On the (im)possibility of fairness

What does it mean for an algorithm to be fair? Different papers use different notions of algorithmic fairness, and although these appear internally consistent, they also seem mutually incompatible. We present a mathematical setting in which the distinctions in previous papers can be made formal. In addition to characterizing the spaces of inputs (the "observed" space) and outputs (the "decision" space), we introduce the notion of a construct space: a space that captures unobservable, but meaningful variables for the prediction. We show that in order to prove desirable properties of the entire decision-making process, different mechanisms for fairness require different assumptions about the nature of the mapping from construct space to decision space. The results in this paper imply that future treatments of algorithmic fairness should more explicitly state assumptions about the relationship between constructs and observations.

[1]  Jun Sakuma,et al.  Fairness-aware Learning through Regularization Approach , 2011, 2011 IEEE 11th International Conference on Data Mining Workshops.

[2]  Salvatore Ruggieri,et al.  Using t-closeness anonymity to control for non-discrimination , 2015, Trans. Data Priv..

[3]  Mark Wilson,et al.  Unfair Treatment? The Case of Freedle, the SAT, and the Standardization Approach to Differential Item Functioning , 2010 .

[4]  Anthony F. Heath,et al.  Equality of Opportunity , 2017 .

[5]  Xiangliang Zhang,et al.  Decision Theory for Discrimination-Aware Classification , 2012, 2012 IEEE 12th International Conference on Data Mining.

[6]  Facundo Mémoli,et al.  Gromov–Wasserstein Distances and the Metric Approach to Object Matching , 2011, Found. Comput. Math..

[7]  Benjamin Fish,et al.  A Confidence-Based Approach for Balancing Fairness and Accuracy , 2016, SDM.

[8]  Indre Zliobaite,et al.  Measuring discrimination in algorithmic decision making , 2017, Data Mining and Knowledge Discovery.

[9]  Michelle Alexander,et al.  The New Jim Crow: Mass Incarceration in the Age of Colorblindness A Case Study on the Role of Books in Leveraging Social Change , 2014 .

[10]  Krishna P. Gummadi,et al.  Fairness Constraints: Mechanisms for Fair Classification , 2015, AISTATS.

[11]  Jun Sakuma,et al.  Fairness-Aware Classifier with Prejudice Remover Regularizer , 2012, ECML/PKDD.

[12]  Aaron Roth,et al.  Fairness in Learning: Classic and Contextual Bandits , 2016, NIPS.

[13]  Toon Calders,et al.  Classifying without discriminating , 2009, 2009 2nd International Conference on Computer, Control and Communication.

[14]  Toon Calders,et al.  Three naive Bayes approaches for discrimination-free classification , 2010, Data Mining and Knowledge Discovery.

[15]  Nathan Srebro,et al.  Equality of Opportunity in Supervised Learning , 2016, NIPS.

[16]  John Langford,et al.  A Reductions Approach to Fair Classification , 2018, ICML.

[17]  Toniann Pitassi,et al.  Learning Fair Representations , 2013, ICML.

[18]  Kedar Dhamdhere.  Approximating Additive Distortion of Embeddings into Line Metrics , 2004, APPROX-RANDOM.

[19]  Angela L. Duckworth,et al.  Personality Psychology and Economics , 2011, SSRN Electronic Journal.

[20]  Toniann Pitassi,et al.  Learning Adversarially Fair and Transferable Representations , 2018, ICML.

[21]  Salvatore Ruggieri,et al.  A multidisciplinary survey on discrimination analysis , 2013, The Knowledge Engineering Review.

[22]  Alexandra Chouldechova,et al.  Fair prediction with disparate impact: A study of bias in recidivism prediction instruments , 2016, Big Data.

[23]  Michael Carl Tschantz,et al.  Discriminative but Not Discriminatory: A Comparison of Fairness Definitions under Different Worldviews , 2018, ArXiv.

[24]  Michael Kearns,et al.  Efficient noise-tolerant learning from statistical queries , 1993, STOC.

[25]  L. Black.  Shame of the Nation: The Restoration of Apartheid Schooling in America , 2007 .

[26]  Krishna P. Gummadi,et al.  Fairness Constraints: A Mechanism for Fair Classification , 2015, ArXiv.

[27]  Carlos Eduardo Scheidegger,et al.  Certifying and Removing Disparate Impact , 2014, KDD.

[28]  Mikkel Thorup,et al.  On the approximability of numerical taxonomy (fitting distances by tree metrics) , 1996, SODA '96.

[29]  Jon M. Kleinberg,et al.  Inherent Trade-Offs in the Fair Determination of Risk Scores , 2016, ITCS.

[30]  Sampath Kannan,et al.  A robust model for finding optimal evolutionary trees , 1993, Algorithmica.

[31]  Andrew D. Selbst,et al.  Big Data's Disparate Impact , 2016 .

[32]  Krishna P. Gummadi,et al.  Fairness Beyond Disparate Treatment & Disparate Impact: Learning Classification without Disparate Mistreatment , 2016, WWW.

[33]  Toniann Pitassi,et al.  Fairness through awareness , 2011, ITCS '12.

[34]  Michael Carl Tschantz,et al.  Automated Experiments on Ad Privacy Settings: A Tale of Opacity, Choice, and Discrimination , 2014, ArXiv.

[35]  Teva J. Scheer.  Uniform Guidelines on Employee Selection Procedures , 2007 .

[36]  Indre Zliobaite,et al.  A survey on measuring indirect discrimination in machine learning , 2015, ArXiv.

[37]  Angela L. Duckworth,et al.  Grit: perseverance and passion for long-term goals. , 2007, Journal of personality and social psychology.