% uh so those five studies are in pnas those five experiments I should say are in pnas now uh the precedings of the National Academy of Sciences they came out in August so there's a link to the article on yeah um if you use that link you don't have to pay the $30 you don't no no no we we paid the additional whatever it was I think it was $5,000 so everyone has free access that's how much they charge you so uh no everyone has free access so that's that but we've gone beyond that and this is where honestly uh I think this becomes for me disturbing because what I realized as as as this work has has moved forward what I realized more and more is this is a lot bigger uh than I thought number one number two it's bigger than me see what I mean I can't do justice to this I can't do the research that needs to be done you know it it would take a massive effort to do the right research massive so in my head I know kind of what would need to be done but I can't do it one thing we realized was that it's not just voting preferences uh it's there other things uh so there's a group now one of the MOX punan institutes in Germany that is looking how at how C might be affecting Health decisions um I'm working with a prominent law professor some of you may have know about him Frank Pasqual who's at the University of Maryland uh we are looking at weather seam can be might be impacting Court decisions how can that be well more and more law clerks just like my lazy friend Mike here I'm lazy too because I I use Google Scholar now most of the time when I'm doing uh you know my scholarly searches I'm using Google Scholar which is nuts it's crazy but more and more law clerks are using Google or the existing search tools that they have used such as Lexus uh and what's the other big one no no no the the other one for law West law they are now more and more uh because of studies showing the power of Google's algorithm uh for for researching cases more and more the other tools the other traditional tools are using Google type approaches to their ordering to the rankings that they do so there's all kinds of areas here that one could look at that's a little scary because again we we don't we have no control over this how long do I have have until 54 all right but I'll but we want to have some chat chitchat right so I'm going to go maybe another 10 minutes and we use the remaining time for questions okay so there's one problem there's one issue is there probably lots of things that this is impacting now we have looked at other things other issues in our experiments for example we were wondering whether we could suppress seam uh we've tried various warnings we did we did three very large experiments uh in connection with the uh the election this past spring in the UK uh and this time we decided to open it up to people from uh many different countries so I think we had people from 39 countries uh different levels of familiarity with with the candidates obviously so uh lots of opportunity to look at demographic issues the point is is what we did though in different groups was give people warnings we told people this is biased it's sometimes at the top of the page sometimes on each item saying this appears to be highly biased in favor of Cameron okay and we actually told them and when you tell people in a in a negative and scary way uh then in fact you can suppress seam to some extent but it's still happening but you can push it down to some extent the only way to push it down completely is with basically an equal time rule as you know the FCC actually still has an equal time rule um and that's what it takes you actually have to mix them up now when you mix them up what are you doing you're you're violating the trust of the user and you're in in effect you're giving them crappy search results right so I can't imagine anytime soon any search engine company just you know mixing up the results uh but that's the only way you can actually suppress seam uh entirely we've also been testing uh an operant conditioning theory of why seam works so well uh and we just completed a very very intriguing experiment on this issue um where uh people in uh in some groups uh well let's put it way everyone in this particular experiment they get some they get to practice on our search engine before they they get into the political part of it they get just to practice on it so they know how it works but uh in half of the groups the the practice consists of uh you know doing Simple searches like what is the capital of Uganda and of course where does the result turn up when they get to the search results right at the top of course so uh you know in those groups they're basically getting what we all get every day which is Mainely that when we do routine searches the result is at the top but in the other groups these people are not getting that they're getting the result anywhere it could be on any page they've got to really you know hunker down and they've got to try to find it so we're with those people we're teaching them that for this particular search engine you know you that that top one is might not be it okay the result could be anywhere well when you do that in the first groups the average VMP that shift toward the candidate of our of our choice is 44% consistent with a lot of other stuff that we've done for the second groups the the the low trust groups the ones that we're we're we're messing with it drops to 17% and that number is not statistically significant ah well that's interesting because what this is saying is that the reason