The tide rolls in
January 25, 2017 6:50 PM   Subscribe

The O.C.E.A.N. model washes away democracies "How good a model is, however, depends on how well it can predict the way a test subject will answer certain further questions. Kosinski charged ahead. Soon, with a mere ten “likes” as input his model could appraise a person’s character better than an average coworker. With seventy, it could “know” a subject better than a friend; with 150 likes, better than their parents. With 300 likes, Kosinski’s machine could predict a subject’s behavior better than their partner."
posted by bitmage (66 comments total) 38 users marked this as a favorite
 
Hari Seldon's hologram is rolling over in its grave.
posted by Behemoth at 7:03 PM on January 25, 2017 [9 favorites]


Don't be so credulous.
posted by ch1x0r at 7:09 PM on January 25, 2017 [12 favorites]


I don't like enough things on Facebook for them to predict anything about me.

On the other hand, I tried their textual analysis on three things I wrote -- a legal case comment, a record review, and the scene from a romance novel -- and they are 99, 99 and 91 percent convinced that I am definitely a man.

So, you know, not thinking all that highly of this process so far.
posted by jacquilynne at 7:11 PM on January 25, 2017 [22 favorites]


Then they enter a subject’s reactions to certain messaging back into the app, from where this new data flows back to the control rooms of Cambridge Analytica.

Brilliant and sickening.
posted by polymodus at 7:32 PM on January 25, 2017 [2 favorites]


Kosinski and his team have in the meantime figured out how to sort people according to Ocean criteria based only on their profile pictures.

What was that about racist algorithms again?
posted by XMLicious at 8:00 PM on January 25, 2017 [13 favorites]


Their test thought I was a conservative man, so.
posted by Miko at 8:00 PM on January 25, 2017 [3 favorites]


Having witnessed how utterly inaccurate the alleged "match" percentages OKCupid generates (based on some magical analysis of hundreds or even thousands of detailed and pointed questions answered by users) generally tend to be, I shall pay this snake-oil sales pitch no mind whatsoever.
posted by Greg_Ace at 8:04 PM on January 25, 2017 [10 favorites]


because I'm a privacy idiot, I went ahead and did the test

apparently liking Bernie Sanders, Amnesty International, and The Wire means that I appear less involved in politics. liking Adventure Time, the Venture Bros, and Tokyo Godfathers make me appear more involved

liking Amnesty Intl' also makes me appear religious, less Christian, and more Catholic

liking the Wire also makes me appear to not to be in a relationship and to be married

liking the NAACP and Killer Mike apparently also makes me appear less intelligent

besides letting me know that people are racist (duh), I'm not sure what you're actually supposed to get out of this test
posted by runt at 8:05 PM on January 25, 2017 [20 favorites]


"[T]his work by investigative reporters Hannes Grassegger and Mikael Krogerus got a lot of attention—and generated some controversy, for apparently having scooped the English-language media with sensational observations about 2016’s most sensational story, the campaign and electoral victory of a fascist dictator in the United States. Perhaps for this reason, the article has not appeared in translation in (or even had its investigative threads taken up by) English-language media outlets, even after nearly two months."

The lack of substance may have something to do with it
posted by stinkfoot at 8:11 PM on January 25, 2017 [3 favorites]


Jesus Christ. This was The Matrix level reveal of scary dystopianism. Also I feel like I got an inkling of this personality reasons for political campaigns stuff from an Adam Curtis documentary so somehow it feels extra true and terrifying.
posted by latkes at 8:11 PM on January 25, 2017


Using my writing (a big chunk, from here), it thinks I'm a shy, slightly neurotic, left-leaning, introverted man (94% male). Using my facebook likes, it thinks I'm sliiiiiiightly masculine and conservative, based apparently on the fact that I follow the Cubs and the Bears on Facebook (largely because I'm too lazy to follow a sports news website so I keep up with my teams on FB). It also thinks I'm probably straight and single because I pay too much attention to sports. And of slightly below-average intelligence.

