Insurers have several ways to ensure compliance with new regulation on their use of AI, according to insurance compliance experts.
More
Insurers' options include forming governance committees to respond, further experimentation with AI technology, and making efforts to balance technological evolution, democratization of AI and fairness concerns. Insurers also need to be aware of how they use AI for pricing and marketing, watching for potential bias.
Speaking in a recent Insurtech Insights webcast, Parul Kaul-Green, founder and CEO of Eudaimon Consulting and previously an executive at Liberty Mutual and AXA XL, recalled creating an AI governance committee from senior leadership to oversee AI strategy, compliance and risk management.
"The chief actuary and CIO were working together," she said. NAIC's guidance "specifically calls for senior management accountability and board level oversight, so that ring fenced that seniority and the seriousness of this particular governance committee."
Kaul-Green recommends:
- Implementing risk-based controls to mitigate unfair practices, discrimination and inaccurate decisions. "These controls should be proportional, and harm prevention should be instituted through these," she said.
- Using new "chain of thought" models to provide transparency and explainability. "This needs to be ensured in AI decision making, so that consumers and regulators understand how AI impacts underwriting, pricing, claims and other insurance functions where it's being implemented," she said.
- Maintaining ongoing monitoring and validation. "Models drift. They are not perpetual. They are self learning. You need a process of ongoing monitoring and validation to detect biases, errors or hallucination that could lead to unfair outcomes," she said.
Regulation can lead companies to "stop experimenting," Kaul-Green added. "I would urge insurers not to do that, because there are immense pain points in our industry. There is a massive opportunity to gain efficiencies."
Insurers should watch for potential bias and examine their use of AI for pricing, to comply with regulation, according to J.P. Wieske, co-founder and principal, American InsurTech Council, who spoke in a recent Insurtech Association webcast. Wieske is also vice president of state affairs at Horizon Government Affairs, a lobbying and consulting firm.
"Regulators look at these issues from a consistency standpoint, certainly concerned about factors that lead into pricing," Wieske said. Regulators are concerned about whether insurers understand the pricing processes performed by AI, he added.
The potential of AI to create bias against serving certain customer segments is also a regulatory concern, according to Scott Harrison, co-founder of the American InsurTech Council. This bias can appear in AI-generated pricing models, he added. "They can use technology to engage in what they refer to as discrimination by proxy," Harrison said. "They find other ways to either weed out people or populations they don't want to do business with."
To check for such biases, regulators have to pull the models apart and have data scientists look at their testing and design, Harrison and Wieske noted. "They want to know that you're going through a process, validating results, validating its use as well as looking at whether there's discriminatory outcomes or an outcome that will lead to an insurance company becoming insolvent over time," Wieske said.
