Dictators constantly face a dilemma: crushing dissent to terrify (but anger) the populace or tolerating protests and offering reforms to keep the public at bay (but embolden dissidents in the process). Instead of relying on gut instinct, experience, or historical precedent, autocrats now have advances in data analytics and ubiquitous passive data to thank for letting them develop new, scientifically validated methods of repression. By analyzing the dynamics of resistance with a depth previously impossible, autocrats can preemptively crush dissent more reliably and carefully.
With machine learning and social network analysis, dictators can identify future troublemakers far more efficiently than through human intuition alone. Predictive technologies have outperformed their human counterparts: a project from Telenor Research and MIT Media Lab used machine-learning techniques to develop an algorithm for targeted marketing, pitting their algorithm against a team of topflight marketers from a large Asian telecom firm. The algorithm used a combination of their targets’ social networks and phone metadata, while the human team relied on its tried-and-true methods. Not only was the algorithm almost 13 times more successful at selecting initial purchasers of the cell phone plans, their purchasers were 98 percent more likely to keep their plans after the first month (as opposed to the marketers’ 37 percent).
Comparable algorithms to target people differently have shown promise somewhat more ominously. For example, advanced social network algorithms developed by the U.S. Navy are already being applied to identify key street gang members in Chicago and municipalities in Massachusetts. Algorithms like these detect, map, and analyze the social networks of people of interest (either the alleged perpetrators or victims of crimes). In Chicago, they have been used to identify those most likely to be involved in violence, allowing police to then reach out to their family and friends in order to socially leverage them against violence. The data for the models can come from a variety of sources, including social media, phone records, arrest records, and anything else to which the police have access. Some