What's Happening?
An issue brief from the AI, Data Science, and Analytics Committee of Actuary.org, titled 'Artificial Intelligence Governance: What Counts as AI?', explores the classification of predictive models, particularly Generalized Linear Models (GLMs) used in insurance,
under emerging state AI governance frameworks. The brief highlights that current technical definitions have not adequately resolved whether these models should be considered AI. This ambiguity leads to inconsistent reporting, increased costs for businesses, and the risk that novel AI tools might be overlooked or miscategorized alongside older technologies. The committee proposes a risk-based approach to AI oversight, where the level of scrutiny scales with characteristics such as learned behavior, limited explainability, consumer-facing use, reduced human oversight, and post-deployment adaptation.
Why It's Important?
This discussion is crucial for the U.S. insurance industry and regulatory bodies. Inconsistent definitions of AI can create a fragmented regulatory landscape, hindering innovation while failing to adequately address potential risks. If GLMs, which have been used for decades, are classified as AI under new frameworks, it could impose unnecessary compliance burdens and costs on insurers without providing additional consumer protection. Conversely, if truly novel AI applications are not properly identified and regulated, they could introduce new risks related to bias, transparency, and accountability. A clear, risk-based approach is essential to foster responsible AI development, ensure fair consumer outcomes, and maintain a competitive and stable insurance market.
What's Next?
The proposed risk-based approach suggests that future AI governance frameworks in the U.S. will likely move beyond rigid technical definitions towards a more nuanced assessment of AI systems. This could lead to the development of tiered regulatory requirements, where AI applications with higher potential for learned behavior, limited explainability, or direct consumer impact face stricter oversight. Industry stakeholders, including actuaries and insurance companies, will need to actively engage with policymakers to shape these evolving frameworks. The goal will be to create regulations that are adaptable, promote innovation, and effectively manage the unique risks associated with advanced AI technologies.
Beyond the Headlines
The debate over 'what counts as AI' extends beyond the insurance sector, touching upon broader legal, ethical, and economic implications across various industries. The challenge of defining AI for regulatory purposes highlights the rapid pace of technological advancement outpacing traditional legislative processes. This ambiguity can lead to 'regulatory arbitrage,' where companies exploit definitional loopholes, or stifle innovation due to overly broad regulations. A risk-based approach, as suggested by Actuary.org, could serve as a model for other sectors grappling with AI governance, emphasizing proportionality and focusing regulatory efforts where the potential for harm or societal impact is greatest. This also underscores the need for interdisciplinary collaboration between technologists, ethicists, lawyers, and policymakers to create effective and future-proof AI governance strategies.