why seam works so well is because we are basically rats in a Skinner box uh footnote I was BF Skinner's last doctoral student at Harvard so I know a lot about rats and Skinner boxes and basically we are rats in a in a Skinner box we are in routine searches day in and day out non-stop day in and day out over and over and over again we are being taught the rule what's higher is better what's higher is trer what's higher is better what's higher is trer what's higher is better what's higher is true so when the time comes when we might want to actually search on something where we we're really not sure and there's a big decision to be made this makes us very vulnerable to being influenced by whatever is higher I'm pretty sure that's why seam is such a big effect in fact it is one of the largest behavioral effects ever discovered and you know we've known for hundred years that lists have impact on people we've known that for a long long time but this is a very unique kind of list effect because it is bolstered by the training that never stops what's higher is better higher is true what's higher is better what's higher is true there's never been a list like that so now does this mean Google is messing with us messing with our elections well if nothing I've said yet scares you this should because this is unprecedented I think a month or two ago Dr Amit singal some of you may know him he's the head of Google search he published an article in Politico criticizing me and my research what what now all it really says is I encourage you to read it in fact I put a link to it delivery so I really read his don't even read my stuff just read his I encourage you to read it because all he really says is Google is a really cool company and we would never do anything like that that's all it says I mean it's much longer you know it's like this long but that's all it actually says so I really really really would encourage you to read it because that if what I've said so far doesn't scare you that should scare you I published a reply there's a link to that as well it's up on huff post but you know the bottom line is this what what the hell they're they're they why would they brush aside research like this if I were running Google I would take this very seriously because I would I would say to the public hey you know we're concerned about this and in fact we want to study it we want to understand it we want to see if there's anything we can do to mitigate this effect we don't want to be flipping elections more money well you see that's that's that's the problem is there's money to be made right and obviously that's why they're just trying to brush it off but you can't brush this off this is too it's it's too big you cannot brush this off they should be the ones studying this maybe they are that's right what don't you think they're not well it's vital to their advertising business so you think they're not I I frankly every single time we discover some new aspect of this it pops into my head and I hate this when it pops into my head it pops into my head Google probably knew this three years ago I'm sure they're way ahead of me in understanding this phenomenon because they probably use this phenomenon for their own purposes now is there any evidence at all that they actually that their search results are biased in ways that serve the company well some of you might be aware that there was an internal FTC report in the US that was made public at least parts of it were made public just within the last year in which FTC staff said that Google does this all the time for many things many many things that they bias their rankings in ways that serve the corporate needs their corporate needs the government of India has concluded the same thing the European Union has concluded the same thing in India and the EU Google is now facing at least $10 billion in total fines specifically because of biased search ranking search rankings that serve their company but that do not serve the consumer and are not in fact objective the European Union the European Parliament excuse me and I've and I've been invited there I've spoken to Regulators there and so on uh I mean they're they're they're serious they're dead serious in what they're doing the European Parliament voted uh a few months ago now it's not it's not a binding vote it's not that kind of thing it's a it's a like a vote of intention they voted by and this is passed by a fairly large margin uh in favor of uh breaking Google up into small companies exactly like our justice department did to AT&T in the 1980s that would be another way of protecting the consumer from seam because if there are lots of little search engines they might be you know specialized I don't know how you know there different ways to go about this but if there were lots of little ones then we use them all equally if if if I don't the point is yes Google probably does things like this because it's it serves the bottom line I understand that and yet and yet and yet I have more and more I've I've actually become less concerned about the people at Google and what they might or might not be doing like the the the little the little pranks you know and so on I've actually become less concerned about the people there yes it's true that uh you know uh maybe Larry page could just type a couple of you know change some parameters whatever right through a back door or whatever and he could he could put Hillary up in no time oh Hillary gee are is anyone aware that a few months ago she hired Stephanie Hannon away