Some countries allow people to be reported for their online hate comments. AI can help, by finding the insults that are punishable by law. Data scientist Marcel Schliebs explains how this works.
Category
🗞
NewsTranscript
00:00Data scientist Marcel Schliebs developed the AI-supported software that SODAN is based on.
00:06It scans through comments on social media, flagging any potentially criminal content.
00:14The AI is already very good at identifying many examples.
00:19The software scans publicly accessible comments for its clients.
00:23It searches for groups of words that constitute known criminal offences,
00:27like insults or incitement to hatred and violence.
00:30Thanks to judicial databases, it is constantly learning.
00:34But one problem remains, and it's not a technical one.
00:42In all honesty, there is not yet much legal precedent in this area.
00:48When it comes to punishable online insults, threats and hate crime,
00:52we are actually leading the way in creating this precedent.
00:58The AI software uses semantic vectors.
01:01That means that similar terms are grouped together.
01:06This way, it also discovers new potentially punishable words.
01:10They are then catalogued for inspection.
01:17Sexist insults like k*** or slut are filed in one space.
01:21Next, words like whore or skank.
01:24They have completely different letters but are semantically very similar.
01:28Of course, that's not enough to say it's also a criminal offence to use this word.
01:32But it is helpful for pre-sorting, to then go over it again and take a closer look.
01:40The final legal check is still performed by human beings, though.
01:44Sometimes, the internet feels lawless.
01:47Franziska Brandmann and her team are working to change just that.