Stanford University (USA) specialists have created facial recognition technology capable of determining a person’s political views from his or her photo. A description of this interesting algorithm has been published in Nature.
The lead author of the development – psychologist Michael Kosinski – used more than one million photos of residents of Canada, the United States and Britain in the process of training the algorithm. Besides photographs, the scientist used data on political views, age and gender of people. All of the data for the project, the researcher obtained on the social network Facebook, as well as on a dating site, whose name is not given. The result is an algorithm capable of determining whether a person holds conservative or liberal views.
The algorithm, as Kosinski specifies, identifies about two thousand factors that determine political preferences. The accuracy of the algorithm is over 70 percent. The author of the development noted that his algorithm should work not only with the residents of Canada, the United States and Britain, but it remains to prove experimentally.
Previously, Michael Kosinski created a similar system, but to determine sexual orientation by photo. As a result, the scientist was criticized – Kosinski’s colleagues decided that his development threatened the privacy of citizens.