he was a person who created the app which downloaded all of the data for the Cambridge analytical study and he also tried to fit the statistical model that's often less reported he tried to fit the statistical model to the data now I approached him quite early on to try and talk to him about things and he wasn't very keen on talking to me at all but then I told him about my research into the efficiency of algorithms and suddenly he was extremely excited to talk to me because he said that's exactly what happened we collected
in all of this data and it just didn't work we got terrible results and so he was talking about correlation levels of nought point one so extremely bad results so the story doesn't start with maths and science but it starts right here in London and it starts with graffiti this is a painting by the graffiti artist Banksy and it can be found actually only 100 meters from here so you can go and look at it after the talk it's just just up the road you just go up and turn left or you probably have to
search on Google and you'll be able to find this this work and the title of this work is shop until you drop and it's painted on the side is painted right in the centre as you know this is the shopping centre of London so it's on bond round round Bond Street Oxford Street and like all of Banksy's art it tries to illustrate the hypocrisy in society that it was painted just in the lead-up to Christmas so the Christmas shopping rush and showing how we're falling especially this is the idea she's falling into more and more
consumerism so how does Banksy's art work well a lot of it is about positioning it's about the spatial positioning of where he painted it painting it right in the center of London really has the effect and was this idea of positioning where the art is located which attracted the attention of some spatial statisticians and it's here where the mathematics came in what these statisticians did is that they thought that they could try and identify the real bang see the key to Banksy's success is that nobody has ever found out who the true Banksy is and
so what the spatial statisticians did is they took all of Banksy's works of art each of these red dots is a GPS recording of the point that he's produced some art around London they did this both in London and they did it in Bristol where his art was first found and they produced what we call a heat map showing the likely location of where Banksy lives assuming he does his art nearer to where he lives so the white areas here these are the hotspots that he's likely to live in the darker redder areas are the
places that he's less likely to live in and the statisticians they had some locations that they thought he might live in these blue dots here and they thought that they could narrow it down a little bit more using this mathematical model so it was this mathematical model that made me start thinking a little bit about more about my own life and what I do in mathematics so when I first read the paper I was really excited I thought this is really cool this is really nice use of mathematics it's just the type of thing that
I would like to do myself if if you're an academic yourself you'll know that when academics see something else that another academic has done they get both really excited and they also get a bit jealous like I wish I'd done that in the first place and that's exactly how I felt I felt like jealous and excited that they've done this but as I thought a little bit more I started to think who do these people think they are I like Banksy and I like all of this painting and I like this idea of him sneaking
through the night doing his graffiti artwork and you wake up in the morning and you see the hypocrisy of our society revealed I really like that idea and then you have these sort of smartass statistician saying oh well there's this heat map and this is where he lives and I started to get a little bit annoyed about it and then another thing struck me or wait a minute you're a professor of applied mathematics you've written a book about soccer Mattox aren't you just as bad as this these these spatial statisticians and it was also at
this time as Jordan mentioned I'd written this book soccer Mattox and this was the sort of typical picture that I was drawing at the time what this is this is a football pitch this is or it's the the penalty area this is the goal and this is a heat map which I produced of the probability of scoring a goal at different positions around the pitch so in my heat map black gives you a very high probability of scoring if you shoot from there yellow is a much lower probability of scoring and so these were the
sorts of things that I was really excited about at the time and I was I'd written the book and I was telling journalists all about this exciting new world of mathematical football and for a large extent journalists thought this was really interesting and I got lots of requests to do interviews and when I did the interviews they'd all say yes yeah that sounds interesting but they'd always say like at the end they'd say um yeah there's something though that we'd really need to ask you David there's something we really need to know it's our readers
who want to know this they want to know isn't mathematics taking the passion out of the game like all the fans they don't want to know all about this statistics mathematics football is all about passion Chi aren't you taking the passion out of the game by doing it mathematically and I would be no no look at this heat map you know it's really exciting do you want to see a passing Network and and and they go okay okay yeah that's fine okay we'll write that no problem okay but I began to have those feelings myself
