A Brief Summary of the AI Bill of Rights: A P&C View
The administration's framework, "The Blueprint for an AI Bill of Rights," has left everyone trying to digest and comprehend its impact. This framework is expected to have a significant effect on how companies adopt AI in the coming years. I'm sure the UK and EU will adopt similar positions soon.
The White House recently released a comprehensive framework called the Blueprint for an AI Bill of Rights outlining principles and best practices for the ethical development and use of artificial intelligence (AI) systems in both government and private sector contexts.
As insurers pursue AI technologies for areas like underwriting, claims, and customer service, companies need to closely review this guidance and consider how to align their own AI governance with the core principles and expectations within.
The Bill of Rights framework contains five overarching principles:
1. Safe and Effective Systems β AI should be vetted for safety, tested rigorously, and monitored continually once deployed to protect the public.
2. Algorithmic Discrimination Protections β AI systems should be proactively assessed for biases and harms to avoid discriminatory impacts on protected groups.
3. Data Privacy β Individuals' data privacy should be protected through minimal data collection, obtaining consent, allowing data access, and limiting surveillance.
4. Notice and Explanation β The public should be notified about AI use and provided explanations of AI-informed decisions and actions affecting them.
5. Human Alternatives, Consideration, and Fallback β Humans should remain involved in AI systems, with fallback and redress available for problematic or unsafe AI.
The Bill of Rights pays special attention to specific high-risk AI applications like those involving criminal justice, employment, education, and healthcare. Insurance is called a "sensitive context" where heightened expectations around data privacy and human oversight should apply.
Some of the significant relevant expectations outlined that insurers may want to address proactively include:
- Proactively testing AI models and data for fairness, accuracy, and representative performance across demographics to avoid algorithmic discrimination in pricing or underwriting decisions.
- Increased transparency through plainer language documentation of AI systems, models, and procedures made available to the public and regulators.
- Review and potential overhaul of data collection, particularly of sensitive data, to align with principles of data minimization, consent, and purpose limitation.
- Evaluation of automated customer-facing tools to ensure clear opt-out to a human alternative is available when appropriate and with minimal burden.
- Ongoing auditing and governance processes to regularly monitor our AI for issues and enable human intervention for high-risk applications.
Implementing the best practices related to artificial intelligence would necessitate investing in IT systems, data management, documentation, auditing, and workforce training. Doing so in advance would significantly enhance the public's confidence in the usage of AI in the insurance sector, decrease compliance and legal risks, and position firms as ethical AI pioneers in the insurance industry as relevant regulations ultimately come into play.
