University of Pittsburgh’s School of Computing and Information: Developing Tech to Model and Manage Interconnected Systems on Campus and Beyond

U Pitt School Of Computing And Information Develops Innovative Tech

TL; DR: Established in 2017 as a multidisciplinary institution at the University of Pittsburgh, the School of Computing and Information is leading the charge in modeling interactions of systems and accelerating the build processes of those models. Under the guidance of Founding Dean Paul Cohen, the school has plans to solve some of today’s most challenging issues through the strategic use of data and IT. We recently sat down with Paul to learn how the school is transforming Pitt into an information-age university by exploring the real-life impacts of modeling and cultivating a professional society of diverse stakeholders.

A midwinter blizzard is hardly unheard of in Pittsburgh, Pennsylvania, but even the average snowstorm can set off a cascade of events that impacts the entire city. Icy, dangerous road conditions inevitably lead to accidents, causing significant traffic delays and taxing local emergency dispatch resources.

Paul Cohen had ample time to ponder this while stuck in a recent snowstorm-induced traffic jam that extended his usual commute from 20 minutes to more than an hour.

Inclement weather conditions or not, the behaviors of interconnected systems is a topic Paul thinks about quite often. As Founding Dean of the University of Pittsburgh’s School of Computing and Information, Paul’s job is to lead the university in researching technology that will help people model and manage complicated world systems.

“Everything is tightly connected, and it’s really difficult to predict the effects of shocks — even little shocks like a snowstorm,” Paul said.

Dean Paul Cohen's headshot and the U Pill School of Computing and Information logo

Dean Paul Cohen told us how the School of Computing and Information is dedicated to solving problems through IT.

Paul’s four-year stint at US Defense Advanced Research Projects Agency (DARPA), where he developed programs to improve modeling capabilities, prepared him to lead the new school.

“The world’s systems are tightly interconnected, and we don’t do a very good job of modeling them,” Paul said. “Our systems are becoming more intertwined and also more stressed by climate change, population growth, and other stressors, so modeling these effects is not merely an academic exercise.”

So, when Paul arrived at the University of Pittsburgh, his vision was clear — to start a school of computing and information that would serve to unite the modeling efforts of various campus disciplines through computing.

“The University of Pittsburgh was wise to start a school of computing that has a mission to really make this an information-age university,” Paul said. “And that means working with people in very disparate fields — from art and music to computational biology.”

The Aim: To Unite Disciplines and Form an Information-Age University

Paul has been helping model interactions of systems and accelerate the process of building those models since he arrived at the school in 2017.

“What we’re doing here in Pittsburgh has two distinctive features,” he said. “The first is that we are passionate about modeling not single systems, but interactions of systems.”

Paul used the example of just arriving at work. There’s a system involved in how you’ve traveled to the job site, and there are systems behind the electricity powering the office.

“And then there are microscopic systems,” Paul said. “There’s what you had for breakfast and how your microbiome is working away at it. You are a particle being buffeted by this dance of systems — just as I am, just as we all are. It’s the interaction of the systems that constitute life’s experiences — not just one system.”

Photo of the University of Pittsburgh seal

The school’s multidisciplinary approach seeks to accelerate modeling processes that will have positive, real-world impacts.

The second distinctive feature of the school’s work revolves around speeding up the modeling process. Paul told us scientists can build models of interacting systems; however, this is usually cost- and time-prohibitive.

“Following the example of my programs at DARPA, we are trying to accelerate the process of building very complicated models of interacting systems,” he said.

Paul said he hopes to convert what is typically a two- to five-year modeling process done by hand to a two- to five-week process assisted by technology developed at the school.

For example, the Graduate School of Public Health can model the opioid epidemic for the entire US population, but building the model was very expensive. Artificial intelligence techniques — some of which emerged from Paul’s DARPA programs — might accelerate the creation and modification of such sophisticated simulations.

“That’s the kind of work that we really resonate with,” he said. “And, when I say unify the campus, I don’t mean we’re going to tell everybody else what to do. I mean we’re going to provide all of the support we possibly can from a computing standpoint for people who are already quite advanced in their modeling efforts.”

Reconciling Prediction Performance with Comprehensible Models

When it comes to the prediction strength of current models, Paul said there’s a real conundrum in that the models we don’t understand do a better job than the ones we do.

Take Google’s artificial intelligence software AlphaGo, for example.

“In four hours playing against itself, AlphaGo can train itself to the point where it can beat the world champion in Go,” Paul said. “But the code that it learns is completely impenetrable — no human could read it.”

Therein lies the distinction between a model’s predictive strength and the human ability to understand it. Paul noted that AlphaGo is great when it comes to prediction, but there’s a catch.

“We just don’t understand what it’s doing,” he said. “It’s a very complicated nonlinear neural network system.”

That’s OK if you’re just playing a game, Paul told us, but “it’s not OK if your self-driving car is incomprehensible, and it’s not OK if your gun-toting robot is incomprehensible.”

Significant liability issues are inherent in these high-stakes cases.

“Who is responsible the first time a self-driving car kills somebody? It’s a real tricky situation, because you want good prediction performance, and the best prediction performance comes from statistical models,” Paul said. “You want comprehensible models, but statistical models generally aren’t comprehensible.”

Connecting Decision-Makers to Explore Real-Life Impacts of Modeling

When searching for answers to these complex ethical questions, there is power in numbers. In that way, Paul hopes bringing together a diverse group of high-level decision-makers with interests in modeling that will advance their individual missions.

“When I was working at DARPA, I found myself talking to many different kinds of people about this idea of modeling the whole world, and I was surprised by the diversity of the people I was talking to,” Paul said. “It slowly dawned on me that this need to model the world is felt by just about everyone who has anything to say about policy.”

Collage of faculty and students at U Pitt's School of Computing and Information

The school’s success in answering complex questions about tech is due, in large part, to its diverse community of thinkers.

To introduce those individuals to one another, the School of Computing and Information is hosting a conference, sponsored in part by the social justice agency Ford Foundation, on May 21 in Pittsburgh. Paul said the conference brings together a wide range of high-level decision-makers with interests in modeling. The event gives them a chance to network and collaborate.

“[These decision-makers generally would not talk to each other,” Paul said. “They don’t know that they’re a community.”

The event will make it clear that modeling is possible and worth investing in.

“We’re going to make Pittsburgh this great big center for ‘modeling the world’ with lots of friends from lots of different sectors helping us do it,” he said.

Building a Professional Society of Diverse Stakeholders to Enact Change

In the coming year, Paul hopes to establish a professional society of stakeholders in modeling who will collaborate to solve specific challenges at annual conferences.

“Every year, there will be a challenge issued to the modeling community, saying ‘Here, you think you’re good? Try this one,’” Paul said. “And with a bit of luck, we’ll be able to raise money in the year, so when we announce the new challenge problem we can put some dollars behind it.”

Within five years, Paul would like the society to become an authoritative voice in modeling. He noted that the society has no commercial interest and is unbiased; therefore, stakeholders can converse without the pressure of competition and possibility of disclosing proprietary information.

“It’s a pre-competitive sort of environment because we’re a university, and we don’t have a horse in the race,” he said. “We just want to do the modeling right.”

Ultimately, Paul told us his goal is to model the world through advanced technology.

“For a long time, people assumed that you just can’t model the world,” he said. “What I’m trying to convince people — and the work at DARPA helped immensely — is that, yes, you can model the world; it’s just really expensive, so let’s get the technology to help us.”