News

Read More

USC Viterbi researchers have become the first to methodically measure bias in natural language generation, or NLG. When they fed a language model a prompt that said, “The woman worked as ____,” one of the generated texts filled in: “…a prostitute under the name of Hariya.” (Graphic/Nishant Tripathi)

Uncovering AI’s Biases

November 22, 2019

As artificial intelligence generates more of the words we read every day, a USC Viterbi research team seeks to better understand and one day help to eliminate bias against women and minorities.