Towards More Responsible Data-driven Decisions
Human decision makers often receive assistance from data-driven algorithmic systems for evaluating objects, including individuals. When there are multiple criteria to be considered, two common ways of supporting the decision making are (i) to assign a score to each object and rank them accordingly, or (ii) to offer a small set of maxima representatives that include the “best” for different users. The evaluations and decisions made based on data can have significant consequences for the individuals evaluated, as well as society. For example, a company may promote high-ranked employees and fire low-ranked ones. University rankings is an example of social impact where it is well-documented that the ranking formula has a significant effect on policies adopted by universities. Another important example is highlighted by ProPublica: judges in the US consider the scores assigned to the individuals based on their criminal record and their background, as guidance when sentencing criminals.My research's goal is to ensure that decisions based on data are made responsibly, that is, properties such as fairness, stability, diversity, and transparency are satisfied.