Four Clemson University faculty have received a grant to develop an artificial intelligence-driven virtual teaching assistant (ViTA) that will provide graduate students the support they need to succeed in the University’s highly interdisciplinary medical biophysics program.
“An AI assistant can enhance both student learning and teaching by providing support tailored to each individual student,” said Laura Finzi, the Dr. Waenard L. Miller, Jr. ’69 and Sheila M. Miller Endowed Chair in Medical Biophysics and a professor in the program. “What excites me the most about this project is the opportunity to democratize access to quantitative, interdisciplinary education and to broaden the scientific background and toolbox of new investigators while maintaining depth and rigor.
“The biggest impact of this project, in my opinion, is to form a class of thinkers who can comfortably straddle different disciplines. These will be the next leaders in scientific innovation and discovery.”
Truly interdisciplinary
The medical biophysics (MBIO) program is taught by scientists and engineers from nine different areas, including mathematical and statistical sciences, physics, chemistry, biological sciences, genetics and biochemistry, material sciences, bioengineering, chemical and biomolecular engineering, and nursing. Faculty Prisma Health also teach.
The MBIO curriculum focuses on understanding how fundamental physics principles underlie complex biological and medical phenomena and how scientists can apply those principles to help solve biomedical challenges and provide better health outcomes for people. Students learn major experimental and computational techniques used to study molecules, cells, tissues and human organs as well as physical approaches to disease detection, diagnostics and drug design.
“The program is one-of-a-kind. It’s meant for grooming a generation of new researchers who have a skill set in various fields that are relevant to clinical applications,” said Ramakrishna Podila, a nano-biophysics scientist and an associate professor in the Department of Physics and Astronomy.
But the program’s truly interdisciplinary nature can present challenges for students, who often come into the program with undergraduate training in either mathematics-heavy majors such as physics and engineering or in life sciences like biology and genetics.
Need support
They typically need support in acquiring and strengthening skills they may not have been exposed to as an undergraduate, Finzi said.
That’s where the AI comes in. The scientists will train a local model similar to ChatGPT, validated by Clemson faculty experts from different disciplines.
“The biggest challenge with AI, if you just use the publicly available search engines, is that it spits out answers that are very confident and nicely written, but, especially when you go to the level of a graduate student, they are many times wrong,” Podila said. “If you’re not careful, you’re trusting wrong information.”
Train chatbot like a student
Instead, the faculty members will train a chatbot like they would teach a student.
“I am a domain expert in biophysics, so I would train the chatbot and make sure the chatbot is a good student in biophysics. Somebody else who is an expert in genetics would do the same for genetics and so on,” Podila said. “The different departments that are involved in this MBIO program would basically train this large language model with their own notes, and they would validate its responses.”
At the end of the first year of the $700,000 National Science Foundation grant, Podila said they would have a highly trained, local ChatGPT-like interface that would be integrated into Clemson’s learning management system.
In the final three years of the grant, the ViTA will be implemented, and researchers will see how it affects student learning outcomes.
Variety of interactions
Students would interact with ViTA in a variety of ways using pre-designed prompt sets to help guide them. ViTa will reference course materials to provide explanations of concepts and offer step-by-step hunts and support in solving problems.
“Virtual teaching assistants have the potential to deliver instruction that is both effective and efficient, while offering personalized, timely feedback to students. They can serve as collaborative partners in the learning process, helping to motivate and engage learners in meaningful ways,” said Danielle Herro, a professor of learning sciences in the Clemson College of Education. “I’m excited to explore graduate students’ perspectives on learning with AI assistants, as this research will help inform and enhance our instructional practices at Clemson — and across broader educational communities.”
The ViTA will store chat histories, allowing faculty to analyze students’ thought processes and improve their teaching methods, Podila said.
Not a replacement
For all its usefulness, ViTA is an assistant, not a replacement for human teachers, Podila said.
“This virtual TA is an innovative approach that will enable students to adapt their learning and develop learning skills toward the ever-growing impact of AI in science, technology and everyday life,” said Emil Alexov, who established the MBIO program in 2021. “It is evident that AI is and will be playing a significant role in all aspects of human life and the fact that the medical biophysics program will use this advanced approach to facilitate the teaching excites me the most. The outcome can be used to facilitate teaching in all other graduate and undergraduate programs.”
