I mean, the article covers the research and all so maybe I'm just out of my depth... but I can't help feeling like there's a fair bit of tech astrology going on here.
I went to uni with the guy who didn't share his stuff with Cambridge Analytica (Kosinski). He presented the likes -> OCEAN analysis and it was pretty compelling (and frightening) stuff.
In particular, with enough likes, it could deduce things such as gender, race and sexual orientation. Obviously in the wrong contexts, this information is extremely sensitive.
They might've extracted useful information from your IP address/geolocation or your useragent, ie maybe figuring out you're accessing the site from tech company X's network. Given tech companies ratio of male to female employees, they may have been able to infer you are male. Also, tech companies pay well, so that's another information nugget about you, etc. The more signals they have, the more precise the outcome.
1) I'm self employed.
2) I'm always connected via a VPN connected to a continent that isn't my home continent.
3) I haven't connected to the internet via a "company" network in over five years.
I could infer that you are male from the way you type, and I guess you're somewhere between 25 and 35. I haven't looked up your username or anything else, it's just a "feeling".
Given millions of paragraphs of text linked to millions of known demographic and other data points (provided by Facebook), a paragraph is all a well-tuned system would need to make quite accurate assumptions about the person who typed the paragraph.
> I could infer that you are male from the way you type, and I guess you're somewhere between 25 and 35. I haven't looked up your username or anything else, it's just a "feeling".
I think you only really need one or two obvious tells for something like that. The guy you replied to used ", etc.". I think that alone puts him into the bracket you've given, so I'm fairly sure you're correct here. Looking forward to hearing if you are.
You could also do sentiment analysis on your dataset and link it to likes. Basically the more signals you have the more you can narrow down your classifications or correlations.
Btw, your inference is correct. Course, one could argue that it is informed by 25-35 males being one of the most common demographics on HN.
I remember posting a random paragraph of text I had previously created into an "analyser" on the Cambridge Analytica website - as an alternative to giving Cambridge Analytica access to my FB profile.
It then accurately deduced my gender, race, sexual orientation, age group, and other key data - all from a single paragraph.
Maybe the "genius", if it can be called that was FB only allowing likes, rather than dislikes. You have to really agree with something to like it, as the default is to not react, and when you do react, most energy is to disapprove of something (for example, negative feedback, online reviews, etc.), so if someone gives something positive feedback it must have nonmarginal significance.
That said, the conclusions are very much expected. If i know what you like, i can know what kind of person you might be.
Not necessarily. It should be fairly straightforward to build correlations like "People who like Breitbart are likely to vote Trump" (and less obvious ones) to help target specifically crafted ads to the people who are most likely to respond positively to them. It's not rocket science, but effective. Basically demographics on steroids.
Isnt this just simply marketing/targeted advertising? There is no voodoo or psychographics going on here.
Honestly, I think people are giving Cambridge way too much credit.
I have seen enough overblown promises in the big data, AI, <insert buzz word> industry to know people just make things sound more innovative than it really is.
People seem to be equating a political profile and targeted advertising with a brain ray that makes people pull the lever for a funded candidate.
Just to take one high level example: Facebook was still "thefacebook" and closed to the public by the time George W. Bush was elected to the white house for a second term. How can we honestly winkwink/nudgenudge a "stolen election" at the feet of facebook or CA when clearly liberal politics were defeated in 2000/2004 with standard political rhetoric? Why are we giving facebook/CA credit for having so much political power?
I hate to use such a broad lens for the recent incident and I hope I'm communicating what I'm intending to here. I just don't see how somehow this incident "changed everything" or even changed much of anything when you look at the history how campaigns get people to vote for their candidate.
People don't want to hear this but targeted memes didn't make people anti-abortion or pro-nazi or whatever conserative boogey-man you pick. Conservative voters are people not a pathology. You may not like their views or the way they vote but this whole "people who don't think the way I do are brainwashed" thing has to go. You're buying that argument at your own peril because conservatives aren't going anywhere.
