What Trump and Vote-leave (Brexit) have in common - Manipulation of the masses through big data

In the german media a topic is heating up pretty much right now. I was looking for counterparts in the english media and was quite surprised, there are only similar articles but none that connect as many things as the original swiss article (https://www.dasmagazin.ch/2016/12/03/ich-habe-nur-gezeigt-dass-es-die-bombe-gibt). Because of this, I'd like to present these thoughts here as well as share my view, which is contrasting in some aspects.
First of all the claim of this article is that there is a company, namely Cambridge Analytica (https://cambridgeanalytica.org), that offers politicians support for their online campaigns. Not only have they supported Trump and Brexit, but they have enabled their success respectively. How does this work? Well, they use a psychological model called "OCEAN", where the letters stand for "openness, conscientiousness, extraversion, agreeableness and neuroticism", the model is also often referred to as "Big 5 personality traits" (https://en.wikipedia.org/wiki/Big_Five_personality_traits). Many of you will know this, since there are a lot of online tests, especially on facebook, that ask you several questions and afterwards tell you, what kind of person you are. You might have asked yourself, why are there so many tests on facebook. The answer is that this company creates these tests, so that users take them and tell about themselves. Therefore they gather data and can predict the behavior of users. Of course not only Cambridge Analytica does so, there are other companies as well. The article then states that Cambridge Analytica does not only analyze these surveys, which are directly designed to measure someones personality but they can also analyze your likes on facebook and from these likes deduct your personality. In contrast your likes are not directly designed to measure your personality, they are just a way of interacting with people on facebook. However this is an interesting answer to the question "What does facebook do with all this data about users?".
The german articles focuses very much on the scientist who invented this model, Michal Konsinski (this part I could not find in english media). His invention, the OCEAN model is portrayed as an invention, which he made, but cannot control anymore. He has shown in 2012 only 68 facebook likes are necessary to predict the skin color of a user (95% accuracy), homosexuality (88%) and if the user is republican or democrat (85%). Kosinski allegedly saying that with more facebook likes it is possible to make a better prediction than the partner of a person and if you have enough likes gathered it is possible to make better predictions than a person can make about himself. The article further states, that when he published these results, he got 2 phone calls. One threat to sue him and a job offer. Both were from facebook. The scientist is contacted by a company wanting to measure the personalities of more than 10 million people. He is skeptical, it means a lot of money for his institute, but he googles the company, it is SCL – Strategic Communications Laboratories, the company that owns Cambridge Analytica. He realizes this is a company which helps politicians to win elections. He declines, but it seems that a colleague from his institute works for them.
After the Brexit referendum, he realizes that his invention is already being used in the real world. Therefore his nightmare has come true. The article goes on telling how Cambridge Analytica has helped Trump to be elected. First, psychological screening using OCEAN, second Big-Data-Analysis and third Ad-Targeting. The first two parts are quite obvious, the latter means sending content to people, considering their psychological profile. I.e. people with high neuroticity get pictures of guns advertised as insurance against criminals such as burglas, also depicted trying to break into a glass window. In contrast extrovert conservative guys get a picture of a guy holding a gun and a child at sunset. Furthermore the campaign is flanked with content aimed at hillary voters to keep them from voting. Potential black Clinton voters get fake videos where hillary denotes black males as predators. The article gives more examples, but they all outline the same concept.

So far I have only summarized the cited article. English articles on this subject can be found here:
http://www.nytimes.com/2016/11/20/opinion/the-secret-agenda-of-a-facebook-quiz.html
http://edition.cnn.com/2016/11/04/politics/donald-trump-political-ads-cambridge-analytica/

I found this topic very interesting, since it explains a lot of things. Also it gives another explanation why trump was elected. In this sense he hired the team with modern technology, whereas in the other explanation it is assumed that people care less and less about facts and we are engaging the "post-truth"-era. I find it relieving that this trend may end, once the more sane guys (the ones who believe in climate change etc.) use this modern technology as well. It will be interested for sure, whether everybody will end up lying about everything just depending on the target of the lies. Or this will lead to a fatigue mechanism, making lies less effective. I don't know what will happen, but time will tell.
In contrast to the original article, I totally disagree to the claim that "the big 5 / OCEAN is the bomb, which has been unleashed". I think this is just a model. There are a lot of other models and it doesn't matter which one you use, as long as it comes with enough dimensions (5 for OCEAN) and these dimensions are independant enough to avoid degeneracy of the mathematical model. I think the bomb being detonated in this story is big data. Up to now, we were like "huh well, facebook shows me ads of vacuum cleaners, when I have googled cheap vacuum cleaner 5 minutes ago, ok" and we thought like, big data is a buzzword, but there is not much behind it. This has obviously changed. But let me get back to the point, why OCEAN is not important. In fact there is a model of 5 independent characteristics of a personality, but in practical use, there are actions users have taken on facebook and these actions are mapped on this model. Afterwards this model is mapped on a predicted behavior. So this model serves as a way to reduce dimensionality. The dimension of all possible facebook actions is really high and thus hard to make predictions from. If there is an encoding of actions to a personality model and further a decoding of a personality model to predicted actions, this is a typical behavior of Autoencoders, a special type of neural networks. However the used technology is far behind what is possible today with neural networks. There are 3 notable milestones in AI (artificial intelligence), 1996 Kasparov was beaten in chess by deep blue, in 2011 Watson has beaten the world champions in Jeopardy and this year alpha-go has beaten the world champion in Go. The technology has changed a lot from one to the next milestone. In 1996 the AI was not using brute force, but it was not far from plain computation of all moves. There was some statistics optimizing the tree search being done. Beating jeopardy with this approach was out of range and new methods were developed that associate terms with probabilities, rank these probabilities and makes predictions, which associations provide the best match. Still this is just a clever version of entering some words and finding an article with most of these words. This is mainly the technology level at which Cambridge Analytica seems to operate. If one looks at alpha go, there is of course a model, which represents how good a state on the board is and how good possible moves are. But the important point is, that this model has not been written down by any human. The machine has developed this model by itself, there was nobody to say "well a 5 dimensional model of personality traits makes most sense". Model free learning means, that the neural network finds its own model and this is not done because humans don't have any models on most problems, they want to solve, but mostly because these networks find better models. A personality model with 5 traits is just something that humans can describe and understand in a reasonable time. However there might be more efficient models. So what I want to say with this point is that manipulation of the masses harnessing the power of big data has not yet reached its climax, it has just begun. This means, even if you doubt this technology has helped trump, its influence will grow and we should consider this instead of ignoring a potentially unpleasant fact.