First, Google does have a system in place to filter repeat offenders, but it's imperfect. When a video or audio file is uploaded, Youtube creates a fingerprint for the file and compares it against a list of banned fingerprints. If someone is trying to upload a file that has previously been banned, it can automatically be blocked. The problem? You only need to adjust the video a bit to change its fingerprint and dodge the ban. Change the audio track, trim a few seconds off the video, and add in a subtitle or credit, and you've now altered the video enough to get it back into the system.
Automatic recognition of illegal videos works very poorly, and is generally not used by any major sites. In the US, for example, the difference between a legal video of someones daughter playing by the pool in a swimsuit and an illegal child porn video of that same child in the same swimsuit by the same pool is simply her position and demeanor. Computers are nowhere near having the capability to comprehend that kind of nuance. Similarly, with these ISIS videos, there's no way that a computer can tell the difference between footage of a Nevada hunter walking across the desert with a rifle while looking for an elk to shoot, and footage of an ISIS jihadi walking across the desert with a rifle looking for an Iraqi soldier to shoot. Or the difference between a video of some terrorist blowing up a building with innocent people inside, or a video shot by a U.S. soldier in Iraq showing his own combat experiences.
Until computers figure out context and nuance, filtering these kinds of videos will largely remain a manual job.
= new reply since forum marked as read