that maybe I was taking the passion out of things by producing mathematical models and here's another place where they're very passionate about algorithms and another opportunity I got in writing so chromatics was to go to google google here in london and talk about the mathematics of football and they were just as enthusiastic as me if not even more enthusiastic they were can't we put heart rate monitors on them so we can monitor them like 24 hours a day and so you know where they're moving around and so and then on the pitch we can have
hundreds of cameras and GPS and all of this stuff and they just thought this was brilliant there was no one there that thought they were losing the passion in the game and what Google are doing and it's not just Google but Facebook also and Microsoft is they want to try and use algorithms and mathematics to understand every part of the world that we live in and they're what they were talking about at Google for example was using medical records to predict when you have liver disease they're using techniques to improve energy efficiency they've also done
it a lot of work on artificial intelligence showing how you can solve games using artificial intelligence there's all this research going on at Google all enthusiastically trying to in their view make the world a better place by using mathematics and algorithms and I think that's good but there are problems here there can be problems with algorithms and a lot of people are worried about algorithms and I suppose I hadn't had those worries before but I want to start by identifying what I think is the biggest problem with algorithms and you can already see it directly
in the Banksy study so this is an interview with one of the spatial statisticians who did the study of Banksy and he said this it rapidly became apparent that there is only one serious suspect and everyone knows who it is if you google Banksy and I didn't want to put the name of suspect you get something like forty three thousand five hundred hits and I'm sure that number of hits has increased since then but they already knew who Banksy was and the reason they knew who Banksy was was not because of any mathematics it was
because the Daily Mail are obsessed with finding out who Banksy is so for over the last twelve years the Daily Mail have written hundreds of articles about who Banksy might be and identified various different suspects and they've come up with one very good suspect I had some fun finding Daily Mail headlines about Banksy so this is scrap your boss thinks he's filmed the elusive Banksy in his Mercedes van and then banks he's caught red-handed this one is my favorite graffiti artist Banksy unmasked as a former public school boy from middle-class suburbia totally shocking they say
he's not a working-class hero that our graffiti artist should be and then of course the mail picked up on the spatial statisticians article they wrote scientists confirm mail on sunday got Banksy's identity right and it's here you can see the problem the spatial statisticians got their information about who Banksy was from the Daily Mail the Daily Mail then riah headline saying that science scientists confirm mail on sunday got Banksy's identity right and it's this feedback between the people who are were the people who are doing the research and the people who are managing to do
the investigative journalism where they confirm each other but they haven't really confirmed each other where did they originally they get the original idea it's it's not necessarily the algorithms that found Banksy so that tells you there's a limitation in if the rhythm really did find Banksy but then you see in the scientific article they wrote the following statement they said that this supports previous suggestions the analysis of minor terrorism related acts eg graffiti could be used to help locate terrorist bases before more serious incidences occur now this was written in a scientific paper and this
is really scary right so what if we find that there's been some sort of graffiti written about that we should assume that these people are going to do terrorist acts and we can we should start to arrest anybody who's done some graffiti and we can track them all down with these algorithms and the point I missed on the previous slide was this algorithm said that if you look at the white areas about 20% of Londoners are potential terrorists so that is a bit of a scary thought and I don't want to be too hard on
on the scientists because I think scientists well I write scientific papers too and sometimes we do get a bit carried away and make draw conclusions that we shouldn't have done but what's really important here is how these algorithms work and that ended up being a lot of what I spent some time investigating what where I wanted to start I I started I when I investigated an algorithms I analyzed three different steps first I looked at how they analyzed us then I looked at how they influenced us and then I looked at how they might eventually
become us and I'm going to start with the part about how they analyzed us and what I wanted to start with in order to really understand how these things worked I actually started by analyzing myself so this is my Facebook page I've revealed it now to the world the picture there is slightly old that serve my daughter is now 15 she's a small one there and my son is now 13 so I don't carry him around on that baby more and so it's not really up today I'm not a big Facebook user but I noticed
that lots of my friends are big Facebook users so this gives me a great opportunity to analyze some people's data so what I did you're probably all glad where some of you are my facebook friends actually you're probably glad that you're not my facebook friends now because what I did is I took my Facebook friends and I put them into a matrix and what you see here these are a list of 30 of my Facebook friends this one's my wife is sitting there the others aren't present so that's good no I I'm joking I have
