Artificial intelligence can identify a gay person using a single photo

<-----sponsored ADS----->


Polish researcher Michal Kosinski says he invented the "gaydar", the gay person detector. It has used artificial intelligence algorithms "trained" with images from dating sites to detect similar features between gay and heterosexual people, both male and female. Based on only tens of thousands of photos, the technology is still not 100% perfect, but most of the results are correct, says Kosinski.

In an interview with The Guardian, Kosinski revealed details of this technology, which he has not yet released publicly. It seems that many of the "psychometric" researchers do not believe the results of AI gaydar technology are relevant as the algorithm has been taught to use photographs that might be similar in terms of clothing or accessories such as glasses and from the point of view of the angle from which the photograph was taken.

Kosinski says that initially, after using a single photograph for each individual his algorithm could have accurate results to detect whether someone is gay or heterosexual in 81% of cases for men and 74% for women . Human detectors, ie people who can easily identify gay people, have a 61% success rate for men and 54% for women. As soon as the number of photos went up to 5 for each individual, the success rate increased to 91% for men and 83% for women .

There are voices that say this technology is dangerous. In the hands of totalitarian or intolerant homosexuals, if they really work, such algorithms could be used to capture and punish suspects. However, Michal Kosinski argues that if he did not disclose information about this technology, it could have been developed separately by the states concerned and used without the public knowing

Kosinski is one of those who has contributed to the development of the technology behind the personality test that garnered Facebook user information and later came to Cambridge Analytica. Although he was not directly involved in this scandal, he was courted by Cambridge Analytica at the time. The researcher claims to have developed an algorithm that can detect even the political affiliation of people using a single picture, the only limitation at this time being bearded people who are harder to "read."

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *