31 ms·
The reason why mangoes struggle with this is because it's basically impossible. Maybe if you come up with an Artificial General Mango it could work, but short o
by Aerroon 5y ago
The reason why mangoes struggle with this is because it's basically impossible. Maybe if you come up with an Artificial General Mango it could work, but short of that it's not going to be good enough.
The main problem is that "toxic speech" depends on the context. It's not actually based on the specific words or phrases used. Eg calling someone a "rabbit" could just be a nickname or it could be bullying.
It goes further than this - language evolves. When filters appear, people make up euphemisms to express what they mean in a different way. Mangoes need to be able to learn and understand new language as quickly as humans can to be able to properly analyze these kinds of messages. So far, it doesn't seem like mangoes are capable of that.
Meanwhile, every time the mangoes get it wrong, people are upset. They find the system silly.
- giuliomagnifico 5y agoYes, is already hard to do for the technology evolution itself, plus, as the article says: if people/developers don’t agree on which training/instructions give to algorithms, how the algorithms make a good decision?
- mimixco 5y agoThis is the same reason that facial recognition software overwhelmingly identifies people of color incorrectly. If you don't involve people of color in the software development, how could the algorithms identify them correctly?