4 ms·

Having built tools to organize and drive human curation, human review only means that you get different errors and different people get fired as scapegoats. Whe

by saulrh 2y ago

Having built tools to organize and drive human curation, human review only means that you get different errors and different people get fired as scapegoats. When you're doing content moderation, rather than simple data curation, you don't even get "far" better results - you frequently get worse results! - and you start traumatizing your moderators enough to destroy them. On top of that, who the fuck would ever use a product if their private notepad was going to be read by a human moderator? There is no easy answer here.

maeil 2y ago
The very least you could do is have human review of these cases where accounts are shut down, have an actual human appeal system. You can't convince me that this wouldn't significantly decrease the false positive rate (without significantly increasing false negatives), learning to better outcomes. Obviously when appealing you'd consent to having your content reviewed by a human. > On top of that, who the fuck would ever use a product if their private notepad was going to be read by a human moderator? There is no easy answer here. Besides, we've gotten to a point where it's not clear if A. every single thing being ingested as data to create a profile of me off (current situation, e.g. Gmail) is better than B. Only a portion that has been flagged being reviewed by a human. There's a lot of low-hanging fruit which humans-in-the-loop would be used for to give far better results. A very obvious example being Meta's key role in the Rohingya genocide. Plenty of outsiders banged on Facebook's door, but they wouldn't listen, costing thousands of people their lives. There's lots of such "obvious" cases. Content moderation is extremely difficult when it comes to things that toe the line, sure, but I wasn't really talking about those.

4 ms·

Having built tools to organize and drive human curation, human review only means that you get different errors and different people get fired as scapegoats. Whe

by saulrh 2y ago

maeil 2y ago
The very least you could do is have human review of these cases where accounts are shut down, have an actual human appeal system. You can't convince me that this wouldn't significantly decrease the false positive rate (without significantly increasing false negatives), learning to better outcomes. Obviously when appealing you'd consent to having your content reviewed by a human. > On top of that, who the fuck would ever use a product if their private notepad was going to be read by a human moderator? There is no easy answer here. Besides, we've gotten to a point where it's not clear if A. every single thing being ingested as data to create a profile of me off (current situation, e.g. Gmail) is better than B. Only a portion that has been flagged being reviewed by a human. There's a lot of low-hanging fruit which humans-in-the-loop would be used for to give far better results. A very obvious example being Meta's key role in the Rohingya genocide. Plenty of outsiders banged on Facebook's door, but they wouldn't listen, costing thousands of people their lives. There's lots of such "obvious" cases. Content moderation is extremely difficult when it comes to things that toe the line, sure, but I wasn't really talking about those.