Researchers expose biases in datasets used to train…

Artificial intelligence (AI) has a bias problem. Word embedding, a common algorithmic training technique that involves linking words to vectors, unavoidably picks up — and at worst amplifies — prejudices implicit in source text and…