Google’s parent company has handed developers an anti-trolling tool.
Jigsaw, an incubator and think tank that sits inside Alphabet (Google’s umbrella company), has released an API called Perspective, which uses machine learning to measure whether an online comment counts as harassment or abuse, as reported by Wired.
It’s already being tested by Wikipedia, the Guardian, the Economist, and the New York Times. Wikipedia said it’s looking to reduce harassment of volunteer editors, which comes shortly after one female editor revealed the levels of abuse she had endured for years.
Jigsaw has built a demo interface which you can test out. Type in a word or sentence, and Perspective will score your comment by toxicity. If you write something that’s scored 80% toxic, chances are that it will offend people online.
Business Insider tried out the tool and results show that it’s still pretty early days for Perspective.
The phrase ‘men’s rights activist’ was deemed less toxic than…
And Perspective recognised some common abusive language…
…but not all
Jigsaw says it’s “still early days,” and asks users to flag it up when Perspective gets it wrong to help it train its model.
Perspective, built by Jigsaw’s counter abuse technology team, is part of a wider joint project called Conversation AI. The project is attempting to clean up online conversation with the help of machine learning.
Get the latest Google stock price here.