An article was published on the website of one of the most authoritative journals on scientific research, Nature.
What is known
The study was carried out by a psychologist from Stanford University.University Michael Kosinski (Michal Kosinski). To test the algorithm, he used photos of more than a million people from Facebook and an unknown dating site. These were mainly people from the USA, Canada and the UK. After analyzing the photos, the algorithm came up with about 2,000 features that were used to identify a liberal and a conservative.
Thus the political orientation wascorrectly classified by the algorithm in 72% of face pairs. This rate is much better than chance (50%), human accuracy (55%) or a 100-item survey (66%). The accuracy of the algorithm was consistent across countries, environments, and when comparing individuals across samples. In addition, the algorithm coped well with the task, even taking into account age, gender and ethnicity (69%).
Naturally, there is no talk about implementation yetalgorithm in real life, but in theory it could make it much easier to identify people, as well as track their location and social interactions. Moreover, unlike many other biometric systems, such facial recognition can be used without the consent or knowledge of the subjects.
By the way, Kosinsky also used a similar algorithm to determine sexual orientation. Its performance accuracy was 76%.
Source: Nature
</ p>