Algorithmic Bias

• DOI: 10.1145/2939672.2945386
• Publication Date: 2016-08-08T18:33:46Z
• Authors: Sara Hajian; Francesco Bonchi; Carlos Castillo
• Keywords: Ask price; Discriminative model; Targeted advertising

Algorithms and decision making based on Big Data have become pervasive in all aspects of our daily lives (offline online), as they essential tools personal finance, health care, hiring, housing, education, policies. It is therefore societal ethical importance to ask whether these algorithms can be discriminative grounds such gender, ethnicity, or status. turns out that the answer positive: for instance, recent studies context online advertising show ads high-income jobs are presented men much more often than women [Datta et al., 2015]; arrest records significantly likely up searches distinctively black names [Sweeney, 2013]. This algorithmic bias exists even when there no discrimination intention developer algorithm. Sometimes it may inherent data sources used (software decisions reflect, amplify, results historical discrimination), but sensitive attributes been suppressed from input, a well trained machine learning algorithm still discriminate basis because correlations existing data. These considerations call development mining systems which discrimination-conscious by-design. novel challenging research area community.