It's actually very amusing, when it sorts my interests and education it is almost exactly upsidedown, with things I actually studied or engage in as hobbies or interests listed as LEAST likely while the things I'm most tenuously attached to are listed as my HIGHEST interest things ... apparently exactly because I use facebook to casually follow things I'm casually interested in -- like sports, some TV shows, etc. -- but things I am really fascinated by or engaged with (law, astronomy, education, architecture) I tend to keep up with through RSS feeds or daily website visits and don't bother to follow on FB.

(It also lists my "like" of Notre Dame as a way it knows that I'm "a Christian - but probably not a Catholic." And this is why algorithms need a human checker!)

It apparently only uses "well-known" likes -- it used 22 of mine and from what I can tell they're all big-brand things like the Cubs, Buffy, the Brady Campaign, etc. I went on facebook and looked, and I have 179 likes, but almost all of them (all but about 22, apparently!) are small local businesses, friends' bands/professional pages, local meetups, and that sort of thing, which it doesn't use in its accounting of who I am.
posted by Eyebrows McGee at 8:12 PM on January 25, 2017 [1 favorite]


I personally think Cambridge Analytics is probably snake oil, but that the Trump campaign's investment in cheap Facebook ads -- supposedly at the suggestion of Kuchner --- instead of more traditional advertising is a significant innovation that may have won Trump the election.

In 2012, the Obama campaign bought a lot of cheaper ads on cable channels instead of buying expensive broadcast spots like traditional campaigns do. They combined it with some demographic targeting to optimize those ad buys. The Trump people don't seem to have done any of that, but I suspect that their Facebook buys represent a similar evolution of political advertising.
posted by chrchr at 8:24 PM on January 25, 2017 [18 favorites]


Hi, I'm a conservative woman who likes Noam Chomsky and Richard Simmons. lighthouses light up my life, and Neil Degrasse Tyson lights my way.
I might be gay, I'm probably Christian, I'm as smart as half the folks and not as smart as the other half.
Life's O.k., but it could go either way.
Somehow Richard Simmons and Lighthouses make me seem more involved in politics while Noam Chomsky and Neil Degrasse Tyson make me appear less so.
Jesus!
I've had carny cold-readers be more accurate, and that's without having my fucking Facebook page in front of them.
posted by Floydd at 8:30 PM on January 25, 2017 [10 favorites]


This interview with Kuchner describes how it went down. In short, they didn't have time to setup an Obama style analytics operation, opting for digital as an expedient option. Meanwhile, the Clinton campaign was content to repeat the the Obama 2012 advertising strategy.
posted by chrchr at 8:33 PM on January 25, 2017 [2 favorites]


Last time I had a Facebook account was 4 years ago, and all my listed interests were elements of the periodic table. Kinda wishing I'd used that account during the past election season to see what sorts of weirdo ads it might have dreamed up for me.
posted by potrzebie at 8:34 PM on January 25, 2017 [4 favorites]


I've got FB convinced I own a mine and am super into megayacht rentals. It is fun and eye opening.
posted by Stonestock Relentless at 8:43 PM on January 25, 2017 [15 favorites]


In This Thread: people drawing conclusions about statistical models based on one or two or five or ten data points.
posted by jsnlxndrlv at 9:25 PM on January 25, 2017 [15 favorites]


My FB likes made the test throw up it's hands and give up, even though I'm a pretty active user at this point. Maybe because Even though I like my friend's posts all the time, I make an effort to never like a corporate or advertising related page.

I feel like this may be pretty normal for Mefites, but not for the population at large, so this is still a bit scary.
posted by fnerg at 10:15 PM on January 25, 2017 [2 favorites]


Using zero data points is not a better idea. Sure, maybe everyone here is an outlier, but there's no compelling reason to suppose so, especially when the article itself contains virtually no specifics and leans heavily on a ominously vague Forer-effect-bait pronouncements from the model salesmen.
posted by RogerB at 10:15 PM on January 25, 2017 [2 favorites]


Some of you guys are totes giving it more data.


When are people going to stop treating politics like a battle and start educating each other and solving problems?
posted by amtho at 10:37 PM on January 25, 2017 [3 favorites]


When are people going to stop treating politics like a battle and start educating each other and solving problems?

