Interesting. I have my suspicions about how this happened. The training data would have been made up of content containing hate speech and comments that aren't. However, the hate speech would be directed at a variety of groups. So, the algorithm would be picking up on the rest of it. For example, the algorithm might detect "Kill all [group]" as hate speech.
If I'm right, the "fix" for the algorithm would be a simple filter that would discard hate speech directed at the acceptable groups.
it's entirely that anti-white hate speech is not normally shunned and censored by people. hell, in many parts of the internet, it's celebrated. so of course there will be more of it found by impartial algorithms.
Interesting. I have my suspicions about how this happened. The training data would have been made up of content containing hate speech and comments that aren't. However, the hate speech would be directed at a variety of groups. So, the algorithm would be picking up on the rest of it. For example, the algorithm might detect "Kill all [group]" as hate speech.
If I'm right, the "fix" for the algorithm would be a simple filter that would discard hate speech directed at the acceptable groups.
it's entirely that anti-white hate speech is not normally shunned and censored by people. hell, in many parts of the internet, it's celebrated. so of course there will be more of it found by impartial algorithms.
Well, yes. I'm just talking about why the algorithm picks it up at all when that's clearly not what its creators intended.