asked for committed well some of their permissions anyway and what's shown here is different categories which they post about so what I did is I took all of the posts that they made and I categorized them to see what they were about so here we have Camila she posted five times about work she posted one thing about a product of advertising one time about culture and sport and for every person I noted the number of times they posted about different things what I wanted to do was see if I could break down and understand these
my friends in the way an algorithm would break down and understand my friends and my aim was to reduce them to two dimensions and so I used a technique called principle component analysis where you basically rotate all of the data round and you tried to find the best way of looking at that data and you try and reduce it down into this two-dimensional picture of the data and the two-dimensional picture I got of my friends looked like this so what we could see is that there was a sort of there was two axes one of
the axes was a public versus personal axes I had some friends who posted a lot about their friends and their family there over here I had lots of friends who posted mainly about either work or what they did or they about news and current affairs and so on and so the work people could be put up here and the current affairs people could be put down here so there was two axes workplace further up culture further down public further to the left personal further to the right now the important thing here is that this isn't
my classification of my friends this is the classification I get by taking my friends putting their posts into an algorithm and asking the algorithm what is the best way to view David's friends and the answer it came up with isn't unreasonable afterwards I asked these friends is this a good classification of you do you use Facebook mostly in this way do you use it mostly for your family and they agreed that this was a very good analysis of how they behaved on Facebook so a simple algorithm this principal component analysis could break down my friends
to two dimensions and could understand and break them down into groups so that's how the app that's how the basic algorithm works at how it works on my friends this is what happens for Facebook this is how Facebook see the world they have all of the world two billion people on Warren axes then they have hundred million likes of various different things on another axes and their task is to reduce this massive massive matrix of data to some form of understanding of all the people around the world I love some of these categories there's so
many of them you know there's yet we've got feminist Peace Network rickrolling big butts and I cannot lie every type of category of anything you can find people are sitting there and they're just clicking like like like like on everything that they supposedly like and filling up a matrix for Facebook of what all different types of people like now we can't expect all of these hundred million likes just to be reduced down to two dimensions so what Facebook do is they reduce it down to tens of thousands of dimensions so they take these two million
people and they reduce them to tens of thousands of advertising categories and you can actually go into Facebook yourself it just takes a few minutes to find out what type of advertising categories they have some of them I thought it's the wrong wedding I know I've been upstaged my arrival here by the Royal Wedding on Saturday so I took a few ones that I thought it would be quite nice with for the royal wedding so the size of these is proportional to how many people are accurately classified by them so about 60 million people around
the world are interested in the British royal family I didn't think about this but 60 million that's the population of the UK isn't it so you're all obsessed with the British royal family but well a 60 million people around the world are interested in the British royal family there's about 5 million people who Facebook have classified as upper-middle class for some reason Britpop has 40 million people so there's 40 million people that Facebook of label does Britpop and then there wasn't one for the upcoming wedding but there was a category for people interested in the
wedding of Prince William and Catherine Middleton and there was about 20 million people or so who were classified in that way and those categories are inside Facebook which you can and you can take them out and have a look at them I think one of the these categories we recognize we understand what they mean but there are other categories which are a little bit more difficult to understand so these are four of my favorites toasts tugboats neck platypus toast I think I found out actually my son is probably in the toast category he sees a
lot of adverts for bread products when he's playing computer games they've made some correlation between computer game playing and having a sandwich after you finished playing computer game and I think probably my son would be in the toast category tugboat I just have no idea why and that that's like 10 million people finger in the tugboat category deck that could be back pain and so on that's the 70 million category and then platypus that's a couple of million people are in the category platypus now of course these categories are funny and but in a sense
the joke is on us because there's 10,000 categories there and they are a genuine statistical understanding of us as individuals so just because we don't understand why they've classified US as platypus doesn't mean that there isn't a genuine statistical relationship which is best characterized by the word platypus so when we we always want to put words on things to describe things but actually inside the Facebook computer there's no words there's just these dimensions and one of those dimensions is best described in words as platypus and I think we can yeah so Facebook understands us in