When one of the two major political parties is no longer engaging in eliminationist rhetoric.
posted by PMdixon at 11:21 PM on January 25, 2017 [5 favorites]


Sorry - the "people" in my comment referred to those parties more than anyone else.
posted by amtho at 12:15 AM on January 26, 2017 [1 favorite]


The thing is, most people here are at least somewhat conscious of their internet use. I usually don't like commercial sites on Facebook - actually I rarely like any sites. The people they are targeting have a different approach - otherwise they wouldn't be targets.
Some of my friends will try every test, and they will go in for every offer or competition. They are also the most likely to believe in and share fake news (this is both rightwing and leftwing). What I found most depressing in the article was the part about negative stories targeting PoC - I can see that happening here as well, from what people repost, and it breeds disillusion, hopelessness.
The marches are a great counteraction to that - something real that can be felt against the rumors and innuendo.
posted by mumimor at 12:58 AM on January 26, 2017 [3 favorites]


Another analytic tool confirms MetaFilter is a special case.
posted by hat_eater at 1:48 AM on January 26, 2017


I tried it and it said I am no more likely to be gay than the general population, despite my liking several drag queens and a page entitled "Bottom Leftist Memes." It also only got 1-2 components of my OCEAN test "correct" (i.e., agreeing with my self-rating).

Obviously, even a weak signal over millions of people would probably result in better click-through and tangible effects in terms of money... but it's also not exactly building a Caprica-esque avatar of me from my Facebook likes.
posted by en forme de poire at 2:17 AM on January 26, 2017 [1 favorite]


The people they are targeting have a different approach - otherwise they wouldn't be targets.

I remember reading how obviously fake, misspelled 419 e-mails are more efficient, because people who bite are essentially prescreened for gullibility.
posted by Dr Dracator at 2:48 AM on January 26, 2017 [5 favorites]


Does it work on MetaFilter favorites?
posted by chavenet at 2:55 AM on January 26, 2017 [5 favorites]


the article has not appeared in translation in (or even had its investigative threads taken up by) English-language media outlets

Here's NYT on Trump, Bannon and Cambridge Analytica, from Nov 2016. There's plenty more in US media. Can someone who made it through the Das Magazin article (I failed this time, may try again later) summarize what's new in that one?
posted by effbot at 3:10 AM on January 26, 2017 [3 favorites]


The NYT article and the German article have nearly all the same information, down to very similar examples and facts, which is a bit odd, did the German reporters read the November article and derive their work off of it?

The side differences are the NYT article saying Facebook/Zuckerberg are best bet to combat this because the new administration probably won't. The German article weaves in a typical academic story, about Kosinski and how his research ideas got implemented and exploited in this way by a rival, and the "good" faculty are really upset about this.

What I found most depressing in the article was the part about negative stories targeting PoC

Those are very stark examples, in particular illustrating how targeted demographics are fed propaganda, and that this actually happened. The point is that the accuracy of the OCEAN profiles doesn't matter; all that's needed is a small difference to differentiate a number of groups and then human agents/operators come in and produce the actual propaganda. The actual metric values are irrelevant in this regard.

In science and engineering experimental methodology, this is the idea that you don't need fidelity to the imagined model constructs for metrics to be useful, a simple example of that is in some contexts you don't need the actual value you just need to know if it was bigger (e.g. relative values), and that's enough to let you decide.

And yes it's sad and concerning that scientific knowledge is being pretty much perverted in this way.
posted by polymodus at 3:38 AM on January 26, 2017 [8 favorites]


It thinks I'm a man, baby!

(I'm not.)
posted by soren_lorensen at 6:10 AM on January 26, 2017


It thinks I am a human.

(I am a dog - now you know.)
posted by otherchaz at 6:27 AM on January 26, 2017 [2 favorites]


Another analytic tool confirms MetaFilter is a special case.

I've worked with Kosinski's data -- a LOT -- in an academic setting, and I really can't stress enough how true this is. First, people like all of you are less likely to have handed over data, less likely to have selected (accurate) age and gender in your Facebook profiles, less likely to have enough data to be a reliable training (or testing) observation, and less likely to have filled out the NEO-FFI (five-factor inventory, OCEAN-- openness, conscientiousness, extraversion, agreeableness, neuroticism).

