Dr. Berk Ustun: Towards Personalization without Harm
Every month, ETAI will invite a speaker to particpate in a monthly speaker series related to ethical issues in biomedical research. The monthly speaker series is part of our module's milestones relating to engagement and reflection on ethics with the AI-READI team.
This month, Dr. Ustun discussed how machine learning models, personalized with sensitive features like sex, age group, and HIV status, can perform better for populations but worse for specific groups, potentially causing harm. He proposed formal conditions for fair group attribute use and outlined practical methods, such as "personalization budgets" and "participatory systems," to ensure fairness in personalized predictions. You can find a recording to his talk here