Google is putting trolls on notice, announcing a new project dubbed Perspective which aims to take the vitriol out of comment sections online. Co-developed with Jigsaw, Perspective relies on machine learning to act as a constantly improving filter, just as a human admin might, only without the risk of the AI getting offended or overwhelmed along the way. Each comment is assessed for what potential impact that could have on the ongoing conversation.
Perspective takes the form of an API that plugs into existing sites, allowing developers to effectively harness the power of the AI in their comments sections. It spits out a score for each piece of feedback, based on training from thousands of previous comments that had been rated by a human admin. However, what’s done with that score is up to the publisher themselves.
For instance, positive comments deemed contributing strongly to the conversation could be given higher priority, and shown further up the page. On the flip side, negative comments like flaming or hate-speech could be hidden altogether, automatically. Alternatively, sites could opt to show a real-time rating, so that readers can see at a glance what might be worth reading and what’s considered less useful.
Indeed, given that Perspective works in real-time, it could even help shape comments as people are writing them. The system could give feedback as people type, “letting the commenter see the potential toxicity of their comment as they write it,” Google says. Alternatively, Perspective could merely act as a filter, pushing what might be negative comments to a human team.
Perspective is initially focused on English language, and on “toxic” comments. However, Google says that the goal is to expand the API to encompass more languages and other perspectives. One day, for instance, the plugin could flag when comments are off-topic, or even poorly argued or sourced.
For now, Google is offering the API for site-owners to try, and has already been working with the New York Times, Wikipedia, The Economist, and The Guardian to streamline the system in the real-world. At the moments it’s a closed beta, with interested parties invited to request a key; Google says they’ll be rolling out access through the year. If you want to see how well it works, there are demos at the Perspectives site which allow you to see how the filtering works in practice on topics like Brexit, Climate Change, and the US election.
If you’re wondering, the “comments” in the images come from an excellent “A Bit of Fry & Laurie” sketch