Gender-stereotyped career and salary advice defines the current failure of AI to eliminate prejudice, according to Mahzarin Banaji. The Harvard professor argues that models reproduce the biases of their creators. This feedback loop risks automating systemic discrimination.