There currently is no feedback loop initiated by the system. As Glorfindel notes, moderators get a system-generated flag that asks them to go investigate and see if anyone needs a little time away, they heard her say, from each other (poorly quoting a Chicago song). But there's no automatic system message providing JIT (just-in-time) guidance to the user writing the comments.
That could change in the near future because:
- Our data science team has come up with a classification engine that pretty reliably identifies comments that are likely to be perceived to be hostile, and,
- Our design team is unifying a bunch of stuff post-responsive changes which gives them a much better framework to display just-in-time messages to folks when we detect problematic patterns.
So, in a month or two, we may show a warning when it appears that what you're writing is at least likely to be perceived (thus, understood) to be hostile, and the warning will assume you have the best intentions but just don't realize that the tone might be off. Likewise, we would be able to let someone know that some of their recent comments didn't go over so well and were removed, and ask them to pay closer attention to how their comments are likely to be received.
Nothing 100% concrete yet, and of course this wouldn't fire if your comments from 2009 were deleted by the time this code started interacting with the comments table, but the plan is to let the system get in front of it when there's a high degree of certainty (and possibly still leave a note for moderators, even just so they know the user saw a warning).
The new AI stuff is wild, I can't wait for the data team to make it more available for everyone to play with. It can also predict stuff like homework assignments in questions with a crazy level of accuracy. But, a big part of being able to do this is the new responsive framework, because prior to that it was near impossible to code dialogs that scaled properly without racking up even more debt.