Google sister fights trolls with artificial intelligence
Big Data Against Hate Speech and Online Trolling: On Thursday, Google's sister company Jigsaw launched Perspective, a service designed to make it easier to identify comments online that are perceived as hateful, unfriendly, malicious, mean, or hurtful.
As part of the Digital News Initiative (DNI), Jigsaw will provide an interface (API) with which publishers can use Perspective free of charge. The service was previously tested in projects with the Wikimedia Foundation and the New York Times to automatically detect hate posts and trolling. Among other things, a data set of more than 115,000 messages from Wikipedia discussion pages was analyzed. The service is initially only available for English-language discussion forums. However, the project could be extended to other languages such as German as part of the DNI.
Protection against DoS attacks
Jigsaw, a company of Google's parent company Alphabet, has set itself the goal of "making the world safer with technology". As part of the DNI, Jigsaw's Project Shield offers publishers a free protective shield against so-called denial-of-service attacks, in which white-collar criminals or state-controlled hackers aim to paralyze publishers' websites with gigantic data packages from millions of hijacked computers from a botnet.
Jigsaw's top manager Jared Cohen said that Perspective's approach goes far beyond matching against a list of suspicious keywords. The artificial intelligence recognizes whether a post is deemed inappropriate by a human user and causes them to leave the conversation. A study has shown that 72 percent of all US Americans have already experienced harassment on the internet, and almost half of all users have also experienced this personally. Almost a third of those surveyed now censor themselves in online comments because they fear retaliation.
"Toxicity" scale
The Jigsaw system determines how "toxic" a comment is on a scale between 0 and 100. What then happens to the comments is up to the publishers. The system can be used to alert the publishers' moderators to suspicious comments, who can then intervene in the conversation.
However, users could also be shown how "toxic" the text is considered to be when writing the comment in order to contribute to a better tone. Readers could also sort the comments so that particularly hateful posts are hidden.
