Google’s Jigsaw Gets a New Perspective: Learning Curve Still a Hairpin

Google’s Jigsaw Gets a New Perspective: Learning Curve Still a Hairpin

Late last year, I commented on Google’s Jigsaw software, created to apply machine learning to detect and remove harassment and abusive content in areas where users interact online.  At the time, I said that no matter how much Jigsaw learned, it would never be smart enough to replace human moderators.

Recently, Google’s Counter Abuse Technology Team released Perspective, the newest Jigsaw tool.  It’s an API that allows users to tap Jigsaw’s library of millions of words and phrases and determine a message’s “toxicity.”  Perspective scans each message and produces a toxicity rating as a percent, based on what panels of users have thought of similar comments.  Each comment, for example, is rated as “8 percent similar to phrases people said were “toxic.””

While several sites, such as the New York Times, are giving it a try (you can try it as well at the Perspective Demo Site), there are many that believe that, while helpful, it will never be used more than as a first pass to flag content for subsequent human review.  Perspective advances the learning curve of using Artificial Intelligence to combat online trolling, but it also further illustrates the continuing value of human moderators.