ASSESSING PUBLIC OPINION ON ALGORITHMIC FAIRNESS Reviewing practical challenges and the role of contextual factors

Abstract

AI ethicists often claim that where algorithmic decision-making is impacting human lives, it is crucial to strive for transparency and explainability. As one form of achieving these, some authors have argued for socio-technical design of AI systems that involves the user in the design process. And while there is no shortage of cases where this step is absent due to blatant disregard for users’ interests, one can say that even where that is not the case, this is no easy task due to a mounting knowledge gap among the general public on the subject of AI. This Master thesis aims to demonstrate the above issue in concrete terms by attempting to collect public opinion on algorithmic fairness. The survey conducted for this thesis asks participants to pick among four different algorithmic models that they think achieves the best fairness in the presented scenarios. Results indicate that (1) contextual factors do play a role, and (2) that attempting to collect public opinion on the subject is challenging as there is insufficient knowledge on the topic and, therefore, poor understanding of the presented options. As urgent as it is to conduct public consultations on algorithmic decisionmaking where human lives are increasingly impacted, it is even more urgent to improve public knowledge on the subject so that people could actually make informed choices. Understanding the complexity of contextual factors offers substantial support in that endeavor.