From what I can see, the trouble is with the confidence bounds. Kosinski isn't accounting for qualitatively different data from what he's already collected.
posted by supercres at 6:55 AM on January 26, 2017 [21 favorites]


I've been teaching data design and social media to art students for the last few years, and I've been known to use the Magic Sauce app in class as a way to demonstrate how badly these kinds of analyses actually work in practice. Their claims about what they can and can't tell about you don't hold up. I've heard numerous pieces of anecdata about how powerful this is, but I don't think I've seen a single instance of an accurate reading.

I've seen numerous versions of this story kick around, and while I have no doubt that Cambridge Analytics are up to no good, my guess is that this is a flagship demo they use to impress clients and gaslight the public rather than their operational methodology. They don't actually need to do any of this stuff when they can reliably weaponise the homophily of social networks using old-fashioned propaganda and grubby direct marketing tools.

I found a lot to think about in Jonathan Albright's Fake News series on Medium. He touches on CA's claims about their psychometric capabilities, but far more interesting is his analysis of the alt-right's propaganda machine in action. This seems far more consistent with Steve Bannon's known modus operandi and doesn't rely on fancy tech that is demonstrably unreliable.
posted by Elizabeth the Thirteenth at 7:36 AM on January 26, 2017 [10 favorites]


...which could also lead to an impression, by those who get access to reports from this data, that people like this group here don't exist. They could get the impression that "people" don't care about privacy, that they don't think both critically and compassionately, and that they engage carelessly with The Internet.

I'm not talking about educated data scientists' impressions, I'm talking about the people they work for and cognitive biases toward only "seeing" the information they're looking at.

Data like this is fun for some and compelling for others and it's hard to persuade people not to use it, so it's going to have outsized prominence for those who have access to it.
posted by amtho at 7:40 AM on January 26, 2017 [1 favorite]


It knew I was definitely not a wolf.
posted by Greg_Ace at 8:33 AM on January 26, 2017 [3 favorites]


Just about every single one of my like is a TV show apparently.

South Park makes me appear more intelligent but Always Sunny in Philadelphia makes me appear less intelligent? I think that's probably backwards.

Dr Pepper is less libertarian and more Christian.
South Park and The Daily Show is more religious?
30 Rock and The Office appear to be single. Aww. Poor NBC.

Also, "0 out of 100 women with your Likes are Lesbian". Sure would suck if I became a lesbian since I'd be single forever, I guess.

Interesting that the Education section leaves out Math, which I guess could be rolled into either engineering or finance. There are only 3 stem categories for me though: Engineering, IT, and Biology. The survey says I prefer working on my own and probably avoid people-centered careers, while also telling me I'd be a good psychologist.

Really this is just telling me that I probably watch too much TV.
posted by LizBoBiz at 8:51 AM on January 26, 2017


For folks incredulous about this, how do you square your doubts with Facebook's valuation? These ads and personal data sales are the core of their profits.

Then you've got Twitter tripping all over itself to get the same revenue streams.
posted by Strange_Robinson at 8:51 AM on January 26, 2017 [3 favorites]


Some more information from a person who helped on research for the article in a reddit post:

I contributed a lot of the research behind the original article. If interested, you can also read about "low information" voters or the disinformation mercenaries who control those databases. My (huge) background dossier is here.
posted by zabuni at 9:31 AM on January 26, 2017 [3 favorites]


I'm surprised by how many reactions in this thread are anecdotal. I suppose that's the domain in which we can best interact with this, but it just seems short-sighted and reactive in nature. I'm actually very scared about this, and its running unchecked.
posted by a good beginning at 9:49 AM on January 26, 2017 [3 favorites]


To everyone submitting them to the system and snarking away at its seeming inaccuracy... if I controlled a unique data-based marketing company with some major wins under my belt, I sure wouldn't be putting my best tech on display in a free public test version. If my clientele were mainly from one part of the political spectrum, and my system detects that the public tester skews opposite... I would even deliberately screw up a little, so as to put them off the scent a bit.

Just saying.

