'Un'Fair Machine Learning Algorithms
34 Pages Posted: 4 Jul 2019
Date Written: June 10, 2019
Machine Learning algorithms are getting widely deployed in real world decision making. These algorithms are required to have protected class (e.g., gender and race) blind decision rules to be legally compliant. Yet, in many cases, these algorithms are shown to be biased and feared to perpetuate structural inequalities found in the society. In response, several “fair” machine learning algorithms have been proposed recently that advocate changing the law to allow use of protected class specific decision rules to adjust for the societal inequalities. In this study, we show that these “fair” algorithms, while conceptually appealing, can make everyone worse-off including the very class they aim to protect. Compared to the current law, these “fair” algorithms limit the benefits from a more accurate algorithm for a firm. As a result, profit maximizing firms could under-invest in learning, i.e., improving the accuracy of their machine learning algorithms. We show that the investment in learning decreases when misclassification is costly which is exactly the case when greater accuracy is otherwise desired. Our paper highlights the importance of considering strategic behavior of stake holders when developing and evaluating “fair” machine learning algorithms. Overall, our results indicate that these “fair” algorithms, if turned into law, may not be able to deliver some of the anticipated benefits.
Keywords: algorithmic bias, fair algorithms, machine learning bias, equality of opportunity, algorithms and bias, machine learning, bias
JEL Classification: M30, M38, 03, K0, K2, D0
Suggested Citation: Suggested Citation