Social media is awesome, not just for private individuals and marketer, but also for authoritarian governments.

Authoritarians are usually toppled by surprise, because they were unaware of or underestimated a trend in the population.

Social media networks like Facebook solve the agency problem for dictators. They offer a massive amount of data about individuals. Combined with machine learning, they provide the perfect tools to control the population.

Let's start with the seemingly innocuous targeting of Facebook. I recently checked in at a public swimming pool. Now I am seeing ads for swimwear. That's okay I guess. All that happened is that the Facebook algorithm took my check in and now has me listed as interested in swimming. I don't mind the occasional Speedo ad even though I am not going to buy new ones anytime soon.

However, the same targeting option can be used by a local politician telling me in an ad or news item, or even a private post, that the other political party is trying to close down the public pool. Of course I am against that. Psychologically, I am even more receptive to that political message because I recently started swimming at that very pool.

For a client, my agency recently targeted people who like Porsche cars. Those come in two main groups – people who actually own Porsches, and people who aspire to one day buy one, but likely never will. When we limited the age to 40 and above, and then created a lookalike audience and invited them to like a specific page, we had a subset of people likely to own a Porsche and their friends.

Porsche owners are homophobes

We then used some manual labor and a machine learning tool to analyze what media outlets these people followed, which airlines they liked, and which hotels and restaurants they had checked into. Amazingly, we were able to identify one hotel, one airline and one newspaper almost all Porsche owners had in common. This allowed us to organize events, place advertisements and create promotions that were uniquely targeted and thus incredibly effective.

Just for fun I then went back to see what else these people had in common. Turns out they reacted negatively to news about gay marriage legalization by a significant margin. And there you have a political insight which again can be used to manipulate with scalpel precision a specific segment of the population.

By analyzing people's messages on Facebook, machine learning can identify anything from need for money (loan ad or party promise to subsidize), depression (article on medication or political fear mongering ad), feeling ugly and unwanted (targeted ad for makeup, feminist agenda) and so on. In Australia a scandal broke recently where advertisers told clients they could now target depressed teenagers.

In a nutshell, the same tools of social media machine learning can be used to sell you soap or influence your political opinion. The Trump campaign allegedly targeted black men in Philadelphia with articles that voting wasn't worth it because "Hillary was just as bad." Such messages work much better than direct political advertising.

Dictators love machine learning

Now imagine what real dictators can do with those tools. Russia's kompromat techniques can be brought to new heights: using machine learning to identify people with specific views, targeting them with nuanced messages, shutting down the dissident leaders etc.

Indonesia is cracking down on homosexuality these days: identifying sexual orientation from social media like Facebook or Tumblr is way easier than randomly raiding gay saunas.

In countries like China, the technology created in the name of free speech and communication is already being used to grade and evaluate human beings. Machine learning is being used to analyze the level of criticism against the government in a specific WeChat post. Moderate articles are allowed, real criticism is blocked, giving the illusion of both benign government and free speech.

One man's ad targeting is another man's suppression tool. As a society, we will have to come to grips with the implications of a society brought together by social media platforms and divided by social media consumption. It will not be easy, and it may get a lot worse before we have solutions.