Google’s new Perspective API can help you not sound like a jerk while commenting
Google and Jigsaw launch machine learning platform Perspective to identify comments that may be labeled as “toxic,” and you can start using it now.
Google and Jigsaw launched Perspective to help “improve conversations” by identifying which comments may have the potential for harassment or abuse, and it is already available to use on their website.
Perspective’s API uses machine learning models to score the perceived impact a comment might have on a conversation. Its first model is being used to identify whether a comment could be perceived as toxic.
The results can be used by developers and publishers to give realtime feedback to commenters, help moderators do their job, or allow readers to more easily find relevant information, according to the official release.
Perspective reviews comments and scores them based on how similar they are to comments people said were toxic or likely to make someone leave a conversation.
You can test this new service by typing whatever you’d like on their website in order to see “the potential impact of your writing” and the toxicity percentage it carries.
While the test seems accurate enough for weeding out key words or phrases that appear to be toxic, it will be interesting to see what perspective Perspective has on context, so let’s give it a shot!
For example, if I were to say, “I fart in your general direction,” the toxicity rating is 56%, but if I use a euphemism for the word “fart,” the score drops drastically.
Thus, “I break wind in your general direction,” becomes just 8% toxic while “pass gas” becomes 14% even though they basically mean the same thing.
This little experiment, although not carried out and repeated using the full scientific method, provides some evidence that Perspective’s toxicity rating is not based upon a deeper context, but rather individual word choice.
In other words, you could write the same concept two different ways and get an entirely different toxicity rating, which is maybe what it is all about — deciphering which trigger words are most likely to get a negative reaction.
In a way, this new API from Google can help teach you how to better insult someone without coming off as being a jerk, or how to soften your tone, so as not to offend anyone while still getting your point across.
Either way, it’s a fun tool that is still under construction, and anyone can play around with it.
For those that suffer from “foot-in-mouth” it can also be very useful to use Perspective’s tool as you hover over the “send” button if you are ever worried about how your message will be generally received.