Most users cannot identify AI bias, even in training data