Facebook researchers are working on an AI system that can detect gender biases in text, as outlined in a paper published by the company.
The researchers are looking at a framework that breaks down gender bias in text through several different factors. It looks at the bias from the gender of the person being spoken about, bias from the gender of the person being spoken to, and bias from the gender of the speaker.
Through these factors, the system tries to understand how nouns describing women are different from the nouns used to describe men.
“Distinguishing between gender bias along multiple dimensions is important, as it enables us to train finer-grained gender bias classifiers,” the researchers note.
The researchers used the system to detect gendered text in certain Wikipedia biographies and found that women’s biographies included more gendered text.
Since AI systems often reflect the gender biases that they are trained through, many companies have been implementing measures to prevent this, as outlined by VentureBeat. For instance, Google recently launched gender-specific translations to prevent gender bias.