Facebook has a major problem with the fake news . Its no a secret. Groups and individuals of diverse filiation use the popularity of the social network to spread all kinds of information, some patently false, that reach the eyes of a vast mass of users who do not exercise their critical sense too much. That said, there is a possibility that the next big political scandal based on a fake news is likely even for experts, and deep fakes could be the perfect tool to light the fuse.
This type of montages use artificial intelligence techniques to replace the faces of a video and even synchronize their features with any type of audio, allowing the creation of surprisingly credible videos. Most publicly available applications leave something to be desired, but using good quality material and the right software some videos may require several viewings to determine their authenticity.
The availability of relatively simple tools has made deep fakes a hobby for some users.
For this reason, and since Facebook is a means of information of first order for millions of people around the world, the social network has expanded its program of checkers to determine the validity of the videos published on the platform. The 27 organizations that collaborate with Facebook to assess the validity of their news may also score videos and static images rated as potentially fraudulent by an automated learning system.
Essentially, Facebook’s artificial intelligence will examine photos and videos until they encounter some kind of suspicious content, which will then be sent to a human team made up of organizations such as Snopes, Politifact or Associated Press. Journalists and fact checkersthey will check their authenticity using visual verification, but also reverse searches and metadata analysis. Receive the human assessment, the artificial intelligence of Facebook will improve the accuracy of your future detections.
The concern for deep fakes began to spread after videos like this created by the University of Washington.
Developing (and training) this artificial intelligence at optimal levels will not be an easy task. Facebook recognizes that right now it is more successful using OCR techniques in videos to detect fraudulent texts inserted in videos that verify the transcription of the audio of a video to verify its authenticity. Making the jump to deep fakes , which can be even more deceptive, will therefore require considerable technical and human effort.