Keeping an Eye on AI: Pitt Engineer Examines the Future of Intelligent Machines

Ervin Sejdic has kept a close eye on the rapidly evolving technology advancing artificial intelligence in machines — an easy task, he said, with Pittsburgh a national hub for this kind of research, whether in health sciences or self-driving vehicles.

Sejdic, an associate professor of electrical and computing engineering in the University of Pittsburgh’s Swanson School of Engineering, is leading research in medicine using machine learning, a branch of artificial intelligence where computing systems can make decisions with minimal human intervention by identifying patterns and interpreting data.

For instance, his team in Pitt’s iMed Lab has created a number of algorithms aimed to help people with swallowing problems. Some of these algorithms automatically detect and track the hyoid bone, a U-shaped bone in the human throat, in X-ray images in about 20 seconds, a task that would take a human 20 to 30 minutes. If the bone is dislocated, patients could have problems with swallowing, talking or breathing. Faster detection could help with better efficiency in identifying best practices in swallowing therapies.

“This could in turn prevent further complications such as malnutrition or pneumonia,” said Sejdic, who became interested in artificial intelligence during his search for graduate schools.

His team is also studying patients’ walking motions, with the goal of having wearable monitoring devices and advanced algorithms predict any instabilities during walking.

“This is a very important problem especially for older adults, as falls and fall-related injuries are one of the major health care costs in the United States,” Sejdic said.

The emergence of autonomous vehicles

Before these two recent studies in AI applications in medicine, Sejdic led testing for the Innova Dash initiative at Pitt in 2014. The research involved a tiny electric car that collected enormous amounts of data.

At that time, Pitt was one of four campuses nationwide where researchers tested four University Electric Vehicles in an effort to reduce campus carbon footprint by collecting massive amounts of data from the cars’ various sensors. The vehicles — not to be confused with autonomous or self-driving vehicles — were connected through WiFi to researchers, students and each other, with data stored in an AI cloud. Data consisted of information regarding the terrain, vehicle speed, hill inclines and declines and energy use.

Today, with advances in autonomous vehicles continuing at a rapid pace, Sejdic keeps close watch on both technological and legislative news around the issue. Among these developments are Pittsburgh Mayor Bill Peduto’s recently signed executive order laying out guidelines for testing these vehicles in the city, with firms like Argo AI, Aurora and Uber leading the way with strong investment backing from larger companies. President Donald Trump also signed an executive order in February meant to promote the development and regulation of artificial intelligence across all sectors, not just automobiles.

“There’s a progressive culture here in Pittsburgh for innovation, with Pitt and Carnegie Mellon University leading those fronts, and AI is part of that innovation culture,” Sejdic said. “Pittsburgh also has unique terrain not often found in other large U.S. cities for these autonomous vehicles; if you drive a half hour outside the city, you drive through rolling hills and rural areas.”

Self-driving vehicles have been the source of intrigue and controversy in recent years. Uber has been testing autonomous vehicles since 2016, but temporarily suspended testing in March 2018 after a self-driving car operated by Uber struck and killed a pedestrian in Tempe, Arizona. Testing resumed later in the year.

Sejdic said while artificial intelligence is more than a decade away from being able to successfully pilot vehicles with enough accuracy for everyday use, the technology is rapidly progressing. However, having autonomous vehicles on the road isn’t a matter of just having the right technology.

“It’s also a matter of law, ethics and other unresolved issues,” Sejdic said. “Let’s say a self-driving car hits a pedestrian. Who’s to blame? Is it the developer of the algorithm used to program the car? Is it the company that packages the algorithm into a product? Is it the car manufacturer? The questions for these moral dilemmas need to be answered before cars can take the road by themselves. The same can be said for the medical field if and when AI misdiagnoses a patient.”

Sejdic said artificial intelligence may move into less consequential facets of everyday life, such as scheduling and budgeting apps on people’s phones. Despite this, he believes AI will not completely take over the way humans live.

“Something that humans have that AI doesn’t is a conscience. That’s a very difficult thing to program since no one person has the same conscience as another,” Sejdic said.