(disclaimer: never joined FB or Twitter)
posted by Artful Codger at 10:11 AM on January 26, 2017 [3 favorites]


I'm surprised by how many reactions in this thread are anecdotal

The article (and pull quote) make hyperbolic claims about the algorithm "predicting your behavior" better than the people closest to you. I agree that in reality the power of something like this doesn't have that much to do with whether it figures out individual MeFi commenters, but it has sort of been framed here as something that could.
posted by atoxyl at 10:19 AM on January 26, 2017 [1 favorite]


One thing to keep in mind is that people are fairly predictable in general and that it doesn't take much to predict especially political stuff. Just now for shits and giggles I predicted more then 90\% of 2012 votes (in the CCES) correctly with just sex, race, and party ID.

I mean, most of this is not rocket science and to some extent bringing this OCEAN stuff in seems like swatting a fly with a sledgehammer. Figuring out that someone is Haitian or connected to Haiti from their facebook stuff does not seem like rocket science, and showing them something negative about Clinton-and-Haiti is likewise not rocket science. The only hard part, honestly, is that creating a bunch of decent ads costs time and money. Yeah, it sort of worked this time, maybe, insofar as Trump won, but we really don't know whether that was more because "They had an amazing and clever technique for manipulating people through facebook!" or more just "It turns out that more white people then you thought will eat up explicit racism with a spoon if you actually give it to them."
posted by ROU_Xenophobe at 10:50 AM on January 26, 2017 [4 favorites]


hey guys you'll never guess which Friend I am
posted by ricochet biscuit at 12:06 PM on January 26, 2017 [4 favorites]


Adding to what I posted above: I'm pretty predictable, when it comes to voting. Anyone looking at my sparse data will conclude that I am a person who will not be seduced by internet ads or by trolls. I'll vote how I usually vote because everything I ever post underlines it. So I am a waste of time. As is everyone being snarky above. Good for you, and for me. I like to know thoughtful and reliable people.
The vulnerable types are like my old construction professor (who I have mentioned before here on the blue), who is a very smart and good guy, but completely ignorant of everything internet, and a YUGE fan of social media. He puts everything about himself out on FB and twitter, and also responds to every quiz and every ad, and he gets a s***load of fake news and propaganda in return. I've tried everything to get him to back down, because he is really not an idiot, but he is old, and the whole thing about internet safety seems obscure to him. If he were American, he would be a Bernie-bro first, and then he would have voted for Trump at the general, he said as much. He is consistently targeted with "news" and "opinions" that matched his pet issues, and only sees alternatives if I and a small handful of friends spam him.
posted by mumimor at 12:49 PM on January 26, 2017


I'm surprised by how many reactions in this thread are anecdotal.

What do you propose we do instead, exactly?
posted by RogerB at 9:02 PM on January 26, 2017


Quit using social media. No kidding. That's the solution.
posted by Strange_Robinson at 10:43 PM on January 26, 2017 [4 favorites]


There's alternately regulating second and third order collection.

Nationalizing Facebook could work too. Once they are cowed, the rest will fall in line.

There are many solutions, but please notice how all the solutions are beyond the pale. This seems deliberate to me though I'm sure I'm just being paranoid.

stallman save us!
posted by Strange_Robinson at 10:48 PM on January 26, 2017 [1 favorite]


Oh, and smash your smart phone. We're ghetoizing ourselves by self selecting into tiny little fetish groups.
posted by Strange_Robinson at 10:50 PM on January 26, 2017


Wow, it really reads a lot into the fact that I like The Godfather. It showed up as being a factor in almost every single one of the determinations that it made about me. (But...the Godfather is widely considered one of the greatest and most popular movies ever made. Everyone likes The Godfather! The knowledge that I like it should be considered very low-entropy information.)

while I have no doubt that Cambridge Analytics are up to no good, my guess is that this is a flagship demo they use to impress clients and gaslight the public rather than their operational methodology.

How on earth does it impress clients, though? If it is never accurate, that's not going to impress anybody.
posted by phoenixy at 1:16 AM on January 27, 2017 [1 favorite]


How on earth does it impress clients, though? If it is never accurate, that's not going to impress anybody.

