How

The chapter proposes some algorithm audit methods in the case that the algorithm is accessible (white box), as well as when it is not accessible (black box). The black box approach analyses algorithms without any knowledge of their codes. This is the only possible approach when the provider of the algorithm does not agree to disclose the code. The white box approach assumes that analysis of the code is possible. Unobtrusive and obtrusive methods coming from the social sciences are considered. Unobtrusiveness may help in detecting bias and unfairness. Four main social research methods are presented throughout the chapter. Experiments may consist of testing the algorithm with a different set of equivalent data to prove that it works similarly. Ethnography concentrates on how algorithms are generated. It helps to specify how cultural, symbolic, and normative values may promote certain images of the social world through algorithms. A survey is justified every time there is the objective of collecting opinions, attitudes, and behaviours of individuals impacted by automation. The personas walkthrough method turned out to be significant in finding assumptions about automated systems that can limit action on the system to people with specific characteristics.