FYI I identify and vote as a liberal. Tilting at windmills like facebook/CA feels great for the ego but I'm trying to explain that it has nothing to do with why people vote a certain way. This is all a big sideshow to assuage the tender injury democrats are suffering from a Trump whitehouse.
So if you wish then let the sideshow be spectacular. Let facebook be the scapegoat and social media be regulated. When all of that and a smug attitude does nothing to stop global warming or another conservative president in the future don't be surprised.
With the new ease of access to people with opinions different from your own, most people now require some self-consistent reason for not interacting with them and their views. The idea that other rational people can have differing views from our own is deeply unsettling as it allows for the possibility that our own views are incorrect.
Pre-social media, this was easy enough: the people with different opinions were just less educated or less informed, and once they became informed they would hold the same views that we do. Social media changed all that by bringing those with different views into direct confrontation where you could immediately connect with someone with different views to your own and educate or inform them.
This fails to work. As soon as we try that, we quickly realize that those with other views are already informed and have just taken a different judgement. So we're left with the problem: either our views are wrong or their views are wrong. Deeply concerning as it's just as likely to be our own views. The only solution then is to shift the blame - it's not our fault or people with other views at fault: it's that there are 'evil actors' working to 'brainwash' people. Therefore our views are correct, and it's just people with other views who have been brainwashed. Simple and clean, no need for self reflection or difficult questions.
Your conclusion is too closed. There are in fact people who are willing to use "nefarious" techniques to boost exposure to marginal ideas to those who may be in a position to swing opinion, or to perform damage control specifically towards those who may be swayed.
That's outside of this bigger experience we're all having about the necessity of respect towards those who hold different opinions and can demonstrate a systematic world view to get there.
I think this is an interesting perspective and worth looking into. For example, targeting democrats in 2016 by assuring them that the election is in the bag so they're less likely to show up to vote.
Though I still think it's necessary to fight the temptation to believe such a campaign is as easy or effective as our conspiratorial nature might want us to believe.
The point of this is to compel Facebook to send notifications to millions of users stating that their personal information was wrongfully used by Trump, et al. It's a campaign tactic by the Democrats: use the gov't (FTC) to force Facebook to send authoritative messages to users shortly before the election this fall badmouthing the opposition. Free advertising. Very clever. Why pay for ads when you can compel Facebook to do it for free?
I think part of the issue is the way FB allows you to use 'dark' posts that are hidden from public view, and also to set up multiple anonymous pages for your ad campaigns.
Together this means you can not only use laser targeted advertising, but show ads to one group you know will appeal, without a second group who would be put off by the message ever being aware of it.
There is also the potential to break rules on election advertising and false claims as the only people to see your ad are the small, well targeted group and an FB reviewer who is checking the ad against their own T&Cs, not the electoral rules of the country the campaign is in.
The second point is interesting and worth looking at further, but I'm less sympathetic to the arguments about this that boil down to "the Trump campaign was more clever than us and made better use of the tools!"
You could have some level of legal oversight (we've never had 100% perfect coverage) without eliminating the use of programmatic targeting.
The OCEAN model mentioned by the CA exec (Nix) was originally based on factor analysis, not ML. The key criticism of factor analysis is just that one can’t infer causality, but in an application like this no one cares about causation — only what works. I’m inclined to to think the Mercers got some ROI here.
You can also use the old fashioned stuff in combination with social media data.
>In focus groups arranged to test messages for the 2014 midterms, these voters responded to calls for building a new wall to block the entry of illegal immigrants, to reforms intended the “drain the swamp” of Washington’s entrenched political community and to thinly veiled forms of racism toward African Americans called “race realism,” he recounted.
Thats kind of my point I guess though I didn't expand on it. They are pitching this like the scraping etc was 'subverting democracy' blah blah... but my contention is they were just doing the kinds of marketing that people know has terrible accuracy and return on investment. Maybe it's the bribery and hookers that were the bits that worked.
