Description: An overview of bias and fairness in machine learning models, as they apply to use cases in life insurance.

Description: As the use of predictive models and novel datasets becomes more prevalent, a focus on consumer protection is being highlighted by regulators, ethicists, and industry insiders. Over time, increased rigor should and will be applied in model development and implementation. These issues are not new. Hundreds of individuals have worked on these issues in the laboratory, the board room, and the court room. The Civil Rights Act of 1964, Uniform Guidelines of the US EEOC, the Principles for the Validation and Use of Personnel Selection and other sources are intended to protect the rights of citizens and consumers and provide guidance to professionals in the development and use of decision-making tools. Drawing heavily on examples and parallels from psychology and employment practices as well as other disciplines, concepts such as model bias, fairness, adverse impact, disparate treatment and the like will be presented. Suggestions to ensure model validity, demonstrate model and data relevancy and minimize the potential for adverse impact to protected classes will also be offered. Each of these considerations will be linked to recent dialogue and concerns in the Life Insurance sector.