“…Methods in the first family consist in pre-processing the data or in extracting representations that do not contain undesired biases (see e.g. Beutel et al [2017], , Calmon et al [2017], Chierichetti et al [2017], Edwards and Storkey [2015], Feldman [2015], , Fish and Lelkes [2015], Gordaliza et al [2019], Johndrow et al [2019], , 2012, Madras et al [2018a], Song et al [2018], Zemel et al [2013]), which can then be used as input to a standard machine learning model.…”