Facebook and Insurance
Chris, Peter and Fraser discuss the insurance industries use of social media data to assess risk in individuals.
For more information on Aleph Insights visit our website https://alephinsights.com or to get in touch about our podcast email podcast@alephinsights.com
Transcript
Hello and welcome to the Cognitive Engineering Podcast produced by Tell Me Studios for Aleph Insights. In this series of podcasts we take a look at interesting topics and discuss what we think they tell us about analysis and decision making. I'm Fraser McGruer and I'm here with Peter Coghill and Chris Wragg of Aleph Insights and this week we're discussing how your use of Facebook can affect your insurance premiums. So Chris can you lead us in and tell us about how insurance companies are starting to use Facebook
Speaker B:information? Yeah so there's some UK insurers now who are starting to look at the text that people put on their Facebook pages and start doing analysis of that text in order to make decisions about what people's insurance premiums should be and I think this raises a number of or a couple of particularly interesting areas to do with what your language, your use of language, tells you about an individual and the extent to which these new types of assessment of individuals are just actually masking other ways of assessing people more traditionally and to what extent they're actually a new way of looking at individuals. So I think there's a couple of interesting
Speaker A:issues that come out of this. Okay, anything you want to come in at this point with Peter?
Speaker C: l just to re-echo that, since: Speaker A:relatively simple going on here which is just about availability of information and the tools that insurance companies and ourselves have at our disposal, they're just quite simply using new technologies which make more information available. That's what it sounds like to me. So Chris, you said this makes you think of certain questions, well let's go into that,
Speaker B:can you develop it a bit further? Yeah so a little bit, so I think they were looking when they were doing this assessment of Facebook content, they're keeping the algorithms quite secretive but some of the things they are looking for are first of all how well you use English, how good your spelling is and your grammar and your punctuation and the other thing they're starting to look at is the way you use language so how particularly looking for things like overconfident language as a marker of somebody who might drive in an overconfident way and so the first, going back to the first point, it raises this interesting issue of you know when you're doing analysis there's lots of text analytics going on at the moment and when you're using language you know when you start assessing for things like punctuation and spelling what are you actually doing there because you know traditionally we have standardized spelling and punctuation in English and the reason given for that has always been about the conveyance of meaning and yet you know I would argue that actually it's not really about the conveyance of meaning you know when somebody puts an apostrophe in the in the wrong place if you've got a sign outside a grocer's or something that says you know orange apostrophe s 50p oranges 50p you know nobody thinks that's about 50p belonging to an orange so but some people are very very fixated on the correct use of apostrophes and that this is essential and I think that's quite interesting because you know what it essentially is doing is it's acting as a signal for something else about the person you are making a judgment but you know you understand something and you know what the marker is and you use that to make an assessment about somebody else are they in your club or not of people who use apostrophes uh correctly and I think that's a um that's an
Speaker A:interesting piece of information it yeah well I guess sort of you know in the case of what we're talking about there is I mean the two things if we're looking at behavior on Facebook and the language people use that could be signaling one thing but if we just focus it just for a moment this slight detour if we're focusing on grammar um which this is let's call it a grammar mistake to me that says that I mean I was a teacher for years and two things I felt was that one certainly because I was teaching English English is just such an unusual language in some ways in the way this grammar works and it's so irregular as well um but also the second thing I realize is that teachers can fall into English teachers can fall into the habit of teaching they think they're teaching English but actually they're teaching grammar and grammar is is arguably a false construct okay and it's not really um yes it does help to improve communication but really it's a set of rules that we've um defined around language well so yeah so what I was going to say basically the signal here is possibly how well educated someone is um or and or um they just might have certain problems they might be dyslexic for example if we're looking at spelling let's say and so yeah so I try because whenever I see oranges with apostrophe it always makes me sort of go as well um but I've sort of have to sort of train myself go well look you know it's just doesn't mean that's a bad person all that means is that I don't know they haven't I don't know how to say this without sounding condescending but they just might not be very well educated let's say which
