The European Union's AI Act has been a topic of significant discussion, especially in sectors that heavily rely on artificial intelligence (AI) for their operations. One such sector is the insurance industry. Views of the European Insurance and Occupational Pensions Authority (EIOPA) were conveyed to the EU commission via email that is publicly available.
I attempt to summarise the concerns of the insurance industry conveyed in this email.
1. Supervision and Regulation of AI in Insurance
The insurance sector believes that insurance legislation and both European and National Supervisory Authorities should remain competent in overseeing and regulating the use of AI. The European Insurance and Occupational Pensions Authority (EIOPA) has expressed its support for this stance. They emphasize that sectorial insurance legislation and the European and national supervisory authorities will continue to oversee AI-related matters in the financial services sector, all while respecting the AI Act's framework.
2. The Unique Role of AI in Insurance
The insurance sector has unique AI applications that differ from other industries. For instance, actuaries and other key function holders in insurance companies play a crucial human oversight role in AI systems used for pricing and underwriting. This kind of oversight doesn't exist in other sectors. Moreover, certain datasets, like a customer's age or disabilities, which might be restricted in other sectors, are permissible for insurance underwriting. This underscores the need for AI applications in insurance to be governed by sector-specific legislation.
3. Collaboration between European Supervisory Authorities and the European Artificial Intelligence Board
For the AI Act to be effective, there needs to be a strong collaboration between the European Supervisory Authorities (ESAs) and the European Artificial Intelligence Board (EAIB). This partnership will ensure that the AI Act's requirements are tailored to the specific needs of the insurance sector. The AI Act should also recognize a more significant role for the ESAs and National Competent Authorities (NCAs) in defining high-risk AI applications criteria.
4. Regulatory Consistency and Avoiding Overlaps
While the Solvency II Directive and the Insurance Distribution Directive (IDD) already have provisions for risk management, governance, and transparency, they didn't consider the unique challenges posed by AI. EIOPA has taken steps to address this gap by creating a stakeholder group on digital ethics in insurance. This group developed an AI governance principles report in 2021, offering non-binding guidance on ethical AI use in insurance.
However, if insurance AI applications were to be classified as high-risk, there could be overlaps with existing provisions, leading to potential inconsistencies. To avoid this, the AI Act should reference insurance legislation, similar to what's done for the banking sector.
5. The Principle of Proportionality
EIOPA strongly believes that the regulation and supervision of AI in insurance should be sector-specific. They have proposed several proportionality considerations:
Definition of AI: The current definition of AI in the AI Act is broad and includes traditional mathematical models. EIOPA suggests a narrower definition that focuses on AI systems with distinctive features, such as machine learning and deep learning.
Relevance of Business Lines: Not all insurance lines are equally relevant to consumers. Only those that impact fundamental rights should have specific requirements.
Claims Management: There's a need to define the concept of claims management more narrowly, considering its impact on fundamental rights.
Impact on SMEs/Start-ups: The AI Act should be adapted to the nature, scale, and complexity of SMEs and start-ups. They should receive the same treatment across sectors.
In conclusion, the insurance industry's reaction to the EU AI Act underscores the need for sector-specific regulations that consider the unique challenges and applications of AI in insurance. Collaboration, consistency, and proportionality will be key to ensuring that the AI Act benefits both the industry and its consumers.