Not so. It's an amazing sales tool, glamorous in every sense of the word. But reading between the lines you can glimpse the man behind the curtain. From the Bloomberg article (emphasis mine):
Of all the microtargeting profiles of myself I had seen, none had flattered my self-concept like this one. Its predictions already seemed more plausible than those of the Democratic data warehouse that had my religion pegged as Lutheran—a prediction likely tethered to the only slightly less dubious profiling of my ethnicity as German.
But then:
A few weeks after I visited their London office, I went to the University of Cambridge Psychometrics Centre’s website to see what I could learn about myself from a psychological assessment detached from political considerations. [...] It took nearly half an hour to complete the 100-item questionnaire before receiving a diagnosis. The modeled assessment I had been shown by the Cambridge Analytica team had been accurate in placing me in the middle range for extraversion and low on agreeableness. But on other personality traits, the company’s prediction had been wildly off.
I think they believe it themselves as well, at least partly. And in a way, it doesn't matter that this is not particularly accurate because their other strategies work so incredibly well. But those are shabby and underhand and lack the mystique of cutting edge tech that borders on magic. It's not such an easy sell.

Honestly, this doesn't make them any less scary. But I think it helps to recognise that they're not evil geniuses, they're basically spammers.
posted by Elizabeth the Thirteenth at 1:53 AM on January 27, 2017 [1 favorite]


The modeled assessment I had been shown by the Cambridge Analytica team had been accurate in placing me in the middle range for extraversion and low on agreeableness. But on other personality traits, the company’s prediction had been wildly off.

So cold reading, basically.
posted by effbot at 4:04 AM on January 27, 2017 [1 favorite]


The text analysis test pegged me as a 47 year old man. I wish I got treated like a 47 year old man in real life. Higher salary and automatic authority, here I come!
posted by aka burlap at 7:01 AM on January 27, 2017


Seriously, though, this really interesting and unsettling. They may not have the algorithms perfectly tuned yet, but they are getting better at it, and using big data to manipulate people is only going to become more and more effective. We should take this seriously.

It feels like an arms race sometimes, the way advertisers and political strategists are all aiming to intrude further and further into our sincere human experience, and the way we have to increasingly alienate ourselves from that experience in an attempt to protect ourselves.
posted by aka burlap at 7:05 AM on January 27, 2017 [2 favorites]


… so, the secret to winning national elections is to just be more cravenly full of shit? Thanks, Facebook!
posted by klangklangston at 11:08 AM on January 27, 2017 [1 favorite]


For folks incredulous about this, how do you square your doubts with Facebook's valuation? These ads and personal data sales are the core of their profits.

To be totally honest I think Facebook, and that strategy, is really overvalued. I strongly suspect that the promise of using machine learning to do this kind of nano-demographic ad buys is hyped beyond what it can actually fulfill right now, an impression Elizabeth the Thirteenth's link does nothing to disabuse me of. I think ROU_Xenophobe has it right here: most of the variance is down to easy-to-measure demographic variables, and none of this would have made a difference if white Americans weren't so vulnerable to the Trump campaign's racism.
posted by en forme de poire at 1:02 PM on January 27, 2017 [2 favorites]


tldr, overfitting is everywhere and fools everyone.
posted by en forme de poire at 1:03 PM on January 27, 2017 [3 favorites]


(Also, assuming Trump's big data strategies were actually unreasonably effective, based on his narrow wins in a few Rust Belt states, is also overfitting, of a more casual variety.)
posted by en forme de poire at 1:04 PM on January 27, 2017 [2 favorites]


(Also sorry to quadruple-post but Cambridge Analytica is not the same as the University of Cambridge research site that people are interacting with, right?)
posted by en forme de poire at 1:06 PM on January 27, 2017


I think ROU_Xenophobe has it right here: most of the variance is down to easy-to-measure demographic variables, and none of this would have made a difference if white Americans weren't so vulnerable to the Trump campaign's racism.

