What is Unfair Discrimination?

“Unfair discrimination” is cited as the major concern across each insurance AI regulation released to date. Perhaps the thorniest question, as this regulation comes into practice, is: what exactly is “unfair discrimination”?

We tip our hat to a piece by Kudakwashe Chibanda for the Casualty Actuarial Society, "Defining Discrimination in Insurance", published in 2022 as part of CAS’s research series on race and insurance pricing, for its deep dive chronicling the evolving definition of unfair discrimination and related terms.

A key concept from Chibanda’s paper is the dichotomy between the traditional definition of unfair discrimination, which emphasizes a consistent relationship between inputs and expected costs, versus some of the more recent definitions that emphasize impact on protected classes, usually using language like “disproportionate impact.”

One major implication of the latter definition is that, in order to consider impact on protected classes, insurers must classify their customers and calculate outcomes by class. These sorts of analyses represent new horizons for most carriers, as protected class information, such as race or sexual orientation, is generally not available and may even be illegal for insurers to collect. 

Furthermore, for regulation that includes both standards, a critical unanswered question is how to balance situations where they may be in conflict. It may be impossible to treat every individual entirely fairly while simultaneously ensuring comparable protected class outcomes, such as in cases where protected class populations differ by various social indicators.

With that context, let’s review the definitions of unfair discrimination and related terms currently on the table, including our read of whether each regulation focuses more on Inputs (i.e., the relationship between inputs and expected costs) or Impact (i.e., impact on protected classes):

Ultimately, understanding what unfair discrimination means for your stakeholders will drive what testing approaches make most sense.

Reference:

• NAIC Model Bulletin

• Colorado SB21-169

• New York Proposed Insurance Circular Letter

• California Bulletin 2022-5