from Google to be her Chief technology officer that Barack Obama's Chief technology off officer was borrowed from Google uh that six other Top members of the Obama administration were all former Google executives that according to the Wall Street Journal uh Google representatives of Google have visited the white house since Obama came into office 10 times as many times as any other company roughly once a week since Obama came into office and yet I'm less concerned about the people than something else the algorithm because what I've realized is again looking at the numbers it's all in the numbers what I realized is you don't need people you don't need you don't need Executives at Google you know fiddling you don't need a Marius Milner fidling who's Marius Milner nobody knows who Marius Milner that's amazing considering where I'm standing right now the fact that you don't know who Marius Milner is astonishing he is the guy that Google said was responsible for the street view Scandal which still hasn't gone away not not not in terms of uh of litigation for more than four years Google Street View Vehicles were driving up and down the streets of more than 30 countries not just snapping pictures but sweeping up signals from unprotected Wi-Fi networks collecting passwords collecting everything about everybody and fact I would bet a dollar that that's probably what street view was meant to do but I don't know for sure this was all blamed on Marius Milner one yeah was in fact motivated by the MIT project but doing Street I don't know you mean the media map that they did to 78 yeah okay but still you have to remember that Google is an advertising company and when and the every single thing they do has to contribute to ad Revenue everything so if you're going to whatever it is if you if if they if they weren't sending these vehicles up and down the streets in order to collect Wi-Fi data right you've got to tell me how simply snapping pictures was going to increase advertising Revenue if you tell me that I'm fine with it but that's always the key to understanding what they're doing their purchase of YouTube that wasn't that was a huge area of information a body of information about users that they didn't have now they had it creating Chrome that was also they had to create Chrome because before Chrome they were mainly getting their information only from people who used their search engine now that they had a browser they you didn't need to use their search engine they you still got all the information about every single search you did without you using their search engine you understand Android Android another step up it allows them to collect information even when you're not in line and don't even I don't even want to think about Google Glass okay uh always think how is how is it feeding the ad machine uh Google cars self-driving cars how does that feed the ad machine my concern is with the algorithm because we know without any doubt that the algorithm always puts One widget ahead of another and in elections it always puts one candidate ahead of another so in the election in India for example last year if you you look at Google's own figures on search activity uh Modi who won was for 60 consecutive days before the final vote was cast uh his search activity was at least 25% more than for the other two candidates now there's a problem here there's a kind of a feedback potential here because if in fact search activity and many other things I that's there are many other factors obviously but if search activity is is one thing that might push someone higher in rankings and if when you push someone higher in rankings that generates more search activity which it does there's a potential for uh feedback to occur for a Synergy to occur so in the pnas pie we actually have a mathematical model and we show what can happen What could happen conceivably even if these both are small effects if they're both small effects nevertheless in just before an election if it's done right you can get an explosion of support for one candidate because of the way those those two factors interact we also developed an equation which allowed US based on easy to measure factors like Pro Ed wind margin we figured out how to predict in advance country by country election by election which elections could be flipped using steam see again think about the resources we have very limited resources if we could figure out things like that obviously Google has much better statistics than you do previously they used try to use these statistics to predict flu outbreaks oh yeah they haven't published anything about using these statistics to predict elections no they haven't have they yeah no read read read singal's criticism of of my work just read that that's that's the scariest thing of all now I I'm we're working on all different aspects of this right now but I just want to finish up by pointing to what I I think is going on here look just looking trying to look at the larger picture there are ways in which Facebook can manipulate elections uh Jonathan a train at Harvard pointed out that uh based on a study that was published a few years ago which was a collaboration between uh researchers at UCSD and uh and employees at Facebook uh that you know Facebook could easily flip an election without anyone knowing they're doing it just by sending out go out and vote reminders to people in one party okay and he he calls that uh digital Jerry mandering okay I've told you about seam the search engine manipulation effect I've also told you