Yeah - ROU_Xenophobe called out age, race, and sex, but back of the envelope says race alone allows me to correctly predict ~70% of the US population's voting pattern (super crude: ~60% are non Hispanic white. ~60% of those vote Republican. ~40% are not non-Hispanic white. ~10% of those vote Republican. 36%x2 - note that this model (White = R, else D) is much more likely to falsely predict an R vote than a D one)
posted by PMdixon at 2:33 PM on January 27, 2017 [1 favorite]


I wonder how much time the skeptical posters here spend on Facebook. What else would Facebook's data be useful for, if not identifying angry and uneducated people?

Have any of you ever even been on Facebook? I know we're all snowflakes here but the average Facebook user is not.

Maybe Cambridge Analytica is entirely snake oil and Facebook's valuation is total bullshit, but it is baffling to me that anyone could say this data could not be used in this way. Maybe it didn't help Trump out this time, his campaign was a dumpster fire that has now spread, but someone will pull this off sooner rather than later.
posted by bradbane at 6:51 PM on January 27, 2017 [2 favorites]


I wonder how much time the skeptical posters here spend on Facebook.

According to my phone's battery usage statistics, about 372 hours a day.
posted by jacquilynne at 7:30 PM on January 27, 2017 [1 favorite]


Maybe Cambridge Analytica is entirely snake oil and Facebook's valuation is total bullshit, but it is baffling to me that anyone could say this data could not be used in this way.

We should definitely be afraid of Cambridge Analytica. They are propagandists of the worst order, they're exploiting Facebook at every opportunity and it's abundantly clear that they're very successful at it.

What I'm saying is that I don't believe this particular hack works, or has much chance of working, ever. When you work with data in large volumes, as I have, you quickly come to realise how difficult it is to use it to draw any meaningful conclusions that support anything other than the most blindingly obvious hypothesis. Something as heterogeneous as Facebook likes is not going to deliver anything approaching an accurate psychological profile.

Big Data, n.: the belief that any sufficiently large pile of shit contains a pony with probability approaching 1

There is indeed a pony, but this isn't it. This is a steaming pile of horse manure. The pony is elsewhere, and it's much, much scarier.

Maybe it didn't help Trump out this time, his campaign was a dumpster fire that has now spread, but someone will pull this off sooner rather than later.

If you want to understand the Trump campaign's success on the Internet, look at the way ISIS operate. They identify lonely and disaffected users and bombard them with messages to bring them onside. They use Twitter and YouTube to create echo chambers of propaganda. They don't need Big Data to do this, just an army of motivated Internet foot soldiers.

Now scale that up to industrial level. Begin by harvesting email addresses from one of your fake news sites and get spamming. If you can throw in a bit of malware to hang out on your target's outdated and vulnerable device, you can reach their friends too. It's not rocket science, it's malicious hacking and gaslighting. Whatever they're up to, it's almost certainly illegal.

These people aren't scientists, they're crooks and con men. This story gives them a veneer of scientific respectability, a fig leaf to hide what they're really up to. Don't let them have it.
posted by Elizabeth the Thirteenth at 2:50 AM on January 28, 2017 [7 favorites]


Vice.com published a version of the article in the original post: The Data That Turned the World Upside Down
WRITTEN BY HANNES GRASSEGGER AND MIKAEL KROGERUS
January 28, 2017 // 09:15 AM EST

Psychologist Michal Kosinski developed a method to analyze people in minute detail based on their Facebook activity. Did a similar tool help propel Donald Trump to victory? Two reporters from Zurich-based Das Magazin (where an earlier version of this story appeared in December in German) went data-gathering.
This is really a really scary and powerful application of psychometrics.
posted by ZeusHumms at 1:09 PM on January 28, 2017


E13 has this story dead on IMHO.
posted by en forme de poire at 4:30 PM on January 28, 2017 [1 favorite]


This is popular because it absolves us of the need to reflect on what has happened, and to empathise with people who have rejected our beliefs. We can simply declare that they were manipulated by technomagic, that they are just like us but have been fooled.

Therefore I expect to see it shared widely.
posted by pfh at 4:44 PM on January 28, 2017 [3 favorites]


"How did Nazis happen?"

"A wizard did it."
posted by PMdixon at 10:29 AM on January 29, 2017 [4 favorites]


« Older My God, it's full of RPG screenshots   |   If you're looking for a shark, you're not going to... Newer »


This thread has been archived and is closed to new comments