Speaker C:is not their fault necessarily anyway um Peter yeah I think that's an important point to really hit home that a language is just a convention um and um it's so there's no absolute rightness in language and language has always been extremely evolving always changing and always things adding and falling out of use so um there is no absolute correctness so I think there's an interesting moral and ethical uh question around how you know assessing people's goodness as a proxy for how risky they are based on their use of language which should be there as something to enjoy in life something you can play with something you can even tink with and you used to your you used to get across whatever meaning it is you want to get across so it's you're not measuring people sort of rightness or worthiness you're actually measuring their conformity to any to an arbitrary
Speaker B:standard yeah and I think I think without you know wanting to get all uh Orwellian but um you know controlling people's use of language is you know on some level an attempt by society to control them and if you look at um you know sometimes standardizations have become quite uh politicized like if you look at the difference between UK you know British and American spellings for example that was that was driven that divergence was driven by you know a political sort of um necessity and uh uh if you you know if you look at the history of of um standardized spellings often they have not led to clarity but um it's about elitism so like words like debt used to be spelt d-e-t but there was this drive to make British uh English spellings um closer to to their Latin roots you know uh which is clearly a form of of elitism you know to some degree so a lot of the standardization around language is a is about isn't it is as I said not about the you know conveying meaning it is about um creating creating barriers for others for others to fail at and so I think the use of that in insurance as Peter said is actually quite quite an interesting I think it's interesting though
Speaker C:because this is this is something that people you it's very difficult not to do so when you meet somebody who with a foreign accent you instantly judge them in some way no matter how hard you try not to you you are inherently fairly xenophobic as a human being you're designed to favor your in group versus the the out group um so you you and and nice cases when you when you meet someone often quite young uh who use words inappropriately they use the wrong word to mean something um uh and it's funny I literally can't stand that yeah literally can't stand that and it's sort of it's entertaining and so you're you're you're you're finding it funny you're you're you're belittling them in some way um but they all but all but I think you know there are genuine you know there are reasonable excuses for this so people there's lots of sort of often referenced in literature Dickens and things talk about the the self-educated and the way you you can pick them out is when they they mispronounce a word because they've not received that word people they talk to don't use that word but they've tried to work out how to how to how to how to um announce that word um from from text yeah and I think you know particularly in in
Speaker B:this country with with our class system you know pronunciation is uh is a really interesting you know um feature there's a there's a great um observation by um the comedian David Mitchell about uh the word valet and uh he does this whole sort of thing about um the fact that uh actually if you had had a valet in you know the Edwardian era you would have said the word valet that's how you would have pronounced it but he can't bring himself to pronounce it the way that it would have been used because he's then scared that people who who well people people will think he's making the the you know the the most basic error as opposed to actually being
Speaker C:more sophisticated and I think the people who who can afford valet valets actually still call
Speaker A:them valets yeah there's another point here which goes back to class which is about being middle class which is this would be a very middle class uh sort of conundrum right because if you were actually upper class you wouldn't you wouldn't care no one's right anyway um so we've sort of talked a lot there about language and grammar etc but this let's bring this background um um to insurance now what strikes me here is you can say well look in some ways this isn't fair it's an unfair judgment etc if we're talking about um misspellings and bad grammar etc however um just from a from a pure let's say commercial point of view um it's a benefit to the insurer if they use whatever tools that are at their disposal um to better price premiums and by extension it should be better for the consumer because everyone should be interested in more accurate um premium assessments so um where do we go from there yeah well I think it's I think it's
Speaker B:an interesting point um and uh you know in many cases I suspect what they have found is proxies for the things they are no longer allowed to discriminate against so overconfident language uh is probably an indication of you being a young man for example uh because you know that's that's how you're sort of conditioned to speak in in the circles you know with you with your
Speaker C:peers um one of the one of the other major drivers which is often uh slated is is the use of occupation because occupations are um they are very polarized often between genders um so if you say you're a civil engineer uh your insurance premium will be much higher than if you say you're a nurse just simply because um your uh your your given profession will have be overridingly
