Investigators: CDEP Affiliate Daniel Björkegren, Joshua Blumenstock
How can machine learning systems account for society's preferences? We explore training machine learning methods to balance multiple objectives. We have developed a theoretical framework which accounts for the fact that welfare objectives tend to be measured with less precision than business metrics like profit. We are developing an application to 'welfare' credit scores based on a digital credit experiment in Nigeria.
Balancing Competing Objectives with Noisy Data: Score-Based Classifiers for Welfare-Aware Machine Learning
International Conference on Machine Learning (ICML) (2020)
