Discrimination in Machine Learning

Link to module

Evaluated December 2021

This module, as described, “. . .analyzes the potential consequences of using location-based predictions to channel law enforcement efforts and resources. The module focuses on the case of PredPol to reveal the discriminatory potential of machine learning algorithms when employed in contexts that meet specific conditions. Based on this case analysis, students learn to discern between discriminatory intent (or disparate treatment) and discriminatory impact (or disparate impact). Through this discussion, students are prompted to reflect on how algorithms operate in specific social, political, and economic contexts, and how their potential for wrongful discrimination changes as a result of the context in which they are implemented.” This module is well-suited for a course in Machine Learning or Data Science.

This module covers material in Social Issues and Professional Practice. 

Instructors who adopt this module will need to identify significant background resources in both philosophical and sociological understandings of discrimination. Interdisciplinary support from faculty who teach philosophy, sociology, or cultural studies could be of benefit. Faculty need to consider how to engage in active connections with students to establish clarity related to wrongful discrimination, disparate discrimination, and indirect discrimination, as well as interpretations of institutional discrimination. The module presents as an isolated unit within a broader course structure. As presented, there is only one reading for students. Additional readings for students on discrimination, or pre-requisite lower division courses in ethics, ethnic and minority relations, gender, or social stratification, would provide a deeper level of understanding and will allow students to accomplish the synthesis across social-cultural-economic interpretations in a more substantive manner. Students who have experiences with discrimination will be more apt to perform well on this module. The “White Collar Crime” article can be found here: https://arxiv.org/ftp/arxiv/papers/1704/1704.07826.pdf. The module may work best as an assignment in an upper-division course in the CS curriculum, and since it does not cover standard CS material, time will need to be carved out to take up this important topic.

When considering assessment, instructors will need to examine the segment on “Lessons Learned” that presents some overall direction. Goals are presented outlining what students are to learn. An instructor can develop a rubric from the overall learning outcomes present in the module. 

The evaluation of this module was led by Colleen Greer and Marty J. Wolf as part of the Mozilla Foundation Responsible Computer Science Challenge. Patrick Anderson, Emanuelle Burton, Judy Goldsmith, Darakhshan Mir, Jaye Nias, and Evan Peck also made contributions. These works are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.