Speaker B:one gender over the other and I think the thing that that people uh object to is um uh sort of um inequitable uh having to pay more for something that they are unable to do anything about so if if you just found a proxy for somebody being young and male there's clearly nothing that that person can can do about that um whereas if you are charging on the basis of some demonstration of of the risk so their individual behavior which which is where I think you know these uh things that monitor a driver driving behavior have um become uh much more popular then then I think people are more prepared to to accept okay you know my driving behavior shows that I'm more likely to crash I'm I'm happy to pay more and not happy but society is happier to judge them on that basis as opposed to something
Speaker A:fixed that they can't change okay but I mean what I would sort of my immediate response to that would be well you may not be able to change it but that doesn't make you any a safer driver so that young
Speaker C:men are more dangerous I think the objection is that statistically it doesn't make you a safe driver but there are plenty of safe young men um so I think that and that the objection comes from you're putting your you're tarring everyone the same brush um so the the the sort of there are two kind of extreme ways you can make it totally fair in inverted commas one you could everyone pays the same rate and it's only modified by the things you can totally choose so not many things left but the value of the vehicle you can choose to drive a cheaper vehicle or you can choose to drive a more expensive vehicle whichever vehicle you can afford so or you can the other extreme is you collect lots and lots of data about people's in the individual's behavior and their policy is entirely tailored to their historical behavior and I think that this is this is happening more and more and more so Aviva have a rate at your driver rating app which collects data about how quickly you accelerate and how quickly you and how fast you brake and where you go and these are all sort of um these are much more choice driven behaviors than things you you're lumbered with at birth but isn't it a question of and you're kind of saying it there
Speaker A:isn't isn't it just the case that here's another set of data that um an insurer can add to the other data points that they've got as you've just mentioned and it's just another aspect which they can throw in there which can make it even more accurate and I think the issue I think this gets
Speaker B:to the heart of um uh the ethics around data data analytics because um yes you're right it and and many people just treat it as well look we're not we're not making traditional judgments that would be considered prejudice we're just looking at the data so it's objective isn't it so we're you know it's it's colorblind or gender blind or or whatever but not if that data is very highly correlated with um you know those things that you're trying as society is trying not to be prejudiced uh against and and so car insurance is one thing which is um you know reasonably contentious but but you know we we probably um accept the way that it's it's it's done but I think what's increasingly alarming is the way this uh certainly in the in the U.S. um judicial system is being uh background data is being used to help um judges reach decisions about whether or not uh to um whether or not an individual to help them guide the sentencing of whether or not somebody's likely to reoffend or not uh and so under those circumstances um you know if you are making if you are making judgments that are based on supposedly objective data but data which is tied very closely to somebody's ethnicity for example uh then that is something we you know we presumably
Speaker A:want to have a look at um okay so that's interesting so immediately probably intuitively I would imagine to most people there's some sort of line there that's crossed and so I wonder what that line is because if we're talking about insurance and we will say okay that sounds fine and someone pays more or less than and broadly speaking people I mean I would go well that's okay that's fair however when it gets and even though you're talking about a future possibility as soon as that future possibility becomes someone um being locked up for longer or executed which is possible in the U.S. I guess I don't know if they would use it in this in this particular way but suddenly some somewhere there a moral line or an ethical line seems to have been crossed and I I have a problem with that so why would I have a problem with it with that but not with um not
Speaker B:with insurance? Well probably because the perceived consequences are are great you know incarceration versus having to pay you know 50 percent more for your for your car insurance um the consequences are greater I suppose but the I think the the point is uh data is being used increasingly to to to guide decisions uh and right rightly so and probably helping us make you know much much better decisions but I think it just highlights that there is a requirement to think very carefully about just how objective some data is when you are using it as to to indicate the characteristics
Speaker A:of an individual. So it sounds like one thing we could say is the with improving technology part of that is um an increased um volume and um new new data sources but with that new those new data sources that we have there's a new sort of we're going to new um moral ethical territory which we have to keep an eye on right is that what we're saying and is that it and we put this to bed?
