Paper ID
2886
Paper Type
short
Description
Notwithstanding the various benefits ascribed to using Data Analytics (DA) tools in support of decision-making, they have been blamed for their potential to generate discriminatory outputs. Although several purely technical methods have been proposed to help with this issue, they have proven to be inadequate. In this research-in-progress paper, we aim to address this gap by helping users detect discrimination, if any, in DA recommendations. By drawing upon the moral intensity literature and the literature on explaining black box models, we propose two decisional guidance mechanisms for DA users: (i) aggregated demographic information about the data subjects (ii) information on the variables that drive the DA output and the extent of their contribution along with information about demographics of the data set being analyzed. We suggest that these mechanisms can help decrease users’ readily acceptance of discriminatory DA recommendations. Moreover, we outline an experimental methodology to test our hypotheses.
Recommended Citation
Ebrahimi, Sepideh and Hassanein, Khaled, "Empowering Users to Detect Data Analytics Discriminatory Recommendations" (2019). ICIS 2019 Proceedings. 39.
https://aisel.aisnet.org/icis2019/cyber_security_privacy_ethics_IS/cyber_security_privacy/39
Empowering Users to Detect Data Analytics Discriminatory Recommendations
Notwithstanding the various benefits ascribed to using Data Analytics (DA) tools in support of decision-making, they have been blamed for their potential to generate discriminatory outputs. Although several purely technical methods have been proposed to help with this issue, they have proven to be inadequate. In this research-in-progress paper, we aim to address this gap by helping users detect discrimination, if any, in DA recommendations. By drawing upon the moral intensity literature and the literature on explaining black box models, we propose two decisional guidance mechanisms for DA users: (i) aggregated demographic information about the data subjects (ii) information on the variables that drive the DA output and the extent of their contribution along with information about demographics of the data set being analyzed. We suggest that these mechanisms can help decrease users’ readily acceptance of discriminatory DA recommendations. Moreover, we outline an experimental methodology to test our hypotheses.