Can the tribunal be taught incorrectly?

TomA3X·5/23/2015, 12:14:40 PM·1 votes·333 views

I was wondering if the new tribunal system might have an issue with being taught incorrectly. How exactly does the system understand which piece of chat was the reason for the report, because couldn't the system start to blacklist phrases like "gj" if they are common with all games in which people are reported?

Edit: For those who don't know, the new tribunal utilises a 'machine learning' backend. This means the system learns 'on the job' about what is a problem and what is not, and then makes a decision based on it's learned experiences. (At least, that's how I understand it)

1 Comments

OhBoyItsaMegaman5/23/2015, 2:02:38 PM1 votes

Even if "gj" appears in lots of games where people are reported, it also appears in lots of games where people are not reported.

As for how it figures out which piece of chat was the reason for the report, there are a lot of little factors that can give that away. Like, if two players are quickly typing things that are obviously negative and obviously aimed at each other, then the first line of their conversation is probably what started the argument, even if it was a seemingly positive phrase. If "best support NA" is typically a positive phrase, it is probably being used as a negative phrase if it starts an argument. So it will be seen as a phrase that is usually positive but can be negative under some circumstances. In order for the system to be completely wrong, it would have to be a positive phrase that's almost exclusively used sarcastically. In which case, yeah, you would potentially see people get punished for using it literally instead of sarcastically. But only if they were also exhibiting other toxic behavior. One line of chat wouldn't be enough to get you banned.