>>14234634overused and repetitive memes aside, the models are simply not good enough because the ones mentioned are trying to do too many things at once.
Machine Learning in its current state is only useful in very specific domains where the inputs are all more or less the same conceptually, e.g., disease recognition in medical imagery, hand written latin text recognition, flower species recognition, etc.
Training data sets can not be large enough for the model to be precise enough, they can only give you approximate outputs, in comparison to more specialized models, that can not recognize shit outside of what they were trained for, but are better than any human at that one task they were designed for.
In this case, Facebook was retarded enough to believe that they could have a highly functional AI despite the unimaginable diversity of content the platform hosts, the AI fucks up probably a LOT, but users never give it much attention, but in the case of mistaking a black man for a primate, people will get offended, making it seem like the AI has some sort of bias, when really it has never been able to do its job properly to begin with.