The like to Ocean numbers in the article are quite interesting, even if a bit predictable.
One thing that surprised me though was that neuroticism was least correlated with all things sports?
LEAST NEUROTIC
“SportsCenter” TV show
ESPN TV network
Derrick Rose athlete
Miami Heat sports team
Football sport
This seems to suggest that either neurotic people are somehow uninteresting in sports, or that sports are really good helping you not be neurotic. Maybe the later is predictable as well, but that nothing else is as anti-correlated with neuroticism is still surprising to me.
That is very interesting indeed. While I believe that causality is more in the other direction, one could speculate that sports fandom is a social activity that lets people act out some ancient instincts that are otherwise suppressed in modern civilization - mostly for good reasons.
Bandwagonning the current best team could be correlated with being non-neurotic. You don't have your personal identity tied up in it, you just like seeing great players do great things. There is something crazy about investigating significant amounts of emotion and energy into rooting for a team with very poor odds of winning... (says me, who's as guilty as anyone).
For curiosity's sake I'd like to see what such models could deduce about me personnally. Google ads and Amazon don't seem particularly apt to deduce who I am.
Last time I checked what ads google was serving me, it was all woman lingerie and services to Londoner expat living in Paris. I'm a young french man living in Paris, without any particular interests in lingerie, I've been to London once 10 years ago. I'd qualify myself as fitting in the norm of young CS students.
>I'd qualify myself as fitting in the norm of young CS students
I say this in total seriousness, and with no ill intent: based on a number of CS twenty-something males that I've met in the Valley and Seattle over the last couple years, an unusual interest in lingerie is not necessarily that inaccurate of a prediction...
Which doesn't entirely surprise me anymore. Competent, self taught hackers tend to represent the upper tail of intelligence, and it seems to me that having such interests from a young age correlates with a number of unusual and perhaps less than socially acceptable interests. Which would probably be influenced by a lifetime of trouble "fitting in", something I suspect is a disproportionately familiar experience to many of us on HN.
I've been running the "Data Selfie" plug-in on Firefox for a bit over a week now. It's pretty interesting. Right now it says:
"You're a laid back, liberal female who doesn't eat out frequently and doesn't prefer style when buying clothes and is less satisfied in life than most".
It also says:
Not likely to like outdoor activities
Not likely to consider starting a business in next few years
I am a male who has been self employed most of my life, and a moderator on one of the busiest "Backpacking Forums" here in the US. So it's got that wrong, but it's true I have very little interest in "clothes style" and don't eat out much.
The "female" part is especially interesting because I'm very much a straight male, but I was a single parent with a 1 year daughter 30 years ago and I did make a very conscious effort to be a "mom" to her. That required I learn some very feminine traits like "kissing boo boos" and nurturing with hugs and kisses and lullabies. I learned how to do these things by closely observing real moms, and asking them for help, or getting my ass chewed by them when I was "doing it wrong".
At the time friends and family were fairly astounded the first few times they saw me go instantly into "Mom" mode when my daughter required my attention.
And, I haven't posted anything, or made any comments about either food or clothes. I can only assume it's my lack of doing that which prompts the conclusions regarding it but I was honestly unaware that others must do that enough to make my not doing it somewhat remarkable.
So, in that very short time period (9 hours on FB) the plugin has made some pretty insightful conclusions and missed the mark on some as well. For example, I've been a hiker almost since I could walk and almost all of my hiking over the past 25 years has been off trail "bushwhacking" into Wilderness areas and most of that has been solo. But, I rarely ever mention my hikes or my work on FB, so again, it would seem to be the lack of those that prompts the algorithm's conclusions.
46 comments
[ 2.5 ms ] story [ 101 ms ] threadIn particular, with enough likes, it could deduce things such as gender, race and sexual orientation. Obviously in the wrong contexts, this information is extremely sensitive.
