Open AccessEliminating unintended bias in personalized policies using bias-eliminating adapted trees (BEAT)
Author(s) -
Eva Ascarza,
Ayelet Israeli
Publication year - 2022
Publication title -
proceedings of the national academy of sciences of the united states of america
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 5.011
H-Index - 771
eISSN - 1091-6490
pISSN - 0027-8424
DOI - 10.1073/pnas.2115293119
Subject(s) - personalization , scalability , computer science , unintended consequences , beat (acoustics) , risk analysis (engineering) , world wide web , business , database , physics , political science , acoustics , law
Significance Decision makers now 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 BEAT (bias-eliminating adapted trees) to ensure balanced allocation of resources across individuals—guaranteeing both group and individual fairness—while still leveraging the value of personalization. We validate our method using simulations as well as an online experiment withN = 3,146 participants. BEAT is easy to implement in practice, has desirable scalability properties, and is applicable to many personalization problems.