at least a little bit about this possible Synergy and basically a digital bandwagon effect and that's not all there's more what what we're talking about here is really new forms of mind control that never existed before in human human history not on on this scale and certainly never in the hands of one small group because I don't care whether they have 37,000 employees or not the fact is you know it's it's a handful of people at the top who make the key decisions that's never happened before never and the fact that they're brushing aside these kinds of things that is a problem so what's the bottom line here the bottom line is this it's a question what if the Mind Control machine what if the Mind Control machine is teaching us not to mess with the Mind Control machine and with that I welcome your questions thank you and your name sir um I'm friend Ze I'm on half and I'm concerned about ramifications of this in terms of how people's mindset is getting possibly manipulated um their attitudes toward climate change global warming ocean acidification and other questions that don't have direct uh economic influence on too many companies but uh do do definitely on some but do do you see evidence of that sort of manipulation see evidence so here's a little anecdote from last week where both I and one of your colleagues here Nate persley spoke to a the the committee on science technology and law of the National Academy and I'm starting to talk about the same research blah blah blah right and the Hand goes up from one of the committee members who it turns out I didn't know at the time is actually a member of the Cali Supreme Court and he goes you know you're telling me all this he said but I just on my on my cell phone I just looked up and I think he said Hillary Clinton and I looked up uh one of the other C candidates he said and the search rankings look fine to me is what he said so see you see the question you're you're you're asking it doesn't have an easy answer because the fact is this this is is this effect is more subtle than the old subliminal you know stimul thing that was that was the rage back in the 1950s some of you may remember this book The Hidden Persuader was number one bestseller in 1957 and then through many many many additions after that this was about advertising companies what is Google it's an advertising company it makes no sense though an advertising company that they would do something to their search algorithm to totally screw their ad wordss and AdSense groups well again we don't know understand what happened and we don't know what the conversations are there yeah we don't I don't and I don't know they're not allowed to talk with each other well and and and you know an answer to your husband's question what I'm saying is you cannot look at search rankings and figure out what's going I can't do it okay I can't do it and I study this I cannot look at search rankings and and tell you you know whether they're biased and one way or another on an issue I can't do ites I mean that toally blew up you know here in California just it was somebody getting the measles in in Disneyland which in another day and age would probably have been a a one night wonder you know see but I guess what I'm saying I'm I'm I'm going to get you in a second but what I'm saying is more and more I'm realizing I don't care about the people the people could at Google or maybe being to some extent and you know the other lesser search engines they could in fact be screwing with us in all kinds of interesting ways you know that's possible but in a way I don't even care I care what the algorithm is doing because the algorithm doing its job is probably right now we estimate because we developed an an equation that allows us to make you know to figure out which elections can be flipped right now their algorithm alone is probably deciding the outcome of upwards of 25% of the national elections in the world outside of China and Russia Mike I was thinking one experiment would be to take uh Bing and Google or is it dog dog something and Google look at the ordering do a simple comparison of the bias of the headlines and see whether the algorithms of those are significantly uh different it would tell us nothing it would say that um Bing had a different orientation I'm not saying it couldn't be done it would tell nothing H it would tell us nothing oh okay fine it would tell us zero why because of personalization would tell us absolutely nothing I mean I I had a one of the Attorneys General of one of the states call me up about this and ask me a gazillion questions and basically uh between us we realized that the only thing you could do is create thousands of fake people and of course marketers do this all the time we you'd have to create thousands of fake people but the problem Google would know they're fake because of these profiles that's the darn problem so even this you'd have to create thousands of fake people you'd have to you you'd have to create different sorts of people you'd have to be looking at the personalized you see results that they're getting and then maybe compare against some baseline but Google would know that they're fake people this is this is an SEO trick is to create fake people Google has very good detectors for detecting people yeah yeah well in fact we paid that is the federal government paid for a PhD pis of a guy named zultan who was one of lar and Serge's colleagues I sat in in this ear defense took photos of it sent back to DC when I first when it was google. stanford.edu I