Nothing more was provided other than a paragraph of text I had previously written, the contents of which had nothing to do with me personally.
1) I'm self employed. 2) I'm always connected via a VPN connected to a continent that isn't my home continent. 3) I haven't connected to the internet via a "company" network in over five years.
I could infer that you are male from the way you type, and I guess you're somewhere between 25 and 35. I haven't looked up your username or anything else, it's just a "feeling".
Given millions of paragraphs of text linked to millions of known demographic and other data points (provided by Facebook), a paragraph is all a well-tuned system would need to make quite accurate assumptions about the person who typed the paragraph.
Again, likes are not required anymore.
I think you only really need one or two obvious tells for something like that. The guy you replied to used ", etc.". I think that alone puts him into the bracket you've given, so I'm fairly sure you're correct here. Looking forward to hearing if you are.
Btw, your inference is correct. Course, one could argue that it is informed by 25-35 males being one of the most common demographics on HN.
It then accurately deduced my gender, race, sexual orientation, age group, and other key data - all from a single paragraph.
No "likes" required anymore.
That said, the conclusions are very much expected. If i know what you like, i can know what kind of person you might be.
Honestly, I think people are giving Cambridge way too much credit.
I have seen enough overblown promises in the big data, AI, <insert buzz word> industry to know people just make things sound more innovative than it really is.
People seem to be equating a political profile and targeted advertising with a brain ray that makes people pull the lever for a funded candidate.
Just to take one high level example: Facebook was still "thefacebook" and closed to the public by the time George W. Bush was elected to the white house for a second term. How can we honestly winkwink/nudgenudge a "stolen election" at the feet of facebook or CA when clearly liberal politics were defeated in 2000/2004 with standard political rhetoric? Why are we giving facebook/CA credit for having so much political power?
I hate to use such a broad lens for the recent incident and I hope I'm communicating what I'm intending to here. I just don't see how somehow this incident "changed everything" or even changed much of anything when you look at the history how campaigns get people to vote for their candidate.
People don't want to hear this but targeted memes didn't make people anti-abortion or pro-nazi or whatever conserative boogey-man you pick. Conservative voters are people not a pathology. You may not like their views or the way they vote but this whole "people who don't think the way I do are brainwashed" thing has to go. You're buying that argument at your own peril because conservatives aren't going anywhere.
So if you wish then let the sideshow be spectacular. Let facebook be the scapegoat and social media be regulated. When all of that and a smug attitude does nothing to stop global warming or another conservative president in the future don't be surprised.
Pre-social media, this was easy enough: the people with different opinions were just less educated or less informed, and once they became informed they would hold the same views that we do. Social media changed all that by bringing those with different views into direct confrontation where you could immediately connect with someone with different views to your own and educate or inform them.
This fails to work. As soon as we try that, we quickly realize that those with other views are already informed and have just taken a different judgement. So we're left with the problem: either our views are wrong or their views are wrong. Deeply concerning as it's just as likely to be our own views. The only solution then is to shift the blame - it's not our fault or people with other views at fault: it's that there are 'evil actors' working to 'brainwash' people. Therefore our views are correct, and it's just people with other views who have been brainwashed. Simple and clean, no need for self reflection or difficult questions.
That's outside of this bigger experience we're all having about the necessity of respect towards those who hold different opinions and can demonstrate a systematic world view to get there.
So, it isn't to motivate somebody to vote for a particular candidate, but to not vote, or prevent the possibility of voting.
Target the opposition, if you will.
Though I still think it's necessary to fight the temptation to believe such a campaign is as easy or effective as our conspiratorial nature might want us to believe.
Together this means you can not only use laser targeted advertising, but show ads to one group you know will appeal, without a second group who would be put off by the message ever being aware of it.
There is also the potential to break rules on election advertising and false claims as the only people to see your ad are the small, well targeted group and an FB reviewer who is checking the ad against their own T&Cs, not the electoral rules of the country the campaign is in.