10000 dimensions we understand each other we tend to understand each other in terms of maybe our gender our race or class and maybe to some degree we understand each other in times of terms of personality we talk about each other's personality this is the way that psychologists understand us they have come up with one of the theories that's predominant in in psychology is this idea of the Big Five personality model these are openness conscientiousness extravagance agreeableness and neuroticism so it's ocean and this has become over the last 30 years or so this has become quite
a standard model that people use to that psychologists use to classify people and that means that they're working in five dimensions so they take questionnaires where you fill in maybe a hundred questions these might be questions there are questions about yourself about how other people perceive you and they take those questionnaires they reduce the number of dimensions to five and then they can tell you on those five axes what is your personality now this guy here Miquel Kaczynski when he was a PhD student at Cambridge University he came up with what can only be described
as a genius idea he came up with the idea the following idea he knew all about Facebook and all the likes that were going on there and Facebook had just started so this was 2007 there wasn't very much to do on Facebook in 2007 so he came up with the idea of making an app where you could do a personality test and if you did the personality test you would get the result and it would tell you you're neurotic or you're agreeable or something like that and this would be a fun thing to do so
these are the questions that he'd asked have you do you have a vivid imagination you could fill in do you hold a grudge and so on you filled in these hundred questions and then at the end of it this was the smartest part he said would it be okay if if we had access to all of your social media data for research purposes and everyone was just like yeah that sounds great yeah yeah that'll be fine okay there you go and four or five years later he had several million people's personalities they also answer questions
about their gender their smoking habits their sexual orientation everything that that he asked them they just clicked into clicked in happily clicked in and he had this massive database of personality and how people what people liked on Facebook then he came up with this really clever idea then to and this was a paper that came out in 2015 this is the personality of various people on these five dimensions these are the Facebook Likes of various people on tens of thousands or hundreds of thousands dimensions you can take both of those matrices and you can link
them together by creating a statistical model so he took these and he could predict the personalities or he fitted a model which predicted personalities based on the likes that people have made and once you've fitted the model you can start to predict on other people so when US mathematicians do something like this we take 90% of the data to fit the model to try and understand people we then take the rest the 10% is left we take that and we try to predict if we could actually get people's personalities right and it turned out he
could get people's personalities right and this is some of the examples so people who are outgoing they like dancing theater and beer pong I actually still don't know what beer pong is I have to find out it is really important if you're a student in America or an outgoing student in America if you're shy so this is really a sorry but this is a list of painful stereotypes so if you're shy you like anime you like role-playing games and you like Terry Pratchett books if you're neurotic you like Kurt Cobain you like the emo lifestyle
and you say sometimes I hate myself and if you're calm this was a bit of a surprise for me if you're calm you like skydiving every day it's supposed to you have to be quite calm if you're going to jump out of a plane you like football so yeah I'm in that category and you like Business Administration see one of the points here is you don't like you don't have to like all of these things at the same time yeah you certainly don't do all of these things at the same time but different those are
the types of signals of somebody who is calm neurotic shy or outgoing now this work by Michael Kaczynski it caught the attention of the following person this is Alexander nix and he was until recently CEO for a company called Cambridge analytic ah which probably a lot of people have heard of or it's very hard not to have heard of them and he gave a presentation in and 16 he was working then on the Ted Cruz campaign for the presidential primaries and he then later went on to work with Donald Trump in the when Cruz dropped
out of the campaign and in this presentation he made a very powerful prediction of how they could use Facebook data his idea was well we can actually predict the personalities of different people and that's going to be really powerful in an election so if you can identify the personality of a neurotic person and that person might be interested in voting for the Republicans and you might take up an issue of guns then you can say when neurotic person is like protect your family with a gun and then if you if that well if you find
the person isn't you neurotic maybe there are more traditional person you can say hand your gun down from father to son and these become the different targeted adversan allottee and you target their that personality with it with a particular advert that was the sort of dream that alexander nix presented for cambridge analytic er and you can see sort of why they ended up getting into trouble later because this wasn't exactly the sort of thing that everybody thought was a brilliant idea but what I've actually found out in when I was writing the book outnumbered is
