Alphabet incubator Jigsaw has released an experimental Chrome extension that’s meant to make our online browsing experience a little bit nicer by freeing it of toxic comments. The extension, called Tune, lets you choose to see varying levels of polite or aggressive comments. “Zen mode” turns off comments entirely, while “volume levels” ranging from “quiet” to “blaring” let through different amounts of toxicity (like attacks, insults, profanity, etc).
The open-source extension uses machine learning to determine how likely a comment is to be perceived as toxic. It uses Perspective, an API created in 2017 by Jigsaw and Google’s Counter Abuse Technology team, and it’s used by news organizations including The New York Times and The Guardian to experiment with online moderation. Below is an example of how Perspective sorts comments by toxicity:
Jigsaw is careful to note that Tune is an experiment that’s still in progress, and it can be inaccurate at labeling which comments are toxic. The extension “isn’t meant to be a solution for direct targets of harassment (for whom seeing direct threats can be vital for their safety), nor is Tune a solution for all toxicity,” the extension’s description reads. Rather, it’s meant to show users how machine learning can be used to improve discussions online.
So far, the extension is available for YouTube, Reddit, Facebook, Twitter, and Disqus comments. Filtered-out comments show up blank with a dot, which can be opened to reveal the actual comment. Using it on my own Twitter feed set to the lowest “Quiet” setting, it incorrectly labeled this “devastating Beto attack ad” that’s actually a parody video as toxic. But can you really blame the AI? Machine learning still has a ways to go before it can detect layers of irony, and maybe this serves as an example that choosing to ignore things we don’t want to see can ultimately do more harm than good — or just make us miss a funny video.