Not super toxic, but supposedly the AI has the ability to block everything that contains namecalling and labeling.
If this is customizable, I could see popular forums and websites using this to create a digital moderator who would lighten the load for the real ones (either because the site is so toxic as to overwhelm them or it’s so big it’s not possible to go through everything).
In a sense, with people on sites like YouTube losing their temper when discussing things as benign as puppies, something like this may actually become a necessity.
Machine learning is on its developing stage, so no wonder it makes mistakes. It was made to hide criticizing comments and leave only positive ones. Applied positive-thinking)))
There’s a comedy series (written series) online called “the BOFH”, it has an episode about trying to make an email filter to do a similar thing. Infact I think it has a few episodes about it, they quickly find that the only way to stop offence is to stop everything. Then the real hilarity ensues.