this is really scary what nicks does present this idea of targeting personality is scary but the problem is it doesn't work and this is the reason it doesn't work so this is some of this is a small subset of the data used by Michael Kaczynski and what's plotted here is neuroticism measured in a cycle psychological test so that's when you do this this ocean test you fill in the questionnaire and what's here is neuroticism predicted by Facebook data and each of these dots is one of a hundred people from the original study but the model
was actually fitted on tens of thousands of people it's not just fitted on these hundred people but they illustrate it here and you can see if you look carefully there is a trend in this data and there's a correlation coefficient here of naught point three so for a scientist that's enough to say there is a statistically significant trend here and that's where Michael Kaczynski could quite rightly publish a good scientific article claiming a relationship between these two variables there is a trend that goes like that slightly more often people with who are have low neuroticism
have a low score and higher woops hi you tinnitus ISM have a high score then but there's lots of exceptions to this rule here are a bunch of people who Facebook thinks and neurotic but when they do a neuroticism test aren't neurotic at all and here are a bunch of people who Facebook thinks are not neurotic but when you do a psychology test on them they're very neurotic so there's a scientifically reliable result but that isn't enough to actually build an algorithm to identify people's personalities and so I found that personality algorithms they just don't
work so what you can do if you know Republicans and you know Democrats you can reliably predict a Republican and a Democrat Democrats tend to follow Barack Obama for example Hillary Clinton so you can identify them and there are some subtle things as well Democrats love Harry Potter for some reason when Republicans aren't so keen on Harry Potter so there's a few subtle things you might not have thought about Republicans apparently love to go camping and they like camping a lot more than Democrats so there's some good signals there to pick out your Republicans and
your Democrats but it only it predicts more neurotic or less neurotic people only with sixty percent accuracy so if I picked out two random people in this audience looked at their Facebook pages and tried to predict with my algorithm which of as most neurotic I'd have about a 60% chance of getting it right and of course if I just said randomly or you're the most neurotic one I'd have a 50% chance of getting it right so it isn't particularly accurate algorithm is scientifically it's it's statistically sound but it's not enough to build an algorithm then
there's another problem is this protect your family with a gun thing that I'm so keen on is protecting your family with a gun neurotic is very different from nirvana emo lifestyle neurotic I mean you don't you don't go up to a Nirvana fan and say protect your family with a gun it's just not going to work they're not going to respond to that thing so the personality there's a lot of subtleties to personalities which aren't captured by these types of algorithms now when I I started working and investigating Cambridge analytical there being quite a few
articles in the newspaper already but I started about a year ago looking at whether their algorithms really work I'd read a few of the Guardian articles about it and I started looking into it but then for some well in March this story really broke on a massive level and it broke when Chris Wylie came out and said that he was the whistleblower he where he he was the the guy behind Cambridge analytical or was involved with the development of their algorithm and he revealed all the documents about how the inner workings of Cambridge analytic are
now I think it's brilliantly he revealed all of these inner workings but I do have a problem with this headline I made Steve Bannon psychological warfare tool now this warfare tool or this weapon that he created it can't have been a particularly powerful weapon because it's based on exactly the principles that we've talked about it has all of the limitations that we've talked about before he it is going to misidentify people it is like a sort of random targeting weapon that will go out and do lots of random stuff and influence people in lots of
different ways but it isn't a sort of dangerous psychological profiling weapon that they might have thought that they were creating if anybody did create a psychological warfare weapon it was this person here this person is Alex Cogan he also worked at Cambridge University and he was the person he was a colleague of Michael Kozinski's and he was a person who created the app which downloaded all of the data for the Cambridge analytical study and he also tried to fit the statistical model that's often less reported he tried to fit the statistical model to the data
now Kogan I approached him quite early on to try and talk to him about things and he wasn't very keen on talking to me at all he'd been accused of being a Russian spy and like vilified in all sorts of different ways on various blogs and so on and so he really just didn't want to talk to me but then I told him about my research into the efficiency of algorithms and suddenly he was extremely excited to talk to me because he said that's exactly what happened we collected in all of this data and it
just didn't work we got terrible results and in fact Michael Kozinski's data set was bigger than the data set they had so they didn't even get as good results as him and so he was talking about correlation levels of nought point 1 so extremely bad results and he told me that he thought that alexander nix the head of cambridge analytic ah he just had very little comprehension of what he was talking about he told me he's trying to promote the personality algorithm because he has a strong financial incentive to sell to tell a story about
