Is the Use of Biased Algorithms Morally Defensible?: Why Biased Algorithms Should be Used Under Certain Circumstances

Abstract

Literature on the ethics of algorithms, particularly biased algorithms, has primarily focused on the algorithms themselves and their direct outcomes rather than the total impact of their use. This has led to a substantial body of work on algorithmic fairness, including both ethical critiques of such algorithms and technical solutions aimed at mitigating bias. However, whether the use of biased algorithms is wrongful has not been questioned. Furthermore, even less has been written on biased algorithms that despite their bias, lead to better outcomes for all affected parties when compared to viable alternatives.

This thesis aims to address this gap by arguing, from a consequentialist perspective, that the use of certain biased algorithms can be morally defensible. Doing so by examining the validity of issues associated with using biased algorithms that are grounded in appeals to fairness, discrimination and structural inequalities. Furthermore, this paper also contends that existing frameworks, such as egalitarianism, in the field of algorithmic fairness are not suitable for evaluating whether such algorithms are morally defensible. This forms the motivation for using a consequentialist approach to examine the issues discussed in this paper. Lastly, this propose directions for further research, particularly on clarifying what ethical values should be considered when assessing whether the use of an algorithm is morally defensible or not.