The issue at hand with any sort of AI is the data. The data sets are pure and can't hide bias that humans otherwise can. This is just once again showing clearly what the data contains and can only be addressed with better/new data. Model weights can help somewhat but the best fix is going to be better data sets.