how cambridge analytical have a secret weapon so he was this was the guy who tried to create the psychological warfare tool and had now come to believe there was no way that he could create a psychological warfare tool one of the issues I have is about how these sorts of things are reported in the media because this is how Kaczynski's original paper was was reported in the media how Facebook knows you better than your friends okay that's very enough Facebook knows you better than your members of your own family then this one's my favorite New
York Times Facebook knows you better than anybody else and so you just got this going to extremes of how well Facebook understood you and really the result was you have a correlation of nought point three between a personality test and a principal component decomposition of your Facebook Likes so it really wasn't a relationship of anything close to this level and what's what's similar I think between the Banksy story I told you at the start and this story is that there's a scientifically interesting result but there isn't a thing that you can later turn into a
algorithm which actually affects people and that turned out to be the case for a lot of the different algorithms I looked at I'm going to tell you a couple more stories shortly but it always turned out that the algorithms the scientific results didn't quite translate into the algorithms that might affect our lives so algorithms try to influence us and you probably can't have escaped if you've heard about Cambridge analytic oh you can't have escaped hearing a lot about fake news so the other big thing about the Trump election was fake news so here's four P
four pieces of fake news so Clinton Foundation staff were found guilty of diverting funds to buy alcohol for expensive parties in the Caribbean Mike Pence said that Michelle Obama is the most vulgar first lady we've ever had leaked documents reveal that the Trump campaign planned a scheme to offer to drive Democratic voters to the polls but then take them to the wrong place at a rally a few days before the election President Obama screamed at a protester who supported Donald Trump now does anyone remember any of these fake news stories to many people but these
are all fake news stories about the election but there's actually a difference between them two of these fake news stories are if you can really call them this they're real fake news stories two of them are fake fake news stories so two of these stories were made up by researchers to show to people as a placebo treatment in a fake news experiment and now I'll show you which one's a which the top the top one is a fake fake news story it was made up by researchers as is the third one and then there's two
here these are real fake news stories that actually appeared on Facebook and on various websites and what the economists who studied this did it was a really clever experiment they showed both the real fake news stories to people and they showed them the fake fake news stories and they showed them some true news stories - and here are the results so if they showed them a big true story they would usually remember the story if they liked very true true stories that were big in the news these are small true stories minor news items they'd
often remember them but over here this is the fake news stories and about 15% of people remembered said that they remembered a fake news story but 14% of people said they remembered a fake fake news story and there's no statistical significance difference there basically people couldn't tell the difference between fake fake news and real fake news and so the possibility are those fake news spreads a lot and people are sharing this stuff all the time nobody can remember it they remember the sort of vague idea of the content of it but they can't remember the
details of it looked a few other things and this is the echo chamber effect now I found to my great relief that it turns out that I don't live in an echo chamber all right well maybe everybody maybe says that but I did a study to see if I did live in an echo chamber this is this is me the center here on Twitter and all of these circles here are the people I follow and they follow me back on Twitter so we talk to each other on Twitter and what the coloring is it shows
if they were more if they were closer to following a pro brexit newspaper or a pro remain newspaper so darker people were more Pro remain and whiter people were more Pro brexit now up here this is an echo chamber this is a cluster of people who I follow and follow me back and follow each other these people are all scientists and they tend to be there's one exception here one contrarian of course but nearly most of them tend to be Pro remain and they are kind of an echo chamber you know we just but they're
an echo chamber in a really nice way it's just moaning about getting your paper rejected and those sorts of things so they're a nice echo chamber I think but it turns out I've got lots of non echo chamber connections I've got connections which go in lots of different directions and some of those people are quite on the pro brexit side of the newspapers that they follow so there is this opportunity to get in different information and myself and a student we did a study of course I can't base the whole scientific theory just on my
own Twitter graph so we'd look we downloaded lots of different people during the brexit so just after the brexit vote and we found that this was very typical of the structure they'll have this group of people who maybe are invective echo chamber but they have all of these nice links that go out to different places where they get different types of opinions coming into them so most of us really don't live in an echo chamber in fact Facebook were very quick out to do a research of this I mean you can or you can't trust
