How do we keep bias out of AI?