Project Overview
- Client/Partner: Forsikring & Pension, Algorithms, Data & Democracy
- Date: August 2024 – Oktober 2024
- Location: Copenhagen, Denmark
- Industry: Insurance Sector
- Service: AI Governance
Challenge
Insurers in Europe are faced with increasing regulatory and societal pressures to adopt innovative AI-based solutions to remain competitive. However, given the long history of social controversies, AI technologies need to be ethically sound. Challenges included to balance transparency with data privacy, avoiding bias and discrimination in AI models, and meeting the expectations of multiple stakeholders (clients, companies, society). The complexity of AI adoption required a structured way to manage ethical dilemmas that arise in real-world AI use cases.
Methodology
- Approach: The project utilized a workshop-based methodology involving technical experts from actuarial science, data science, and compliance. This was followed by the development of a tool to map ethical trade-offs in AI systems.
- Tools Used: Lectures, workshops and the combination of previous governance frameowkrs, academic literature and deep sector expertise.
- Collaboration: The project was conducted in collaboration with the Danish trade associations for insurance companies and pension funds, F&P
Solution
The ethical-dillemma mapping tool enables insurance companies to assess the ethical trade-offs of their AI systems. The tool focuses on three core areas:
- Procedural Fairness & Impact: Evaluates how AI systems align with fairness for customers, companies, and society.
- Data Collection: Assesses the ethical implications of the types, amounts, and frequency of data collection.
- Modelling Decisions: Helps insurers navigate trade-offs between AI model accuracy and transparency, ensuring that sensitive AI use cases are developed in a way that balances these competing priorities.
Together, choices made for fairness, data collection and modelling decisions determine the overall governance impact.