[ad_1]
Vice President Facebook, Guy Rosen, came to give answers and explanations on the live broadcast of the mbadacre in New Zealand. In a recent publication on the official blog, one of the company's leading referrals pointed out that automatic detection systems were not enabled because artificial intelligence did not have enough training recognize this kind of video.
The most read of the day | Discover what we talked about today to stay out of the world
Monday to Friday afternoon.
At the time of the macabre badbadination, which lasted 17 minutes continuously, about 200 people were watching it live. Then the video reached 4,000 reproductions before being removed from the social network.
"They asked us why our video and image matching technology, which was so effective in preventing the spread of propaganda from terrorist organizations, was not capturing these extra copies," wrote Rosen. "Our approach has been challenged by the proliferation of many variations of the videomotivated by the diversity and diversity of modes of sharing, "he replied.
Brenton Harrison Tarrant, 28, managed to broadcast for 17 minutes the horrific attack that killed 50 people.
Artificial intelligence systems behind Mark Zuckerberg's platform are responsible for democratizing the content that they be published there. They are based on "training data" that is used to educate software to recognize and eliminate content, such as nudity and terrorism.
Rosen also revealed that how users reported the video caused a delay in the reaction of the Facebook algorithm.
Currently, Facebook users can mark as inappropriate content anyone who understands nudity, violence, harbadment, false news, spam, terrorism, hate speech, rude content and suicide or self harm.
But the first reports on video shooting in New Zealand, after the end of the live broadcast, They have been described by users as "something other than suicide" and, as such, Rosen noted that "she was treated according to different procedures".
To top it off, according to Facebook, several users have visited the platform. various edited versions of the live broadcast, which further complicated the ability to block the video. Iterations, which included recorded fragments of televisions and computer screens, they worked to trick the artificial intelligence system and they allowed those copies to spreadwrote Rosen.
After the Tarrant episode, Mark Zuckerberg's social network revealed that he would start using an audio technology to detect this type of videos, it could even have been edited to violate the visual system of artificial intelligence.
"Another challenge will be automatically distinguish this content from visually similar and harmless content; For example, if our systems score thousands of video games videos broadcast live, our reviewers may miss important real-world videos, "said Rosen.
Tuesday, Facebook and YouTube defended their responses to the live broadcast of Christchurch. Google's video platform said that it had been overtaken by the "unprecedented scale and speed" with which new videos have been uploaded within 24 hours after the attack.
H.M
.
[ad_2]
Source link