Vector Researcher Develops Fairness Model Accounting for Individual Preferences
Researchers at the Vector Institute for Artificial Intelligence have developed a new machine learning fairness model that incorporates individual preferences, addressing limitations in traditional binary fairness definitions. Led by Safwan Hossain and supervised by Nisarg Shah, the team adapted economic concepts of envy freeness and equitability to create a more nuanced algorithmic framework. Unlike standard models that often overlook how different individuals value outcomes, this approach recognizes that fairness is subjective and dependent on personal satisfaction. For instance, receiving an unwanted loan type can be as unfair as being denied a loan entirely. The resulting model, detailed in the paper 'Designing Fairly Fair Classifiers Via Economic Fairness Notions,' is generalizable and applicable to various sectors, including targeted advertising and healthcare. By integrating these economic fairness notions, the model aims to reduce bias in data-driven decisions where individual preference plays a critical role. Hossain is currently extending this research to the health sector, emphasizing that personalized care requires algorithms that align with patient desires. This development marks a significant step toward creating AI systems that are not only statistically fair but also perceived as fair by the individuals they serve.
Wire timeline
Vector Researcher Develops Fairness Model Accounting for Individual Preferences
Researchers at the Vector Institute for Artificial Intelligence have developed a new machine learning fairness model that incorporates individual preferences, addressing limitations in traditional binary fairness definitions. Led by Safwan Hossain and supervised by Nisarg Shah, the team adapted economic concepts of envy freeness and equitability to create a more nuanced algorithmic framework. Unlike standard models that often overlook how different individuals value outcomes, this approach recognizes that fairness is subjective and dependent on personal satisfaction. For instance, receiving an unwanted loan type can be as unfair as being denied a loan entirely. The resulting model, detailed in the paper 'Designing Fairly Fair Classifiers Via Economic Fairness Notions,' is generalizable and applicable to various sectors, including targeted advertising and healthcare. By integrating these economic fairness notions, the model aims to reduce bias in data-driven decisions where individual preference plays a critical role. Hossain is currently extending this research to the health sector, emphasizing that personalized care requires algorithms that align with patient desires. This development marks a significant step toward creating AI systems that are not only statistically fair but also perceived as fair by the individuals they serve.
Vector Institute for Artificial Intelligence