By Josh Hershman
Bias continues to be high on regulators’ agenda. This was the takeaway from the 2022 NAIC Summer National Meeting which just wrapped up on August 13th in Portland Oregon.
The NAIC and its members are laser focused on algorithmic bias and modeling bias through big data, algorithms, ai/ml utilization through every part of the insurance life cycle, including marketing, underwriting, pricing, claims and fraud. The key takeaways were:
- Regulators are focused on fairness more than ever before.
- Fairness is not an issue that is going away.
- It’s also a complicated issue that is driven by each individual jurisdiction.
- Some industry insiders and regulators believe regulators already have the tools through ERM or other actuarial principles to address these issues, while others think regulators are not equipped to manage the growing complexity and spectrum of fairness considerations.
Rest assured these conversations will be ongoing, but the solutions and how they will be implemented are still being determined.
To get a model act through the NAIC takes a significant amount of time and energy. Meanwhile, algorithmic bias may be a big enough issue that some jurisdictions move quicker than the NAIC. Already, New York DFS issued Circular Letter No. 1 on January 18, 2019. Colorado, meanwhile, passed SB169 in 2021, and Connecticut issued a notice in 2021 and then updated it requiring a certification from industry in April of 2022. California followed with their own bulletin.
Each of these jurisdictions have attempted to address the algorithmic bias issue differently. We will see more of this approach over the coming months while the NAIC works through the issues as well. Once the direction of the NAIC is clear then the jurisdictions that did act will likely update their processes to meet the model act, especially if the model act becomes a part of the NAIC accreditation.
The 2022 NAIC Summer National Meeting offered some solutions, including possibly requiring the industry to use an auditing or validating process before it uses AI/ML models, or even adding labels to models similar to food nutrition labels. It will be interesting to see which jurisdiction moves next and what that next move looks like.
For some jurisdictions a National solution might not come quick enough. At some point insurance companies are going to be subject to a fairness regime, one way or another.