The most popular video hosting daily checksmillions of videos on the presence of banned or copyrighted materials, passing along the most interesting videos in recommended ones. Artificial intelligence, which is used in almost all sections of Youtube, helps to do this quickly and efficiently.
Search forbidden videos
In the first quarter of 2019 in automaticYouTube mode has deleted 8.3 million videos. Algorithms do not work perfectly, sometimes completely “clean” materials are deleted, but in most cases the AI system justifies itself. Scenes of violence, a demonstration of dangerous actions, drugs - all this is successfully detected even at the loading stage.
Before applying machine learning, malwarevideos were deleted using user complaints. According to statistics, only 8% of harmful material was seized before the 10th viewing. After the introduction of AI in the protection of Youtube, 50% of the prohibited content does not even reach 10 views.
Users are unhappy ifoffensive videos still gain hundreds of thousands of views, falling into the recommended ones. Brand companies often order ads on Youtube and don’t want them to appear in banned videos that offend the advertiser’s reputation.
External pressure factors force developersYoutube is working hard to improve artificial intelligence to search for prohibited content. For this purpose, machine learning is used, which helps to organize the work of the "bad video classifier" on the home page and in the "Next" section. Big data is used to analyze the comments of viewers, which help to identify inappropriate content or “clickbait” - a video with a loud title, the content of which does not match the name.
The next section contains videos thatsimilar in content to the video you are watching. The included “Auto Play” function displays the first movie in the list, located in the “Next” section. This list is formed in two stages. The first analyzes the user's browsing history. At the second stage, an independent ranking system is created, which assigns each video a rating.
You must also consider that on Youtube everythousands of videos are uploaded that can be listed as “New.” Without real-time processing using artificial intelligence and big data technology, YouTube simply could not cope with the issuance, and instead of the long-awaited new release of Amkal, a 7-year-old boy would have another music video or other video uninteresting to him in auto play.
Youtube as a training ground
Huge data stream is perfecttraining ground for AI. It turned out that the challenges spread on YouTube can be not only interesting, but also useful for studying augmented reality. As a result of processing more than 2000 videos with the popular mannequin-challenge in 2016, programmers got a good base for studying the depth of field of the video to improve VR technologies, the ultimate goal of which is the effect of full presence.
The terrible trend of streaming terrorist acts began several years ago and continues now. The shooter in Galle and the terrorist attack in New Zealand are far from the first such cases.
The US has already demanded that on all platforms withBy the possibility of streams, software was developed that allows identifying dangerous individuals in live broadcasts before they cause damage to civilians. AI and YouTube are already working on AI to remove terrorist content in collaboration with the Department of Justice and law enforcement.