A new study from Stanford University uses facial recognition to tell if you are a liberal or a conservative | Instant News


Today, the idea is bad: an AI system claims to be able to use…your face to accurately predict your political affiliation. Yes really.

Stanford University researcher Michal Kosinski published his research (you can read Here) In the diary nature Earlier this week.You may remember the name of Kosinski who took the lead Similar research As early as 2017, it claimed to be able to accurately detect a person’s sexual behavior based on the alignment of their facial features.At that time, civil rights groups and scholars General dunk Based on these findings, Kosinski is called “Ruck” published.

According to a Registration form Kosinski submitted an application for this new research to the Open Science Center, and he started this latest research in September 2017, shortly after the first round of anger began to fade. According to the research paper, this time, Kosinski took a sample of more than one million faces from an unnamed “popular dating site” (mainly used in the United States, the United Kingdom, and Canada).In addition to this sample, Kosinski’s original document also describes how some of these faces were extracted from Facebook-centric U.S.-centric Facebook profiles My personality, This is a psychological testing application that Facebook eventually disconnected from In 2018, Cited some Very bad Data practice.

In all cases (whether it’s somewhere on the dating site or somewhere on their Facebook profile), these users said they were “conservative” or “free.”

These facial batches are fed Open source The facial recognition algorithm then reduces each face to approximately 2,000 different data points (“including its core features”). After providing enough photo data points to the algorithm, Kosinski finally got a set of “average” points that can be associated with free or conservative faces, and the algorithm can successfully predict a person from their faces The political background averages about 72% of the time taking pictures.

At least according to this article, it is difficult to distinguish a red face from a blue face. Kosinski tried to isolate some facial features, such as the glasses they wore in a given photo, or the way their face tilted towards the camera, and tested whether these features can also be used as predictors of the person’s political orientation . It can be said that these functions are not as targeted as the black box of the algorithm.

In other words, Kosinski described some common points in his work notes Discuss this article. For example, people who consider themselves liberals are more likely to face the camera directly and are more likely to express “surprise” in photos. At the same time, in addition to having more whites, men and older people in general, conservatives eventually showed more “disgust” than liberals.

But nothing in the author’s comments can really cover why This research was even completed first. As VentureBeat Point out Within the scope of his own research, Kosinski’s entire research paper is based on the pseudo-scientific idea of ​​so-called “appearance”, that is, the ability to maintain a person’s entire personality through the arrangement of facial features.Psychology researchers have been saying for many years An algorithm that claims to be able to classify someone based on their face as a bank robber, a political scientist or a grumpy Republican is actually no better than accidental.

Kosinski’s notes attempt to refute these findings, saying that “many studies show that people can determine the political views, personality, sexual orientation, honesty, and many other characteristics of others from their faces” without… really pointing to any facts that support the research. .Maybe because of All evidence It seems to point in the opposite direction.



image source

to request modification Contact us at Here or [email protected]