NJIT Professor Receives NSF Grants Totaling $2.8M to Design Solutions for Responsible AI

If you're a human resource specialist grappling with AI-based decision systems for hiring good candidates, or a data scientist trying to develop transparent and accountable ranking algorithms for decision-making in critical socio-technical contexts, you will likely benefit from the research of Aritra Dasgupta, an assistant professor in the Department of Data Science at NJIT's Ying Wu College of Computing.

Dasgupta is the recipient of two collaborative National Science Foundation (NSF) grants totaling approximately $2.8 million (out of which NJIT’s funding is $860,000) from the Information and Intelligent Systems (IIS) and the Future of Work (FW-HTF) programs.

The IIS grant ($400K) will fund the development of human-in-the-loop techniques for data scientists to transparently design and validate algorithmic rankers in critical areas like educational program admissions, hiring and institutional rankings. The FW-HTF grant is for an interpretability-by-design framework named Trapeze ($460K) that helps HR specialists in their meticulous balancing act of talent acquisition.

The overarching theme of both these projects will be designing and scrutinizing responsible AI solutions, according to Dasgupta, who directs NJIT's Intelligible Information Visualization Lab. He explained that “being responsible here means having agency and control over different stages of data-driven model design, allowing data scientists or decision-makers to gain confidence that the models can be safely applied in critical socio-technical contexts.”

Dasgupta also emphasized the highly collaborative nature of the projects, noting that “most complex global challenges nowadays require innovations that can only stem from the amalgamation of diverse expertise. This is reflected in our project teams.”

The assistant professor is collaborating with Associate Professor Julia Stoyanovich, who directs the Center for Responsible AI at New York University, on both projects. Other collaborators include experts in databases (University of Michigan Professor H.V. Jagadish, on the IIS project) and in cognitive and organizational psychology (Rice University Professor Fred Oswald and Michigan State University Professor Ann Marie Ryan, on the Trapeze project).

Dasgupta’s expertise in developing interactive visualization techniques is at the core of the technological interventions planned for both projects. He reflected on the role of visualization using an analogy: “We all are proud of NJIT’s recent improvements in rankings published by The Wall Street Journal and the U.S. News, primarily because of corrections in methodology focusing more on student outcomes. Now, imagine if, instead of institutions, individuals, like job candidates, are ranked, and because of problematic methodology, candidates from under-represented groups are unfairly demoted.

“Our methods will help proactively visualize the sensitivity of rankings to slight changes in the input data or the scoring formula, the complex structures with the data, and help balance candidate quality, fairness, diversity, et cetera,” he added.

According to Dasgupta, poorly designed AI models can produce incorrect and inconsistent results that fail to match candidates appropriately to job requirements, or that limit the visibility of well-suited candidates. HR personnel who manage talent acquisition must meticulously sift through large pools of candidates to find people who not only meet the job requirements but are the “right” cultural fit and face a difficult balancing act.

“This is why we named our solution Trapeze,” he said, with a smile.

The ideas proposed in this project are based on work that he published together with a doctoral student, Jun Yuan. “Jun consistently brings innovative ideas during our
brainstorming sessions, and he has developed a happy knack of rigorously pursuing them to fruition, which is great to see, as his advisor,” said Dasgupta. Yuan is Dasgupta’s first
doctoral student and a Ph.D. candidate in YWCC’s newly launched Ph.D. in data science program. Dasgupta is seeking highly motivated doctoral student aspirants to join his group and work on these cutting-edge problems.