Tufts University Innovates AI-Driven Study Aids to Enhance Learning in Physical Therapy Education

Pioneering safe and effective educational tools for Tufts DPT students.
Artificial Intelligence Stock Image -- shows technological neuron system with AI in the center

In the rapidly evolving realm of artificial intelligence (AI), Benjamin Stern, MS, DPT, an assistant professor in the Tufts University School of Medicine's Doctor of Physical Therapy Phoenix program, is spearheading efforts to develop tailor-made study aids using natural language processing. The aim is to support students in mastering the fundamental aspects of physical therapy while at Tufts.

Recently, their team successfully secured a Tufts Data Intensive Studies Center (DISC) award. The Tufts team along with Stern is comprised of Peter Nadel, digital humanities and natural language processing specialist at Tufts Research Technology, Tara Dickson, PhD, DPT, assistant professor in the Tufts University School of Medicine Doctor of Physical Therapy Phoenix program, and David Hammer, PhD, professor of physics and astronomy in the Tufts School of Arts and Sciences.

Their research endeavor revolves around building a model for integration into the Tufts DPT curriculum, with prospects of extending its use to other disciplines. The team envisions the model serving as a time-saving asset for faculty in preparing course materials, while offering extensive content customization to explore new frontiers. Simultaneously, it aims to provide students with an accurate and dependable system for retrieving essential documents. Beyond these immediate goals, they aim to stimulate a broader discussion within Tufts about the potential applications of AI and large language models as educational tools within and beyond the classroom.

“We have an early prototype of a learning tool that creates brief summaries and short quizzes from my lecture transcripts,” said Stern. “As we move forward, we’re going to try and place a safety backstop or guardrails and then elevate them as high as possible to try and get model responses as accurate and reliable as we can. We’ll try different approaches to accomplish this, likely combining a few that work best for us. Right now, we’re using a process called retrieval augmented generation. This has improved our model output because as opposed to responding based on the original model training data, the model retrieves information from documents (lecture transcripts in this case) held in a database.”

With the DISC award funding, their plan is to further refine the application to enhance its practicality. Stern also acknowledges the potential risk of students relying too heavily on the AI and emphasizes their commitment to minimizing inherent biases associated with the model's training and use.

“We’re incredibly lucky because Tufts has some pretty amazing technical resources,” said Stern. “Peter and I share the belief that customized AI study applications will improve the student learning experience and student performance. These tools will empower learners across all fields of study, but they’ll also allow faculty to focus on higher-order learning objectives.”

While the research project progresses, Sterns has already initiated the integration of AI into the Tufts DPT Phoenix program's Primary Care Course this semester. Students are actively engaging with chatbot outputs, pinpointing errors, and exploring potential use cases like image recognition, where the model can interact with users about various images, such as radiography, assistive devices, or exercise equipment. During the class discussions, they also delved into the potential pitfalls of relying solely on model responses, citing instances where critical details in interpreting x-rays could be overlooked with dire consequences.

Sterns and his team are eagerly anticipating staying ahead of the curve, examining how faculty and students can safely incorporate AI within the classroom setting.