is it even AI? They claim it's a language model analysing human bias but it's blindingly obvious that it is just a human curated blacklist.
Maybe they're using a language model for part of speech tagging or to suggest alternative words with similar meaning but there's no way an AI decided that "motherboard" is taboo.
Yes an AI in this case would be more intelligent than whichever human designed this feature. Google is actually a pioneer in the space of AI comment toxicity filters.
Even a toy solution would be smarter, you could compare the TF-IDF of each word in different populations (woke, average, toxic) to identify non-woke language, if that was the intention - not saying it’s right.
This is clearly just someone who ran a regex over a dictionary and didn’t check the output of their script for scunthorpes such as “motherboard”. It’s so low tech for Google, 1990s technology.
Maybe they're using a language model for part of speech tagging or to suggest alternative words with similar meaning but there's no way an AI decided that "motherboard" is taboo.