By MARTY LEVINE
“Models and data-fueled applications are everywhere, and that is exactly where data science skills and applications are needed — everywhere,” says Associate Vice Provost for Data Science Michael Colaresi, who heads up the provost office’s new Responsible Data Science (RDS@Pitt) initiative.
In a brief video explaining the need for the effort, he cites the “harms of gaps in available data” and the “convenient but distorting modeling assumptions” that have “real costs.”
“Too many data science programs,” he says, “acquaint students with the context of data only from introductory example datasets, but leave them blind to the ways that data and modeling can be used to lead but also mislead. … At RDS@Pitt we broaden the focus to responsible uses and applications of the ever-evolving tools of data science.”
RDS@Pitt, Colaresi says, aims to look at both “where data comes from and who decides?” and “who is influenced in what ways?”
The initiative has now formed its first three communities of practice, focused on the data science of addiction, on sustainable practices and on equity and access in data science.
The last group will look at whether using data science is worthwhile for particular issues. Data science’s emerging technologies can be very energy inefficient, Colaresi notes, “so you don’t necessarily want to burn down half of a forest answering silly questions.”
The data science of addiction group will be working with the School of Pharmacy and other schools here, looking at the issue from molecular to societal levels. “You don’t necessarily want to develop a drug that no one is going to use,” Colaresi says. “You don’t want to ... chase ghosts.”
Responsible data science encompasses responsible artificial intelligence technologies (AI) as well, says Kendra Oliver, managing director for Responsibility Data Science at Pitt. The new program will also include an RDS@Pitt scholars program, with 16 Pitt students who have “wildly diverse training,” she says — from neuroscience, economics and urban planning to criminal justice, psychology and environmental studies. They will each develop a “framework” focused on Pitt’s student data use, for instance, or on health-care privacy issues, and share the results during a campus Data Science Day.
“The world is a messy place,” points out Colaresi, and the value of RDS@Pitt is in the discussions it prompts, and in encouraging people to build its values into their applications ahead of their use. “We have to be very intentional that, just because it’s in code, it isn’t necessarily very clean,” he says: Just because some of the results can be expressed in mathematical expressions, that doesn’t mean those results are unbiased.
“There are incredibly exciting things we can do” with AI and other technologies, Colaresi adds. “But it seems like you need to give up your values to use these,” faculty and students are too often concluding. These technologies lack “a larger human understanding” of the problems they are being used to examine, and of the solutions they suggest. Reliable data science “really does help people use these technologies.”
“People … just feel like things are changing so fast,” he says. “It shakes their foundation of what they can do. ... We need to shift how the University engages to make sure we are keeping up,” changing parts of the curriculum to reflect new thinking in responsible data science much more often than annually.
This is hardly just an internal Pitt issue, he adds, particularly because many places outside the University team with Pitt and rely on expertise here to collect and use data in a smart, useful and responsible manner.
Pittsburgh already employs Pitt resources at the University Center for Social & Urban Research, for instance, in the city’s quest to forecast tax revenue accurately. The University has helped city officials to build different predictive models, Oliver says. When people ask why the forecast went up or down, “they have to have an answer for it,” Colaresi adds. RDS@Pitt should help to make more such connections between campus and community.
“We hear over and over again that, because data science is all over the University, people don’t know where to go, especially from the outside — that we need to have a welcoming front door,” he says. “People want to learn from our staff and what they are doing.”
Marty Levine is a staff writer for the University Times. Reach him at martyl@pitt.edu or 412-758-4859.
Have a story idea or news to share? Share it with the University Times.