A couple months back the company announced it would be increasing its ability to find and flag extreme videos by implementing machine-learning, and the results are in: AI has proven to be a dramatic upgrade over humans when it comes to flagging terrorist content.
On Google’s blog the company revealed a significant increase over previous efforts, including being able to flag posts before a human can 75 percent of the time. Researchers also report the AI is able to review twice the content humans can; that number is expected to increase as they further develop the system.
YouTube is going beyond simply flagging and removing extremist videos, or re-directing searches for that content in order to debunk and de-legitimize extreme viewpoints, as we previously reported.
Google will also impose stricter guidelines on all videos, even for content that doesn’t actually violate policy — according to a post on its official blog yesterday.
In the next few weeks Google’s AI will begin policing videos that get reported as being hate-speech, offensive, or violent in new ways. Even if it doesn’t violate any specific YouTube policies, it can be placed in a state of limited access. This restriction would keep posters from monetizing videos, and block comments, likes, and search prioritization.