could see the bias immediately uh the the bias was it would give you standord results over birthday results and we we were paying a million dollar to both St and Berkeley and um my my good test case actually was to try to find the LI Hill report which was a very anti artificial intelligence report and if you tried to do it using any search engine at that point in time the thing is all the people who had vested interests in search engines you know I mean they don't even reference it they just simply criticize James Li Hill and uh being in fact was the first search engine that actually got a copy now I I got a copy of the lighthill report by a completely human way of asking on the net as opposed to using search enges but um you know it went it went for a long time I watched um google.com as opposed to google. stanford.edu uh ultimately um lower McCarthy passed away basically he was one of the biggest critics of the Lio report but um it is now possible to find to find criis you know embarrassing kinds of results um now I'm not the only person you should talk to I mean there's this's a whole long uh discussion about what can go on with search engines and just before we walk in the room I was talking to you about their colleague Luis Grana who invited all the guys who written search engines but uh I'm not even sure Google is now the leading search engine since you can't use Google in China right I have I have friends in China and I asked them about B B has spoken here only within the last year and one of the things that that the people of ba do do of course is they watch very closely what first Google does and also what I mean the great firewall they know in China how to get around the great firewall yeah and they will do Google searches but they do it only sparingly in order the great firewor doesn't shut down their their their net access but you know I mean Google is far from any way shape or form oh the these things are happening in China with B undoubtedly and I I was on a panel actually earlier this year in Germany with one of the top exe Executives at Yandex which is the big Russian search engine and uh there's no question that the endex has been used deliberately uh for you know to manipulate well that shouldn't surprise us right so uh all I'm saying is you know it's not just in Russia and China yes so there's another horrible complication in this area and not another one because really remind me of your name sorry H Hal of course IBM go back one letter so yeah independ put your advertising hat back on there are a whole bunch of people that are out there trying to manipulate search results so that the people who pay them for manipulating search results you know will be happier because their ranking goes up that's competitive that's completely different that's comp no it isn't because it's it's putting it's spamming search engines yeah I understand but it's competitive M buy Power right it is no no but the point is that that the algorithms that you want to do one thing okay I want to get rid of that spam so that Google algorithm is getting good at that now we get to discuss whether it gets rid of more than spam and as I was just going to point that that the uh penal penalizing sites which try to manipulate their ranking is one of the larger elements of the search engine of every search engine but Google in particular that is to say uh if you try to use CEO they're going to pandaly you if they find you doing at least some of the more straightforward simple ones and whereas a very straightforward presentation of information uh gets ranked more highly than one that has an attempt attempts to uh to scam the uh search engine system well worst things are are coming because the page rank you know has has just been announced it's gone no no no no no there's something new because I just read it a couple days ago it can't be gone already is it gone already it's been gone for quite a while since Panda well it it's a it doesn't give you anything it's the page rank's there there's a number there but whether it helps anybody or the vast majority of searches are whatever whatever is going on the point is we don't know we it's not transparent we have no control of it we have no say and there's there's a lot of there's a lot of power and influence here more than than any small group of people in the world has ever held before and in the US uh I know from some of the people I spoke to in Washington unaccountably mysteriously when we investigate something almost always and there have only been maybe two two significant exceptions almost always the investigating agency pulls back mysteriously and so I asked about why why does this keep happening and they said well you know campaign contributions I said wait no these These are these are career you know bureaucrats in Washington these aren't they don't run for they're not elected he said no but they're appointed and you know I I think there's a serious problem here because the reason why the EU and why India and by the way now also Brazil and Canada why they're being aggressive is because they view Google as an American company this is true yeah I can I can attested that yeah but here we like that we like that and we love Google we all love Google Google is awesome I mean their toys are incredible right and they're getting more incredible by the day and their their PR is amazing and right so we are probably you know if there really is some danger here the danger is probably here more than anywhere any other questions I think we're this is Back To Top