Google: Robots Better than Humans at Spotting Extremist Videos

Google says its artificial intelligence system is better at identifying guideline-breaking videos than human moderators. The claim is somewhat subjective, however.

The company, which runs YouTube, made the claim as part of an update on how it deals with 'questionable' videos.

Last month, Google said it was refining its use of computerized systems for examining videos and checking whether they breach content guidelines. It says that as a result of the improvements, three-quarters of removed videos are now caught automatically before the company receives a complaint from a human viewer.

AI Works Quickly and Accurately

It also says the accuracy with which the automated system has improved dramatically and "in many cases our systems have proven more accurate than humans at flagging videos that need to be removed." That's an arguable point, however, as whether a video breaches the rules is ultimately a subjective issue. (Source: telegraph.co.uk)

The improvements to the automated reviewing were designed to do a better job of assessing context. For example, the same images used in a news report might be acceptable on YouTube, while being banned when used as part of a pro-terrorist recruitment video.

As well as being faster and sometimes more accurate, the key to the automated systems is that they can be scaled far more easily than recruiting more human moderators. In the past month Google says it's been able to double the number of extremist videos taken down and halve the average time they stay up before being removed.

Controversial Content Carries No Ads

There's still room for humans, however. Google says that since its last update on the topic, it has added 15 organizations to its list of advisors who help it identify extremist content as radicalization tactics.

Google also said it's just about ready to start implementing a new classification of videos that don't break the precise wording of its terms and conditions but are near the knuckle. It says videos which contain "controversial religious or supremacist content" will be placed in a special "limited state" category that means they don't carry advertisements, have comments switched off, and don't have the usual "like/dislike" feature. (Source: googleblog.com)

What's Your Opinion?

Do you trust automated systems to accurately detect a video that is offensive and breaks guidelines? Are such systems the only way to cope with the sheer number of videos YouTube hosts? Is the "limited state" category a sensible way to deal with controversial content?