For O’Neil, evaluating the quality of an algorithm is not only contingent on assessing the data employed to train it, but determined by its impact on everyone who might be affected by the decisions it makes.