Earlier this week three US Senators introduced a new bill aimed at requiring large companies to test algorithms used in consumer decisions and correct any outcomes that are “inaccurate, unfair, biased or discriminatory”. The Algorithmic Accountability Act would apply to companies with annual revenue over $50M and that have data for over one million people or devices, and has provisions for protection of that data. The Act also calls for the FTC to create further regulation that would require impact studies for certain automated systems.
As with other emerging technologies, regulators have taken an interest as stories emerge about bias or discrimination in pricing or eligibility for consumer products. In many cases these early attempts at regulation are broad and require further definition. The Algorithmic Accountability Act follows the same trends, and questions. For example, the Act appears to focus on bias in the algorithms and not on how biased data can affect the results of algorithms.
Deployment of algorithms for AI, machine learning, image recognition and other areas are at the beginning stages and in limited deployment at insurance carriers (see our Emerging Tech report). Most use cases are not related to rating and pricing, but there is interest in using algorithmic technologies to improve marketing and underwriting processes, and insurers, perhaps more than in other industries, are fully aware of the regulatory consequences of improper use of algorithms and associated data.
Insurers should continue to monitor potential legislation like the Algorithmic Accountability Act, but also include in their pilot projects provisions for documenting, testing and evaluating the algorithms and data deployed for the use cases they are considering.
If you’d like to discuss how insurers and solution providers are approaching technologies that use algorithms, please contact me.
Add new comment