Facebook's research but it was published in the best scientific journal in the world science and what they showed is that people tend to be exposed to so these are these are conservative people so Republicans this is if they took their news at random they would get 40% liberal type of news and they almost get they get 35% liberal type of news so conservatives see a lot of news which disagrees with their point of view liberals they tend to be in a little little bit more of an echo chamber liberals see more liberal news than they
see conservative news and that's because they share liberal news with each other so they don't see as much conservative news as if they just pick stories that random so though so most of us don't live in an echo chamber although if you're a liberal then you're slightly more likely to live in a bit more of an echo chamber than the typical conservative person in general I don't I don't want to overplay that because my overall point is that most of us don't live in a you know much of an echo chamber okay so finally I'm
just want to say one last thing about the last part of the book I wrote I spent a lot of time talking to researchers who were looking at artificial intelligence so both Facebook and Google and for some reason artificial intelligence research seems to be blue so when I went into their website all of their logos were blue so blue is the color of artificial intelligence and all of these places are working on artificial intelligence and I think I'm tempted to tell you that you have to go and read the book in order to find out
the risks of artificial intelligence the what I did is I downloaded a lot of the tools that they use and I went through them one by one and tried to work out how intelligent are these tools and yeah they weren't that smart or they're not as smart as maybe a lot of people think they are a lot of them are mathematical methods that we've known for about for some time applied in it in a new type of way and is quite fun to find out the limitations of these types of artificial intelligence I'm one of
the people who believes that artificial intelligence is a long general artificial intelligence is a long off I want to finish by saying cuz I've moaned a lot about Facebook and Google and so on I want to just finish by telling you about a couple of three of the good guys in these stories this was a person who really impressed me this is Julie address or she's an undergraduate student at Dartmouth College and what she did as her undergraduate project is she deconstructed an algorithm called compass which is used in sentencing decisions or parole decisions in
the United States and she actually managed to do an experiment where she compared how this algorithm worked with just how members of the public answered these same sense of sentencing questions they got hardly any information they just got the basic description of the crime and had to make a sentencing decision and the algorithm came up with very similar recommendations so she summarized her main findings for me she said what I found was that a major commercial software that is widely used to predict recidivism that's that reoffending is no more accurate or fair than the predictions
of people with little to no criminal justice expertise who responded to an online survey so this computer this sophisticated computer model was no better than a bunch of randomly recruited Mechanical Turk O's who were paid $1 to do this task this is another person who I met a lot of respectful this is Glenn McDonald and any of you use hoops Spotify you use Glenn out Glenn's algorithm he is the guy who decomposed the whole of music and created music genres which Spotify used to tell you what you're going to listen to next and he was
an amazing person to talk to he was really kind of emotionally involved with the music that he listened to and because of his emotional involvement with the music he told me that when he was given when he was in a start-up and Spotify bought this startup and when Spotify bought it they asked him he has to be called a data alchemist in data scientist because he sees his job not as searching for abstract truths about musical styles but it's providing classifications that make sense to people this process requires humans and computers to work together so
that was very much how he emphasized it for me you shouldn't think of it as data science you should think of it as some form of data alchemy there's going on in a lot of these companies and finally I wanted to say a little bit about Joanna braceland she did a study breaking down gender differences in how we use languages so she took Google's algorithm for classifying languages and she did things like the following things so my name is David of my generation the most popular female name is Susan if you try and find out
inside this algorithm where we're closest to it turns out that I'm much closer to the word intelligent than Susan is and Susan is much closer to the word resourceful than I am and there was an inbuilt gender biases with inside these algorithms but she was kind of she was very interesting because she was very skeptical about artificial intelligence and I thought it would be nice to end my own presentation on what she said because I thought back to what she told me and I'd asked her about the rise of algorithms and whether she thought they
would become as smart as us and she told me I was asking the wrong question we've already exceeded human intelligence in so many ways our culture has produced forms of mathematical artificial intelligence for solving problems for thousands of years from the early geometry of the Babylonians and the Egyptians through the development of calculus by Newton and Leipzig through the hand calculator allowing us to perform faster arithmetic to the modern computer the connected society and our present-day algorithm with algorithmic world and I think is that thought that I'll leave you on thank you you
