Dataset used in NLP task since 2003 is highly biased

Research by Scale AI on CoNLL-2003, a dataset used to train and test NLP models found it to be highly biased. This could have potentially lead to supression of search results involving women or biases in voice assistants.

While most discussion against AI involves FRT, biases in NLP can also have huge repercussions. The article by one zero explains them.