You could have some level of legal oversight (we've never had 100% perfect coverage) without eliminating the use of programmatic targeting.
It was targeted miss-information - directly at the people they needed to vote.
>In focus groups arranged to test messages for the 2014 midterms, these voters responded to calls for building a new wall to block the entry of illegal immigrants, to reforms intended the “drain the swamp” of Washington’s entrenched political community and to thinly veiled forms of racism toward African Americans called “race realism,” he recounted.
https://www.washingtonpost.com/politics/bannon-oversaw-cambr...
Perfection is a red herring, as long as it's more accurate than the alternatives people will want to use it.
FWIW in retrospect I'm happy that already years ago I stripped my Facebook account to the bare minimum
One thing that surprised me though was that neuroticism was least correlated with all things sports?
This seems to suggest that either neurotic people are somehow uninteresting in sports, or that sports are really good helping you not be neurotic. Maybe the later is predictable as well, but that nothing else is as anti-correlated with neuroticism is still surprising to me.I could think of a time, around 2012, specifically before Rose injured his knee, when Miami Heat and Derrick Rose fans were not very neurotic.
I don't think "sports are really good helping you not be neurotic", but I can see how rooting for a good team might be.
Bandwagonning the current best team could be correlated with being non-neurotic. You don't have your personal identity tied up in it, you just like seeing great players do great things. There is something crazy about investigating significant amounts of emotion and energy into rooting for a team with very poor odds of winning... (says me, who's as guilty as anyone).
Last time I checked what ads google was serving me, it was all woman lingerie and services to Londoner expat living in Paris. I'm a young french man living in Paris, without any particular interests in lingerie, I've been to London once 10 years ago. I'd qualify myself as fitting in the norm of young CS students.
I say this in total seriousness, and with no ill intent: based on a number of CS twenty-something males that I've met in the Valley and Seattle over the last couple years, an unusual interest in lingerie is not necessarily that inaccurate of a prediction...
Which doesn't entirely surprise me anymore. Competent, self taught hackers tend to represent the upper tail of intelligence, and it seems to me that having such interests from a young age correlates with a number of unusual and perhaps less than socially acceptable interests. Which would probably be influenced by a lifetime of trouble "fitting in", something I suspect is a disproportionately familiar experience to many of us on HN.
That is what you think.
"You're a laid back, liberal female who doesn't eat out frequently and doesn't prefer style when buying clothes and is less satisfied in life than most".
It also says:
Not likely to like outdoor activities Not likely to consider starting a business in next few years
I am a male who has been self employed most of my life, and a moderator on one of the busiest "Backpacking Forums" here in the US. So it's got that wrong, but it's true I have very little interest in "clothes style" and don't eat out much.
The "female" part is especially interesting because I'm very much a straight male, but I was a single parent with a 1 year daughter 30 years ago and I did make a very conscious effort to be a "mom" to her. That required I learn some very feminine traits like "kissing boo boos" and nurturing with hugs and kisses and lullabies. I learned how to do these things by closely observing real moms, and asking them for help, or getting my ass chewed by them when I was "doing it wrong".
At the time friends and family were fairly astounded the first few times they saw me go instantly into "Mom" mode when my daughter required my attention.
And, I haven't posted anything, or made any comments about either food or clothes. I can only assume it's my lack of doing that which prompts the conclusions regarding it but I was honestly unaware that others must do that enough to make my not doing it somewhat remarkable.
So, in that very short time period (9 hours on FB) the plugin has made some pretty insightful conclusions and missed the mark on some as well. For example, I've been a hiker almost since I could walk and almost all of my hiking over the past 25 years has been off trail "bushwhacking" into Wilderness areas and most of that has been solo. But, I rarely ever mention my hikes or my work on FB, so again, it would seem to be the lack of those that prompts the algorithm's conclusions.
(https://addons.mozilla.org/en-US/firefox/addon/dataselfie/)