One health-care scheduling gap closed by adding race to its objective

An appointment queue can hide the decision that shapes it. In one machine-learning scheduling system in health care, Black patients waited approximately 30% longer than non-Black patients. Researchers eliminated the disparity by putting race into the system’s optimization objective while maintaining scheduling efficiency.

That finding shows why AI systems are not only prediction engines. They can allocate resources, shape opportunities and influence institutional processes. When historical inequities are embedded in data, feedback is incomplete or institutions interact in reinforcing ways, disparities can emerge less visibly and compound over time.

The example comes from one of 13 studies in a special issue of the INFORMS Journal on Computing focused on responsible AI and data science for social good. The issue spans six areas: judicial systems, education, communication, health care, bias and fairness, and interpretability. Ram Ramesh, a professor at the University at Buffalo School of Management and an area editor of the journal, says ethical consequences cannot be handled only after systems are built; social goals have to shape data collection, learning, optimization, interaction and governance.

Ramesh describes fairness as a design objective rather than an external constraint added to an otherwise optimal system. The appointment result is a narrow but useful demonstration: changing what an algorithm optimizes altered a measured disparity without reducing the reported scheduling efficiency. It does not settle the broader problems created by biased historical data or incomplete feedback.

Concretely, teams building AI for scheduling or resource allocation can treat fairness as part of the technical brief from the beginning, alongside the data and performance objectives. For people affected by those systems, that choice can show up as a measurable difference in waiting time. The reported result offers a specific route to improvement while leaving the wider task—finding and preventing less obvious disparities—firmly on the engineering table.