Speaker C:Well I think it's interesting why why why why do you uh care more about the moral implications of data in judicial processes versus uh car insurance Fraser because I'm sure if you were going if you were applying for a job and uh you found out later on you didn't get the job because this company had deliberate policy of hiring more women or more older people or or some other fairly arbitrary thing you you probably feel quite annoyed about that but this is this is the same thing that's happening with your car insurance you're being charged more simply because you're a man.
Speaker A:I think yeah it's perceived consequences uh and you're right I mean I'm fairly unemployable anyway but um so that's why you do our podcast that's right that's why I'm here um that's yeah I mean you're quite right you know I mean I would be really you know it would it starts to get slightly you know minority report dystopian sort of um territory doesn't it um but I think I but I think
Speaker C:I think there is hope in the future I think if we are careful about the data we collect um and you collect more of the right kinds of data and deliberately exclude those data which which have strong proxies for something which you don't want to be discriminating against then
Speaker A:then the future is bright. Yeah and look I'm close to wanting to wrap it up but are there
Speaker B:any other proxies that we've not talked about here? Well I think just an alternative view point which is um you know that that there is an argument that says um actually if you're if you're trying to achieve something um like uh optimize the pricing of uh car insurance or ensure that people who are more likely to commit crimes you know um go to go to jail uh and those that um have got a greater chance of rehabilitation don't uh then there is a there is the alternative argument is that actually you should you should be extremely ruthless about your use of data and the the fact that something um correlates that is unpalatable to society with with the outcome if it if it helps you if it helps you optimize your results you should you should go down that route and I think you know in in counter to that uh that idea you know there needs to be a rationale for for why it is society is considering um not just picking the best predictors of something even if they happen to be very closely associated with somebody's gender ethnicity social class whatever.
Speaker C:Peter? Well I think the the the problem with insurance having this potential immorality to it is not because the algorithms are particularly chosen that's a sub that that is a consequence of insurance being a private enterprise the insurance is not in the business of doing what's right for people insurance is in the business of making money and maximizing their profit and the way they've done that is finding sets of algorithms that work um so arguably if if if it's collectively decided that insurance car insurance is something that everybody needs who drives a car then perhaps it should be organized extra publicly and and provided provided by the state as something
Speaker A:that you get much as other utilities are. I think for me the if if I have if it's something I have concern about this is the what what if what if I'm not a Facebook user for example is that people are coerced into a certain kind of behavior that um and this does happen already with job interviews in the U.S. and um I can't remember how Facebook profiles are used but what if you're someone who doesn't have a Facebook profile what if you've made a either a conscious or unconscious decision to say well I'm not going to be a Facebook user and then suddenly you'll put it at a disadvantage
Speaker B:but that's my issue yeah and that will be incorporated into the into the data you know it'll be a Facebook user or not not a Facebook user and that may be used to provide some level of information about you but to me the the interesting thing that people involved in predictive analytics uh should be asking themselves is um even if you could predict something more um more accurately using data should should you that's that's the question I think you know we could we could get a situation where you could very very um you know very very accurately predict something but it would be based on all sorts of measures that we as society don't want to discriminate against even if actually in terms of the data set you've got that that helps you predict the out the outcome well so so it's it's a moral question as well as a mathematical question
Speaker A:I like that and I think unusually I'd like to leave things on a question rather than an answer because I think that's a really good question just because you can do something should you just because you can predict more accurately is that something we should do um okay also unusual I feel quite serious about this suddenly and I'm quite discussing quite heavy profound stuff here that sort of um has great consequence for for all mankind or humankind that's listening so we'll stop there um thank you as always for listening to the Cognitive Engineering Podcast I'm Fraser McGruer we've been here with Chris Wragg and Peter Coghill of Aleph Insights thank you for listening until next time bye
