Google Will Use AI to Flag Offensive Videos (GOOG, T)

Posted April 06, 2017

"When we spoke with many of our top brand advertisers, it was clear that the videos they had flagged received less than 1/1000th of a per cent of the advertisers' total impressions", Google chief business officer Philipp Schindler said.

Google now says that it has applied new machine-learning systems that, in the past couple of weeks, have identified five times more videos that are not "brand safe". "While they know that no system can be flawless, they appreciate the actions we've taken and know we are taking this seriously and are committed to getting better and better", Mr Schindler said.

Google's YouTube division was doing its best to play down the controversy surrounding its advertising practices over the weekend, days after a number of large clients pulled their spending from the video platform.

"It has always been a small problem", with "very very very small numbers" of ads, Schindler told Recode.

Companies including Omnicom and GroupM are aiming to tackle brand safety issues using third-party verification technology, while YouTube itself is reportedly exploring partnering with third-party firms to improve protections, while also calling the problem "very very very small".

Some publishers and ad agencies have called on Google and rival Facebook to more actively police the content they host online. And his statements about the boycott showed how convoluted Google's position is.

Google aims to implement most of the changes by Sunday, according to the memo, a copy of which was obtained by Bloomberg News.

OpenSlate - a social video analytics company - is able to rank YouTube content for quality and brand safety and provide contextual insights to the agency's advertisers. That includes language that promotes negative stereotypes about targeted groups or denies "sensitive historical events" such as the Holocaust.

"It is natural for users to read up on "exciting news" on various websites out of personal curiosity rather than an endorsement of the content of the sites visited".

Some researchers argue digital platforms should rely on humans to make these editorial decisions.

The Wall Street Journal (WSJ) recently targeted the world's most popular YouTuber, Pewdiepie, with claims that the Swede was making frequently making anti-Semitic jokes.

Nor is the company willing to alter YouTube's fundamental formula. It's also likely to share best practices with advertisers akin to what it's done with SEO algorithms: "We'll give you best practices on things that will affect your search rankings-this will be the same thing-but no secret sauce will be shared", Ryan said. The predicament for Google is to manage the placement of ads for a user without betraying the trust of it's customers. After the British government pulled ads from YouTube in recent weeks, companies including Johnson & Johnson, AT&T, Sainsbury's, Toyota, Volkswagen and the BBC followed.

The executive likened Google's ad business to an airline: Each faces long-tail risk beyond its control. "You can't guarantee it".