By Kevin Manne

BUFFALO, N.Y. —  As generative AI becomes more autonomous and deeply integrated into society, researchers need to evaluate who benefits, who may be harmed and how systems can be optimized for social good, according to research from the latest issue of the INFORMS Journal on Computing, co-edited by Ram Ramesh, PhD, an area editor of the journal and professor of management science and systems in the University at Buffalo School of Management.

Through 13 studies, the special issue, titled Responsible AI and Data Science for Social Good, examines how responsible AI and data science can create meaningful societal impact across six areas: judicial systems, education, communication, healthcare, bias and fairness, and interpretability.

“Responsible AI cannot be achieved by addressing ethical consequences after the systems are built,” says Ramesh. “Social goals must be incorporated throughout the AI lifecycle, from deciding what data to collect and how models learn, to determining what they optimize, how they interact with people and institutions, and how they are governed.” 

The researchers say AI systems do far more than generate predictions. They can allocate resources, shape opportunities, influence institutional processes and ultimately affect long-term societal outcomes.

As one example from the healthcare field published in this issue, recent research found a machine-learning appointment scheduling system had Black patients waiting approximately 30% longer than non-Black patients. Researchers were able to eliminate that disparity by incorporating race into the system’s optimization objective while maintaining scheduling efficiency.

Bias can also develop in less obvious ways, the researchers say. Historical inequities embedded in data, incomplete feedback and interactions among different institutions can all produce disparities. And, because algorithms can influence the data generated by future decisions, those disparities can compound over time.

“Fairness should not be viewed as an external constraint imposed on otherwise optimal systems but as a core design objective,” says Ramesh.

Ramesh jointly edited the journal issue with Kaushik Dutta from the University of South Florida Muma College of Business; Ajay Kumar from Emlyon Business School; Zhiling Guo from the University of North Texas G. Brint Ryan College of Business; Martin Bichler from the Technical University of Munich School of Computation, Information and Technology; Dursun Delen from Oklahoma State University Spears School of Business; and Paul Brooks from Virginia Commonwealth University School of Industrial and Systems Engineering.