I've said before that comments on internet posts are a double-edged sword; on the one hand, you can have fantastic conversations with them or you get a toxic cesspit with people spitting whatever drivel comes to mind. Google is hoping to combat the latter with its newest machine learning initiative called Perspective.

In cooperation with Jigsaw (another one of Alphabet's attempts to tackle global security challenges), Perspective will launch with an API that publishers and platforms can use on their sites. It will review comments and score them based on how closely they resemble those that were previously identified as "toxic" by other readers and moderators. As an example, check out the GIF below: that statement is similar enough to other comments that people have called "toxic" in the past.

From there, it's up to the publishers or platforms to decide what they want to do with the data. They could flag any comments for later human review, show trollish commenters the impact of their words as they write them (I'd like to see that), or let readers sort comments by the level of toxicity to help filter out the dregs.

Over time, Perspective will learn to identify the different levels of the poor and antagonistic use of language, especially as it is exposed to the wide world. This is limited to English for now, but Google plans to expand it to other languages. There is also a hope to improve the model to recognize off-topic or unsubstantial comments.

Source: The Keyword