Trustworthy artificial intelligence for the greater good

Carnegie Mellon University has consistently pushed the frontiers of artificial intelligence (AI) research, pioneering solutions to pervasive problems in healthcare, transportation, manufacturing, and many other disciplines. Now, as AI rapidly grows in reach and sophistication, it’s crucial for AI developers to be able to prove that their programs are safe and trustworthy.

Toward that goal, Ding Zhao, associate professor of mechanical engineering, created the Trustworthy AI Autonomy course in 2021. His main objectives for the class are to help students understand AI autonomy and train them in crucial research skills. Students work on three individual research projects to build a solid background in trustworthy AI, as well as read recent papers to gain exposure to ongoing research. At the end, each student picks one paper and writes a presentation on it, receiving constructive criticism from another student.

Over the course of the semester, students also complete what Zhao calls an extended challenge. Working in teams of three, they’re tasked with designing a protocol for testing self-driving cars in a virtual environment. This year, the team with the best protocol received free tickets to the upcoming conference for the IEEE International Automated Vehicle Validation Conference, where experts from academia, industry, and government will come together in Pittsburgh to discuss trustworthiness of self-driving cars.

Zhao first began teaching Trustworthy AI Autonomy at CMU in Pittsburgh, and shortly after, the course was made available to CMU-Africa students in a hybrid learning capacity. Students from the two locations work together on the final presentation, as well as the extended challenge. “I think this course helps students feel like they can make an impact in this ever-changing field,” said Zhao. “Whether in Pittsburgh or in Kigali, all of my students are at the forefront of the world.”

I think this course helps students feel like they can make an impact in this ever-changing field. Whether in Pittsburgh or in Kigali, all of my students are at the forefront of the world.

Ding Zhao, Associate Professor, Mechanical Engineering

To supplement the lectures, readings, and research projects, Zhao also invites AI experts to share their knowledge with the class. This past spring, he invited a speaker from Google DeepMind, an AI research lab currently supporting research at Zhao’s lab. “CMU is in a leading position in robotics and AI,” said Zhao. “We are excited to link our students to leading industrial partners.”

Dirac Murairi (MS EAI ’25) is a student at CMU-Africa who took Zhao’s course on trustworthy AI. He’s passionate about making AI accessible through everyday devices such as mobile phones, smart watches, and personal computers. “I enjoyed the whole course, especially the part where we learned about adapting AI to different tasks,” Murairi said. “It showed me how we can make AI flexible to suit our needs, which I found both impressive and revolutionary.”

Another of Zhao’s students, Mark Phillip Matovic (MS EAI ’25), agrees that there’s a lot to learn in Trustworthy AI Autonomy. “The course expanded my understanding of advanced AI applications, broadening my previously limited perception,” Matovic said. “Professor Zhao introduced me to remarkable alternative applications, providing insights into areas like reinforcement learning and generative modeling, which I’m eager to explore further.”

Every year at the conclusion of the course, many students are excited to connect with others who have taken Trustworthy AI Autonomy. This informal community now consists of hundreds of students and professionals from both CMU-Pittsburgh and CMU-Africa, all exchanging ideas and resources with the shared goal of making a positive impact on the world.

“I tell my students that they should also know AI is only a tool,” Zhao said. “It has advantages and limitations just like other tools. Their goal is to make AI more reliable, more affordable, and more inclusive, so that it can be used for the greater good.”

More praise for the Trustworthy AI Autonomy course

One alumnus of this course was hired by a startup, VeyTel, Inc. to help several hospitals advance their analysis of x-rays of lungs for patients with COVID-19, RSV, pneumonia and acute respiratory distress syndrome.

“At VeyTel, we’ve had the privilege of working with highly motivated interns from Carnegie Mellon University, whose contributions have significantly advanced our products. Particularly, Nischal Suresha (MS AIE ’23) and his team, inspired by CMU’s 24-784 Trustworthy AI Autonomy course, have made remarkable strides. Their work on segmenting human lungs from CXRs, generating artificial imagery using GANs to enhance our dermatology training sets, and pioneering diffusion modeling approaches to differentiate disease characteristics in clinical images, has been invaluable in pushing the boundaries of our AI capabilities.”

 -Kevin Mitchell, director of data science and AI, VeyTel, Inc.