Anti-discrimination Analysis Using Privacy Attack Strategies

Abstract

Abstract. Social discrimination discovery from data is an important task to identify illegal and unethical discriminatory patterns towards protected-by-law groups, e.g., ethnic minorities. We deploy privacy attack strategies as tools for discrimination discovery under hard assumptions which have rarely tackled in the literature: indirect discrimination discovery, privacy-aware discrimination discovery, and discrimination data recovery. The intuition comes from the intriguing parallel between the role of the anti-discrimination authority in the three scenarios above and the role of an attacker in private data publishing. We design strategies and algorithms inspired/based on Frèchet bounds attacks, attribute inference attacks, and minimality attacks to the purpose of unveiling hidden discriminatory practices. Experimental results show that they can be effective tools in the hands of anti-discrimination authorities.

Managing bias and unfairness in data for decision support: a survey of machine learning and data engineering approaches to identify and mitigate bias and unfairness within data management and analytics systems

“…Many data mining papers, dating from 2008 to 2016, deal with discovering and measuring discrimination within datasets, the results being potentially useful for "debugging" the datasets for later training machine learning models. They investigate scenarios of direct and indirect discrimination, further complicated by additional privacy concerns [151] and cases where the protected attributes are unavailable. Methods.…”

Measuring discrimination in algorithmic decision making

“…Discrimination discovery aims at finding discriminatory patterns in data using data mining methods. A data mining approach for discrimination discovery typically extracts association and classification rules from data, and then evaluates those rules in terms of potential discrimination [28,39,40,47,49,53,54]. A more traditional statistical approach to discrimination discovery typically fits a regression model to the data including the protected characteristics (such as race or gender), and then analyzes the magnitude and statistical significance of the regression slopes at the protected attributes (e.g.…”

How Data Mining and Machine Learning Evolved from Relational Data Base to Data Science

“…The process is guided by legally grounded measures of discrimination, possibly including statistical tests of confidence. An alternative view of "discovery as attack" is investigated in [75], in which attack strategies of privacy models are used to unveil discrimination hidden behind redlining practices. Discrimination against individuals has been instead modeled with a k-NN approach, following the legal methodology of situation testing, and applied to a real case study in research project funding [76].…”