Preventing Algorithmic Bias in the Development of Algorithmic Decision-Making Systems: A Delphi Study

In this digital era, we encounter automated decisions made about or on behalf of us by the so called Algorithmic Decision-Making (ADM) systems. While ADM systems can provide promising business opportunities, their implementation poses numerous challenges. Algorithmic bias that can enter these systems may result in systematical discrimination and unfair decisions by favoring certain individuals over others. Several approaches have been proposed to correct erroneous decision-making in the form of algorithmic bias. However, proposed remedies have mostly dealt with identifying algorithmic bias after the unfair decision has been made rather than preventing it. In this study, we use Delphi method to propose an ADM systems development process and identify sources of algorithmic bias at each step of this process together with remedies. Our outputs can pave the way to achieve ethics-by-design for fair and trustworthy ADM systems.