To train Perspective, Jigsaw pointed it at hundreds of thousands of comments that were identified as toxic by human reviewers. From there, the machine learning software would look at other comments and score them based on similarities. As with all machine learning systems, Perspective gets smarter and more accurate each time it finds new examples of toxic comments and every time humans guide it and correct things that it may mis-identify.
Jigsaw hasn't just been testing and training Perspective in its own labs -- the company says it has partnered with The New York Times to test it. Currently, the NYT has moderators who review every single comment before it is posted; that team has to review about 11,000 comments every day. But that desire for manual review means that the publication only has comments open on about 10 percent of its stories. Through working with Jigsaw and Perspective, the NYT's moderation team can go through comments more quickly and open up comments on more articles.
Perspective isn't the only tool Jigsaw is working on to make the internet a more hospitable place, either. Last year, the company let any news organization sign up to use Project Shield, a tool that defends against DDoS attacks. As useful as that may be, Perspective could be something that impacts the way people view publications and news sites every day -- provided it works, of course. Here's hoping the NYT and Jigsaw have some hard data to share about how Perspective has impacted the site's comments section soon.