Why is it considered acceptable risk to deploy something that can't recognize large swaths of the population. If your FR fails in tests to recognize Oprah Winfrey - it's broken.