Instagram now uses AI to block offensive comments – The Verge

Instagram is introducing an enhanced comment filter today meant to wipe out nasty remarks using AI. The app first began offering a comment filter last September, but it was a very simple approach: Instagram would only remove comments that contained words and phrases it had specifically identified as offensive. (Users could also add their own custom banned phrases.)

Now, the system is getting a lot smarter. It uses machine learning to identify comments that seem offensive, giving the system some ability to take into account the replys context, potentially catching more bad comments and cutting down on false positives at the same time. Wired has a big story on how the system was made, and it mentions that when a comment gets flagged, itll be blocked for everyone except the person who wrote it, so they wont know their remark didnt get through.

An AI spam filter has secretly been in place since October

One other notable change here: Instagram is turning the offensive comment filter on by default, whereas the earlier filter had to be enabled. Youll still be given the option to turn it off from inside the apps settings, and Instagram still includes the ability to block custom words and phrases.

The filter only works in English at launch, but Instagram says its working to expand it to other languages over time.

Instagram is also announcing an AI spam-filtering system today, too. The spam filter has secretly been in place since last October, but its only being revealed today. Given that no one has noticed it in the past nine months, the filter probably isnt blocking too many comments that it shouldnt. That filter is active in nine languages, including English. (As a side note: Instagram really needs a better system for blocking spam accounts, as well. I set my profile to private recently in order to cut down on spam followers, but now Im just getting follow requests from spam accounts instead.)

View original post here:

Instagram now uses AI to block offensive comments - The Verge