Formal fairness constraints often “worsen” outcomes. No coherent account of fairness is possible without applying some normative perspective.1
In general, you need to collect information on protected characteristics to avoid discriminating on that basis.2 People often find this counterintuitive.
Even after car insurance reform, Detroit drivers have higher rates
Workday AI Bias Case Tests EEOC Definition of Employment Agency
UnitedHealth sued over use of algorithm in Medicare Advantage plans
AI Chatbots Are The New Job Interviewers
Huskey v. State Farm: Class action alleging discrimination by State Farm resulting from machine learning systems disproportionately selecting claims from Black policyholders for extra screening
Alex Chohlas-Wood et al., “Designing Equitable Algorithms,” Nature Computational Science 3, no. 7, 7 (July 2023): 601–10, https://doi.org/10.1038/s43588-023-00485-4.
M. Lindholm et al., “Discrimination-Free Insurance Pricing,” Astin Bulletin: The Journal of the Iaa 52, no. 1 (January 2022): 55–89, https://doi.org/10.1017/asb.2021.23.