How to BEAT unintended bias in algorithmic-based personalized policies?

What is it about?

Decision makers use algorithmic personalization in many domains. An inherent risk of personalization is unintentionally discriminating certain groups. We propose a solution to this problem which we call BEAT (bias-eliminating adapted trees). Key insights from our work: 1. Existing solutions that recommend removing protected characteristics (such as race, gender, age) from the data and relying on other attributes may exacerbate the bias. 2. Our solution, BEAT (Bias-Eliminating Adapted Trees), guarantees both group and individual fairness, while still leveraging the value of personalization. 3. BEAT is easy to implement, scalable, and practical (thanks to GRF). 4. BEAT can be used to evaluate the tradeoff between fairness and efficiency in the algorithm’s output (e.g. profits, access to resources, engagement). 5. BEAT is applicable to many personalization problems, and can be used for prediction-based resource allocation, such as loans and credit card access. 6. BEAT can also be used for interventions, such as advertising, CRM, and medical treatments.

Featured Image

Photo by Michael Dziedzic on Unsplash

Photo by Michael Dziedzic on Unsplash

Why is it important?

Decision makers use algorithmic personalization for resource allocation decisions in many domains (e.g., medical treatments, hiring decisions, product recommendations, or dynamic pricing). An inherent risk of personalization is disproportionate targeting of individuals from certain protected groups. Existing solutions that firms use to avoid this bias often do not eliminate the bias, and may even exacerbate it. We propose a solution that ensures balanced allocation of resources across individuals (guaranteeing both group and individual fairness) while still leveraging the value of personalization.